Daily Archives: October 12, 2021

Link Building Do’s and Don’ts: 10 Tips For Safe Backlinks

By | October 12, 2021


As your website gains authority and visibility in the search engines, you will inevitably attract toxic backlinks.

Toxic backlinks are links from websites that are not designed to educate, inform, or even entertain. They are created purely for ‘black hat’ purposes and are never even visited by real people.

Here’s a useful guide on how to spot toxic backlinks.

So why does your website attract these kinds of links? It’s a question we’ve all asked ourselves, at one time or another.

Sometimes it is due to ‘negative SEO’. 

That’s where a competitor tries to damage your rankings by directing low quality backlinks to your website. Alternatively, toxic links can come from people who want to charge you a fee to have them removed.

Wherever they come from, links from spammy websites will damage the SEO of your website.

Ideally, you should ask the linking websites to remove your link. But very often they will ask for a fee to remove the backlink. That, after all, is why they linked to you in the first place.

Google has provided a remedy to this: the Google disavow tool. When you upload a list of linking websites that you want to disavow, Google will ignore the links pointing to your site from those spammy websites.

You can identify toxic backlinks and disavow them in SEO tools such as Ahrefs, SEMrush, and Monitor Backlinks.



Source link

On-Device Machine Learning and Prediction

By | October 12, 2021


Sharing is caring!

On-Device Machine Learning Tasks Such As Prediction, Training, and Example Collection

This newly granted patent relates an on-device machine learning platform and associated techniques that enable on-device prediction, training, example collection, and other machine learning tasks or functionality.

Before reading this post, there are many help, support, and Blog pages from Google about On-Device Machine Learning and Federated Learning worth a look. Theseare not getting much discussion in the SEO industry, and they probably should get more. There are support and information pages from Google that are a few years old, already:

In recent years, machine learning has gotten used to provide improved services to users of computers. In particular, many applications or other computing programs or systems rely on machine-learned ****** to produce inferences based on input data associated with the program, device, and user. The applications can use the assumptions to perform or influence any task or service.

One conventional training scheme for solving machine learning problems can include collecting at a centralized location, like a server device, for training examples from many computers and smartphone user devices. A machine-learned model can then get trained at the centralized location based on the collected training examples.

Besides, the trained model can get stored at a centralized location.

The user computing device must send input data to the server computing device to infer the model. The device can wait for the server device to put the machine-learned model to produce inferences based on the transmitted data. The device can then receive the assumptions from the server computing device again over the network.

Inferences and Training Examples Go-Between A Computing Device and A Server

In such scenarios, the training examples and inferences get transmitted between the user computing device and the server computing device over a network. Such network transmission represents a security risk as the data sent over the network may become susceptible to interception. Besides, such network transmission increases network traffic which can result in reduced communication speeds. Further, the latency associated with transmitting the data back and forth over the network can cause delays in providing the application’s services.

More recently, specific applications have included machine-learned ****** stored and implemented on the user device. But, this architecture is both challenging to put in place and resource-intensive. For example, the application must keep, manage, train, and put machine-learned ****** in such a scenario. The inclusion of the model and related support services within the application itself can increase the data size of the application, resulting in a larger memory footprint.

Machine learning within the application can also must more frequent application updates. For example, the application may need to get updated as the underlying machine learning engine gets updated or otherwise advances. Application updates can network usage and downtime for the user as the update gets downloaded and installed.

Furthermore, machine learning within the application can also complicate application development, as more services need to get built into the application itself. Thus, developers may get required to learn and stay abreast of the intricacies of different machine learning engines.

One aspect of this patent gets directed to a computing device. The computing device includes processors and non-transitory computer-readable media. The non-transitory computer-readable media store: applications implemented by the processors; a centralized example database that stores training examples received from the applications; and instructions that, when executed by the processors, cause the computing device to put in place an on-device machine learning platform that performs operations.

On-Device Machine Learning Operations

The operations include:

  • Receiving a new training example from a first application of the applications via a collection application programming interface
  • Determining context features descriptive of a context associated with the computing device
  • Storing the new training example together with the context features in the centralized example database for use in training a machine-learned model

Another aspect of the present disclosure gets directed to non-transitory computer-readable media that store instructions that, when executed by processors, cause a computing device to put in place an on-device machine learning platform that performs operations. The operations include receiving input data from the first application of applications stored on the computing device via a prediction application programming interface.

The operations include:

  • Deciding context features descriptive of a context associated with the computing device
  • Employing at least a first machine-learned model of the ****** stored on the computing device to generate at least one inference based at least in part on the input data and further based at least in part on the context features
  • Providing at least one deduction generated by the first machine-learned model to the first application via the prediction application programming interface

The patent is at:

On-device machine learning platform applications or other clients
Inventors: Pannag Sanketi, Wolfgang Grieskamp, Daniel Ramage, Hrishikesh Aradhye, and Shiyu Hu
Assignee: Google LLC
US Patent: 11,138,517
Granted: October 5, 2021
Filed: August 11, 2017

Abstract

The present disclosure provides systems and methods for on-device machine learning.

In particular, the present disclosure gets directed to an on-device machine learning platform and associated techniques that enable on-device prediction, training, example collection, and other machine learning tasks or functionality.

The on-device machine learning platform can include a context provider that securely injects context features into collected training examples and client-provided input data to generate predictions/inferences.

Thus, the on-device machine learning platform can enable centralized training example collection, model training, and usage of machine-learned ****** as a service to applications or other clients.

Performance of the On-Device Machine Learning Functions

This patent is about building systems and methods for on-device machine learning. It describes an on-device machine learning platform. It also includes associated techniques that enable on-device prediction, training, example collection, and other machine learning tasks or functionality, which may get referred to as “machine learning functions.”

The on-device machine learning platform is from computer programs stored locally on a computing device or terminals, such as a smartphone or tablet. Those get configured when executed by the user device or terminal to perform machine learning management operations. They can enable on-device machine learning functions on behalf of locally-stored applications, routines, or other local clients.

At least some of the on-device machine learning functions may get performed using machine learning engines implemented locally on the computing device or terminal. Performance of the on-device machine learning functions on behalf of the locally-stored applications or routines may get referred to as “clients.” These may provide a centralized service to those clients, which may interact with the on-device machine learning platform via application programming interfaces (APIs).

Generating Predictions and Inferences Using Context Features

The on-device machine learning platform can include a context provider that securely injects context features into collected training examples and client-provided input data to generate predictions or inferences. Thus, the on-device machine learning platform can enable:

  • Centralized training
  • Example collection
  • Model training
  • Usage of machine-learned ****** as a service to applications or other clients

More particularly, a computing device such as, for example, a mobile computing device, like a smartphone, can store or otherwise include applications (e.g., mobile applications). The computing device can also include and put in place the on-device machine learning platform and machine-learned ******. For example, the device can store the machine-learned ****** in a centralized model layer managed by the platform.

According to one aspect of the present disclosure, the applications can communicate with the on-device machine learning platform via an API. It may get referred to as the “prediction API.” It may also provide input data and get predictions based on the input data from the machine-learned ******.

The on-device machine learning platform can download the URI content with a uniform resource identifier (URI) for a prediction plan.

These can be instructions for running the model to get inferences or predictions, or model parameters. These get created by interacting with a machine learning engine to build the model.

It can show the prediction plan and parameters, and we can get inferences or predictions with the model. Besides, the platform can store the content so that it can get used for later prediction requests.

Thus, on-device machine-learned ****** can communicate with the on-device machine learning platform via a client and service relationship. In particular, the machine-learning platform can be a standalone multi-tenant service that can get referenced by applications. A given application is not required to store, manage, train for machine-learned ******. You can simply communicate with the on-device machine learning platform to request and receive inferences from the ******.

Storing Training ****** Received From Applications

According to another aspect of the present disclosure, the computing device can include a centralized example database that stores training examples received from the applications. For example, each application that is a client or tenant of the platform can have its collections of ****** stored within the centralized example database.

The groups can get supplemented and managed in an online fashion. In particular, the on-device machine learning platform can receive training examples from the applications via an API (which may get referred to as the “collection API”) and manage the standards’ storage in the centralized example database.

The on-device machine learning platform can cause storage of each training example received from an application (e.g., within its corresponding collection) according to options parameters associated with the application providing the training example. As one example, the options parameters can include a time-to-live parameter that defines a period for which training examples get stored (e.g., and are after that deleted). The options parameters can get predefined and adjusted through instructions provided to the platform via the collection API.

Injecting Context Features Descriptive of a Context Associated With the Computing Device

The on-device machine learning platform can securely inject context features descriptive of a context associated with the computing device into the training examples. For example, upon receiving a training example from an application, a context provider component of the on-device platform can determine context features and store such context features together with the training example in the centralized example database.

For example, the context features and the data provided in the new training example can get stored as a single database entry. The particular context features that get determined and then injected or otherwise associated and stored with a training example received from a specific application can get specified by the options parameters for such particular application.

As described above, these options features can get adjusted or predefined via the collection API. Thus, an application can control (e.g., via defining the options parameters) which context features or context types get injected into its training examples.

Context features get injected on the service side, such that the context features never need to become directly available to applications. The centralized example database is not directly accessible by the applications. The context information stored with a specific training example is not accessible even to the application that provided the training example.

Grouping or Categorizing Context Features According to Different Context Types

The context features can get grouped or otherwise categorized according to many different context types. In general, each context type can specify or include a set of context features with well-known names and well-known types. One example context type is device information, consisting of the following context features: audio state, network state, power connection, etc.

The content provider requests the value injected for a given context feature from the device (e.g., from a context manager) at the time/point of injection. Alternatively or additionally, the context provider can register as a listener to context updates and maintain a context feature cache of current values for the context features based on the context updates. Then, when context features get injected, the context provider can simply access the context feature cache and inject the current value maintained in the store for the particular context feature.

Besides context features to training examples at storage time, the context provider can also inject context features at inference time. In particular, when a specific application or other client requests to use the prediction API for an inference generated based on client-provided input data. The context provider can inject or provide supplemental context features for input into the corresponding machine-learned model alongside the input data. Thus, inferences can get made on context information and the client-provided input data, which may improve the accuracy of the stereotypes.

The training examples and context features described herein get provided to illustrate example data stored with training examples or used to provide inferences by the on-device platform. But, the data is not collected, used, or analyzed unless the user provided consent after knowing what gets collected and how it gets used. Besides, certain information or data can get treated in or more ways before it gets stored or used so that personally identifiable information gets removed or stored in an encrypted fashion. Further, the user can get supplied with a tool to revoke or change the scope of permissions.

According to another aspect, since specific applications or other clients may have permission to access only certain context features or context types (e.g., defined or controlled by a device user), the context provider can perform client permission control. The on-device machine learning platform or another device component can maintain a mapping of which clients can access context types.

When context features get injected, such as a training example for storage or to supplement client-provided input data at inference time, the context provider can check the permission status of the application or another client relative to the context features or context types to get injected.

For example, the permission status for a particular application and a context type can describe whether such application has permission to access such context type. The context provider will inject features included in context types that the application has permission to access. This permission prevents an application from accessing, even in second-hand parts or types that it can not access.

Like the time-to-live options parameter described above, each context feature can have an end period associated in addition to that or assigned to that. This end period information can get related to each training example that contains context features.

After the end period for a particular context feature provided in a specific training example, the value for such a feature can get deleted or otherwise removed from such a training example. Or the entire training example can get deleted or otherwise removed.

In response to a change to a permission status for a particular application or another client relative to a specific context or context type feature, the on-device platform can delete any entries for such context features associated with training examples from the centralized example database related to the particular application. Besides, the corresponding ****** can be re-trained on the remaining data after deleting the context feature values.

Training The On-Device Machine Learning Model With Applications

According to yet another aspect of the patent, the applications can communicate with the on-device machine learning platform using an API. It may get referred to as the “training API” to cause re-training or updating of a machine-learned model based on training examples stored in the centralized example database. Given a URI for a training plan, such as instructions for training the model, the on-device machine learning platform can train the model by interacting with a machine learning engine to cause movement of the model by the engine. It can do this based on prior collected examples. For example, the training can get performed in the background at scheduled times and when the device is idle.

After re-training the model, the re-trained model can provide inferences as described elsewhere herein. These inferences will have higher accuracy since the model has been re-trained on data specific to the user. Thus, the on-device machine learning platform can enable centralized example data collection and corresponding personalization of machine-learned ****** as a service to applications or other clients.

The machine learning platform can upload logs or other updates about the machine-learned ****** to the cloud for detailed analytics of machine learning metrics. The on-device platform can determine an update that describes the parameters of a re-trained machine-learned model or changes to the parameters of the machine-learned model that occurred during the re-training of the model.

Updating the Learning Modle Using Federated Learning

federated learning

The platform can send the update to a central server computing device, such as “the cloud.” for aggregation with other updates provided by other computers. Thus, the platform can enable participation in a process known as “federated learning.” “Federated Learning” is when a device determines a local update to a model based on locally stored data and then communicates the local update to a cloud service (e.g., in a privacy-preserving and communication efficient manner) for aggregation to generate a global update to the model.

Each application can enclave certain functionalities (e.g., for every functionality) offered by the on-device platform. For example, the platform can authenticate an application before accessing the platform gets returned to the application via a factory. The returned interface can then represent a sole view of the application’s enclave in the venue. In one example implementation of this process, when an application connects to an API of the platform, the application can provide a signed package token that verifies the application’s identity. The application is not able to get the API interface without passing this authentication.

Machine Learning Engine Variations

According to another aspect of the present disclosure, the on-device machine learning platform can completely abstract from an underlying machine learning engine. For example, the machine learning engine can be a TensorFlow engine, a neural network library, or other engines that enable the implementation of machine-learned ****** for inference and training.

Due to such abstraction, the machine learning platform can treat model artifacts as blobs generated in the cloud and then shipped to devices (e.g., via dynamic model download). They are then interpreted by matching engines. In such a fashion, the machine learning platform and its supported applications can be resilient against changes to the machine learning engine and agnostic/flexible to a particular motor or engine type employed.

According to another aspect, a toolkit that is complementary to the on-device platform can provide a set of tools (e.g., Python tools) to create and simulate ****** in the cloud before they get shipped as artifacts to devices. The toolkit can generate from the same source artifacts (e.g., Python source artifacts) for different versions of machine learning engines or even different engine types (e.g., mobile-focused TensorFlow Lite versus a neural network library, etc.).

The on-device machine-learning platform can get included in or implemented as an application, such as, for example, a mobile application. For instance, in the context of the Android operating system, the on-device machine-learning platform can get included in an Android Package Kit (APK) that can get downloaded and updated. In one example, the on-device machine-learning platform can fit in or implement as a part of a more extensive application that provides many different support services to other applications or the device itself.

For example, besides the on-device machine-learning platform, the more extensive application can enable the computing device to interact with a digital distribution service (e.g., downloading applications and updates from an “app store”) and other services. In another example, the on-device machine-learning platform can be included in or implemented as part of the device’s operating system rather than a standalone application.

The On-Device Machine Learning Model and Technical Impact and Help

The systems and methods of the patent provide many technical effects and benefits. As one example of technical impact and help, the on-device machine-learning platform can personalize machine-learned ****** based on locally-stored device-specific training examples, thereby leading to higher accuracy inferences. Similarly, as described elsewhere, the on-device platform can enable participation of the device in “federated learning,” in which local updates get aggregated to generate a global update, thereby leading to improved global model accuracy for all individuals.

Another example of technical effect and benefit is that the on-device machine-learning platform can enable the secure inclusion of contextual signals into training examples and inference inputs. That is, context features can get added to training examples or inference inputs in a manner that maintains privacy and complies with user-defined permissions. By including context information, the accuracy of the inferences provided by the machine-learned ****** can improve.

Another example of technical effect and benefit, the on-device machine-learning platform can provide a centralized service so that applications do not need to manage (e.g., train and run) machine-learned ****** or interact with machine-learning engines. As such, a given application does not have to store, manage, train, and put in place machine-learned ****** but can instead communicate with the on-device machine learning platform to request and receive inferences from the ******. Focusing on communication can enable the data size of applications to be smaller. It can also simplify the development and deployment of applications or other clients as application developers are not required to learn the intricacies of each different machine learning engine but can instead rely upon the usage of the platform APIs.

Like the previous effect and benefit, the on-device machine-learning platform can also easily update a single centralized service rather than all applications. For example, when a new version or type of machine learning engine gets launched, only the on-device platform must update to interact with the new engine because the applications do not interact with the machine but can use the forum to do so on.

This updated approach can drop the need for applications to ensure that they are compatible with the latest versions of machine learning engines. They can instead rely upon the on-device platform to stay up-to-**** as the engine technology advances.

Yet another example of technical effect and benefit, the on-device machine-learning platform can improve communication network efficiency and usage. That is, under past paradigms where machine learning gets performed by a server rather than on-device, various types of information (e.g., input data, training examples, inferences, model parameters, etc.) got transmitted by the server to the device over a communications network (e.g., the Internet). But, since the present disclosure enables on-device prediction, training, example collection, and other machine learning tasks or functionality, such information is not required to get transmitted (at least in every instance) over a communications network. Thus, communications network traffic, efficiency, and usage get improved. Since the input data, training examples, is not transmitted to and from a server, the security of the data may get increased.

An Example Computing Device That Includes An On-Device Machine Learning Platform

The computing device can be any type of computing device, including, for example, a desktop, a laptop, a tablet computing device, a smartphone, a computing device that can get worn, a gaming console, an embedding computing device, or other forms of computers. The computing device can be a mobile computing device and a user computing device.

The on-device machine learning platform can enable on-device prediction, training, example collection, and other machine learning tasks or functionality, which may get referred to as “machine learning functions.”

The on-device machine learning platform may be a computer program stored on a smartphone or tablet, which gets configured to perform machine learning management operations that enable on-device machine learning functions on behalf of locally-stored applications or other local clients.

At least some of the on-device machine learning functions may get performed using machine learning engines implemented on the computing device. Performance of the on-device machine learning functions on behalf of the locally-stored applications or routines (which may get referred to as “clients”) may get provided as a centralized service to those clients, which may interact with the on-device machine learning platform via application programming interfaces (APIs).

The on-device machine learning platform can include a context provider that injects context features into collected training examples and client-provided input data to generate predictions/inferences. Thus, the on-device machine learning platform can enable centralized training, example collection, model training, and usage of machine-learned ****** as a service to applications or other clients.

More particularly, the computing device can store or otherwise include applications (e.g., mobile applications). The computing device can also include and put in place the on-device machine learning platform and machine-learned ******. For example, the device can store the machine-learned ****** in a centralized model repository managed by the platform.

Predictions Based On The Input Data From The Machine-Learned ******

According to one aspect of the present disclosure, the applications can communicate with the on-device machine learning platform via an API (which may get referred to as the “prediction API”) to provide input data and get predictions based on the input data from the machine-learned ******. Given a uniform resource identifier (URI) for a prediction plan, such as instructions for running the model to get inferences or predictions or model parameters, the on-device machine learning platform can download the URI content – the prediction plan and parameters and get assumptions or projections by running the model. This approach is by interacting with a machine learning engine to cause the implementation of the model by the engine. Besides, the platform can cache the content (e.g., within the repository) to get used for later prediction requests.

Thus, on-device machine-learned ****** can get accessed by an application by communicating with the on-device machine learning platform via a client/service relationship. For example, a respective machine-learned model can be provided for each application and managed by the platform. Applications can share a single machine-learned model, or a single application can have two or more ******.

The machine-learning platform can be a standalone multi-tenant service that can get referenced by applications. As such, a given application does not have to store, manage, train, and put in place, machine-learned ******. It can instead communicate with the on-device machine learning platform to request and receive inferences from the ******.

The Computing Device Can Include A Centralized Example Database That Stores Training Examples Received From The Applications

According to another aspect of the present disclosure, the computing device can include a centralized example database that stores training examples received from the applications. In particular, the on-device machine learning platform can receive training examples from the applications via an API (which may get referred to as the “collection API”) and manage the standards’ storage in the centralized example database. For example, each application that is a client or tenant of the platform can have its collections of ****** stored within the centralized example database. The displays can get supplemented and managed in an online fashion.

The on-device machine learning platform can cause storage of each training example received from an application (e.g., within its corresponding collection) according to options parameters associated with the application providing the training example. As one example, the options parameters can include a time-to-live parameter that defines a period for which training examples get stored (e.g., and are after that deleted). The options parameters can get predefined and adjusted through instructions provided to the platform via the collection API.

The on-device machine learning platform can securely inject context features descriptive of a context associated with the computing device into the training examples. For example, upon receiving a training example from an application, a context provider component of the on-device platform can determine context features. It can store such context features together with the training example in the centralized example database.

The Injection of Context Features

For example, the context features and the data provided in the new training example can get stored as a single database entry. The particular context features that get determined and then injected or otherwise associated and stored with a training example received from a specific application can get specified by the options parameters for such particular application. As described above, these options features can get adjusted or predefined via the collection API. Thus, an application can control (e.g., via defining the options parameters) which context features or context types get injected into its training examples.

Context features get injected on the service side, such that the context features never need to become available to applications. In particular, the centralized example database is not accessible by the applications. The context information stored with a specific training example is not accessible even to the application that provided the training example.

The context features can get grouped or otherwise categorized according to many different context types. In general, each context type can specify or include a set of context features with well-known names and well-known types. One example context type is device information, consisting of the following context features: audio state, network state, power connection, etc.

The context provider requests the value injected for a given context feature from the device (e.g., from a context manager) at the time/point of injection. Or additionally, the context provider can register as a listener to context updates from the context manager and maintain a context feature cache of current values for the context features based on the context updates. Then, when context features get injected, the context provider can access the context feature cache and inject the current value maintained in the store for the particular context feature.

Besides injection of context features to training examples at storage time, the context provider can also inject context features at inference time. When a specific application or other client requests, via the prediction API, for an inference to be generated based on some client-provided input data, the context provider can inject or provide supplemental context features for input into the corresponding machine-learned model alongside the input data. Thus, inferences can get made based on context information besides the client-provided input data, which may improve the accuracy of the assumptions.

According to another aspect, since specific applications or other clients may have permission to access only certain context features or context types (e.g., defined or controlled by a device user), the context provider can perform client permission control. The on-device machine learning platform or another device component can maintain a mapping of which clients have permission to access which context types or context features. When context features get injected, either into a training example for storage or to supplement client-provided input data at inference time, the context provider can check the permission status of the corresponding application or another client relative to the context features or context types to get injected.

For example, the permission status for a particular application and a context type can describe whether such application has permission to access such context type. The context provider will inject only context features that get included in context types that the application has permission to access, thereby preventing an application from accessing (even in a second-hand fashion) context features/types to which it does not have permission to access.

Like the time-to-live options parameter described above, each context feature can have an end period associated in addition to that or assigned there. This end period information can get related to each training example that contains context features. After the end period for a particular context feature provided in a specific training example, the value for such a feature can be deleted or otherwise removed from such a training example. Or, the entire training example can get deleted or otherwise removed.

Furthermore, in response to a change to a permission status for an application or another client relative to a feature of context or context type, the on-device platform can delete from the centralized example database any values or entries for such context features or types that get associated with training examples related to the particular application. Besides, the corresponding ****** can be re-trained on the remaining data after deleting the context feature values.

Training APIs

Re-Training Using the ‘Training API’

According to yet another aspect of the patent, the applications can communicate with the on-device machine learning platform via an API, referred to as the “training API,” to cause re-training or updating of a machine-learned machine model based on training examples stored in the centralized example database. As an example, given a URI for a training plan (e.g., instructions for training the model), the on-device machine learning platform can run training of the model (e.g., by interacting with a machine learning engine to cause movement of the model by the engine) based on earlier collected examples. For example, the activity can get performed in the background at scheduled times and when the device is idle.

After re-training the model, the re-trained model can provide inferences as described elsewhere herein. These inferences will have higher accuracy since the model has been re-trained on data specific to the user. Thus, the on-device machine learning platform can enable centralized example data collection and corresponding personalization of machine-learned ****** as a service to applications or other clients.

According to another aspect, the machine learning platform can upload logs or other updates about the machine-learned ****** to the cloud for detailed analytics of machine learning metrics. For example, the on-device platform can determine an update that describes the parameters of a re-trained machine-learned model or changes to the parameters of the machine-learned model that occurred during the re-training of the model (e.g.,,, a “gradient”).

The platform can send the update to a central server computing device (e.g., “the cloud”) for aggregation with other updates provided by other computers. Thus, the platform can enable participation in a process known as “federated learning,” in which a device determines a local update to a model based on stored data and then communicates the local update to a cloud service (e.g., in a privacy-preserving and communication efficient manner) for aggregation to generate a global update to the model.

According to another aspect, to protect the applications from each other, each application can have its enclave for certain functionalities (e.g., for every functionality) offered by the on-device platform. For example, the platform can authenticate an application before accessing the platform gets returned to the application via a factory. The returned interface can then represent a sole view of the application’s enclave in the medium. In one example, when an application connects to an API of the platform, the application can provide a signed package token that verifies the application’s identity. The application is not able to get the API interface without passing this authentication.

Machine Learning Appplications and Permissions

Each application’s enclave within the platform is account independent. Thus, many charges associated with the same user profile on the computing device can share the same training data and state. This reflects that many accounts are for the same user, and different users on a computing device would use other user profiles instead.

For certain functionality (e.g., accessing context), permissions get required. An application that wants to use a particular context in the platform, even if it never directly touches the context because it stays within the forum, has permission to access the specific context.

All relevant permissions can get verified in the client and then passed on to the platform called, letting the platform operate with this set of permissions. The platform can request that the user consent to the venue having access to all licenses. Context may also need that a particular user gets logged in. Such users can get specified by the application for those cases or determined by an optional field for context injection. But the user may not get detected by the platform. The API itself does not need authentication with such a specific user account.

According to another aspect of the present disclosure, the on-device machine learning platform can completely abstract from an underlying machine learning engine. For example, the machine learning engine can be a TensorFlow engine, a neural network library, or other engines that enable the implementation of machine-learned ****** for inference and training.

Due to such abstraction, the machine learning platform can treat model artifacts as blobs generated in the cloud and then shipped to devices (e.g., via dynamic model download). They are then interpreted by matching engines. The machine learning platform and its supported applications can be resilient against changes to the machine learning engine and agnostic/flexible to a particular engine or engine type employed.

According to another aspect, a toolkit that is complementary to the on-device platform can provide a set of tools (e.g., Python tools) to create and simulate ****** in the cloud before they get shipped as artifacts to devices. The toolkit can generate from the same source artifacts (e.g., Python source artifacts) for different versions of machine learning engines or even different engine types (e.g., mobile-focused TensorFlow Lite versus a neural network library, etc.).

The on-device machine-learning platform can get included in or implemented as an application, such as, for example, a mobile application. For instance, in the context of the Android operating system, the on-device machine-learning platform can get included in an Android Package Kit (APK) that can get downloaded and updated.

In one example, the on-device machine-learning platform can get included in or implemented as a part of a more extensive application that provides many different support services to other applications or the device itself. For example, besides the on-device machine-learning platform, the more extensive application can enable the computing device to interact with a digital distribution service (e.g., downloading applications and updates from an “app store”) and other services. In another example, the on-device machine-learning platform can be included in or implemented as part of the device’s operating system rather than a standalone application.

An Example Machine-Learned Model Deployment

In particular, an application developer can interact with a toolkit to generate and test a model. The model can get split into or represented by an inference plan and a training plan.

A “plan” can include a protocol buffer (AKA “protobuf”) that contains a graph (e.g., a TensorFlow graph) and instructions on how to run the graph. For example, a plan can be a declarative description of a sequence of operations to perform on a graph (e.g., a TensorFlow graph) that embeds the chart itself. The plan can describe how to query the collection for training data, feed it into the graph, and produce and deliver outputs.

On device machine learningtraining place

The patent illustrates two alternatives (but complementary) deployment schemes. In a first scheme, the inference and training plans are deployed to a cloud server. The cloud server provides the inference plan and the training plan to a device.

The device can put in place the inference plan to generate inferences. The device can additionally put in place the training plan to perform on-device training based on stored data, which can also get referred to as “personalization” or “personalized learning.”

In a second deployment scheme, the inference plan gets deployed to the cloud server as described above. The cloud server provides the inference plan to a device. The device can place the inference plan to generate inferences.

But, alternatively, to deploying the training plan to the cloud server, the training plan gets deployed to a federated server in the second scheme. The federated server provides the training plan to the device. The device can put in place the training plan to perform on-device training based on stored data. After such on-device learning, the device can provide an update to the federated server. For example, the update can describe parameters of the re-trained model or changes to the model’s parameters that occurred during the re-training of the model.

The federated server can receive many such updates from many devices and total the updates to generate an updated global model. The updated global model can then be re-sent to the device.

The device can further provide logs or other updates about the machine-learned ****** that can get used by the developer (e.g., in conjunction with the toolkit) to get detailed analytics of machine learning metrics. Example metrics that can be computed based on the logs include plots, graphs, visualizations of check-in request outcomes, traffic (e.g., volume), loss and accuracy model metrics, phase duration, or other metrics.

Example Personalization and Federated Machine Learning Data Flows

The patent depicts three different learning data flows, which may, in some instances, get used in a complementary fashion.

In a first data flow, training data gets generated on a user device. The training data gets uploaded to a central authority which then trains or re-trains a machine-learned model based on the uploaded data. The model goes to the user device for use (e.g., on-device inference).

In a second data flow which can get referred to as personalization or personalized learning, the training data created on the user device gets used to train or re-train the model on the device. Such a device then uses the re-trained model. This personalized learning enables per-device ****** to get acquainted and evaluated without centralized data collection, thereby enhancing data security and user privacy.

In a third data flow which can get referred to as federated learning, the training data created on the user device gets used to train or re-train the model on the device. Thus, the actual user-specific training data is not uploaded to the cloud, thereby enhancing data security and user privacy.

After such on-device learning, the user device can provide an update to a central authority. For example, the update can describe parameters of the re-trained model or changes to the model’s parameters that occurred during the re-training of the model.

The central authority can receive many such updates from many devices and add the updates to generate an updated global model. The updated global model can then get re-sent to the user device. This scheme enables cross-device ****** to get trained and evaluated without centralized data collection.

central authority

The On-Device Machine Learning Platform

Many different implementations of the on-device machine learning platform described herein are possible. The example on-device machine learning platform can include or put in place the primary process and a background process.

The primary process can handle all API requests. The main function can provide a collection API service that provides training examples, collection services via a collection API; a prediction API service that provides inference generation services via a prediction API; and a training API service that provides model training services via a training API.

The collection API service can persist training examples with automatic retention policies. The training API service can execute training sessions at scheduled times and conditions as an invisible process, drawing data from an example collection. The prediction API service can allow clients to run inference based on a given model, resulting from the trainer or external sources.

The background process can only host training and other periodic maintenance tasks. It can be transactional and designed to get teared down. The background process obtains its state solely from the primary process.

As discussed further below, the context provider injects context information into examples, both for the collection API service and the prediction API service. The storage component can enable and perform storage of ****** (e.g., in the centralized example database) and bookkeeping state. It can get based on LevelDB.

Many prediction engines can get accessed by the prediction API service, depending on the prediction plan type. Prediction and training plans and model parameters get provided by or otherwise managed by the artifact manager. The artifact manager can support retrieving artifacts from a cloud server, from application assets, and files. It can also help mutable artifacts, for example, store training results consumed by a predictor or another trainer.

The background process can host many training engines which get chosen based on the training plan. For Federated Learning, the background process can communicate with a federated learning server to upload training results for accumulation using privacy-preserving techniques (e.g., secure aggregation).

The log manager can upload logs about the machine-learned ****** to the cloud server for detailed analytics of machine learning metrics.

The collection API service is a facility that allows, manages, and performs storage of training examples in a centralized example database for later retrieval by the background process, such as to perform background training. For example, the collection API service can interact with a storage component to manage the storage of the training examples in the centralized example database.

The collection API can look like this: once a client has gotten authenticated, it gets access to an object as demonstrated in the example code below (where the task is an approach to represent asynchronous API calls; Task may get ignored or listened to for observing errors):

The `options` parameter can contain at least the name of a collection. If no further options get provided, the default or earlier configured options can get used. Example options include time-to-live content and context for a learning event before it goes to the database.

The training API and corresponding training API service can schedule the background process to perform training. The background process can implement or interact with training engines to pull data from an example collection and execute a training plan. A plan can be a declarative description of a sequence of operations to perform on a graph (e.g., a TensorFlow graph) that embeds the graph itself.

The plan can describe how to query the collection for training data, feed it into the graph, and produce and deliver outputs.

Each training plan type can get associated with a training plan engine. The on-device machine learning platform can thus get extended by new kinds of plans, making it capable of representing any machine learning execution that fits into the general model of background training.

An Example API To The Trainer

Learning.getTrainerClient(options) can take chances that contain at least a trainer session name and create or reconfigure a training session. Session names can be constants, like package names, chosen by the application, and the session itself can be eternal. The options can also specify the plan type, the method of obtaining the plan, and any parameters specific to a plan type. Plans may get obtained in different ways depending on the plan type; for example, for federation, the program can get downloaded from the federated learning server; for personalization, it might get contained in the assets or downloaded from a cloud server.

The schedule can be either continuous or one-off. In both cases, training will only get scheduled if device conditions allow. Exercise will only get scheduled or otherwise performed if the device is both idle and charging.

TrainerClient.stop( ) can allow cancellation and removal of a training session.

The prediction API can allow a client to feed input and derive predictions based on a trained model. As the trainer, the predictor can be plan-driven, where the plan is what to perform on a graph and how to get inputs and outputs.

As one example, an example prediction API code is as follows:

Learning.getPredictorClient( ) can return a predictor based on the given options. The options can specify how to get plan and model parameters for prediction. They can also specify which context features should get automatically injected into candidate examples before getting passed into the prediction engine.

predictRank( ) can return a prediction for a ranking problem derived from the given context example and the specified candidates. More application-specific prediction methods can get introduced over time.

The below code illustrates one example usage of the three example APIs.

First, options for configuration can get defined. Typically those options can get obtained by the application from the phenotype config component, but for reasons of simplification, they can get defined as static constants:

Note how URIs can get used to referring to artifacts describing training and prediction plans, as well as model parameters. The methods can encode a graph (e.g., a TensorFlow graph) and information on how to execute the graph. The plans can get created by the tools, such as python tools, included in the toolbox. The model parameters can be opaque representations of weights associated with a plan. URIs can refer to a “model repository” (mrepo:), implying that they get downloaded to the device (e.g., from the cloud server), but can also refer to files cached locally (file:). For example, an artifact manager can manage the download of the model artifacts from the server and other model management tasks.

In the case of a file artifact, dependencies can get defined between the APIs. For example, TrainerOptions can get limited to generating a file artifact with trained parameters consumed by the prediction API service. The on-device machine learning platform can internally deal with such input-output dependencies by delaying operations or rejecting them with an appropriate error code if a required input has not yet become produced.

Given that configuration, the platform can include some API code where training examples are continuously fed into the collection. As one example, the following code can consist of adding training examples:

Each time an example gets added to the cache, the context specified with the COLLECTION_OPTIONS can get added as a feature. The user does not need to limit the size or lifetime of added data, which can get dealt with based on the provided options.

To schedule training, an application can typically, at creation time, ensure that background training gets configured using current options and scheduled. If training has gotten already designed before and configuration has not changed, it will be not affected by this example call:

Finally, another piece of example code can use the prediction API to leverage training results. As one example, this can look as provided in the example code below:

The context provider of an on-device machine learning platform can inject context features into learning events. This can happen on the service side, such that context never needs to become directly available to applications.

Example Points Where Context Might Get Injected

1. Before an example gets stored into instance collection. The injected context can become specified by CollectionOptions.

2. Before an example gets passed to the prediction engine. The injected context can become specified by PredictorOptions.

In general, each context category can specify a set of features with well-known names and well-known types that get added to the example (e.g., TensorFlow model proto). The value that gets injected for a given context feature might get requested from the system at the point of injection, or it might be a cached value that the platform’s internal context provider periodically updates.

Example context features include: Audio State; Day Attributes; Calendar; Detected Activity; User-Specific Places (e.g., “home” vs. “work”; Network State; Power Connection; Screen Features; User Location; User Location Forecast; WiFi Scan Info; Weather; or other context features.

A user may get provided with controls allowing an election about both if systems, programs, or features may enable the collection of user information. These may include such as training examples and context features. That is if the user sent the content or communications from a server. Besides, specific data may get treated in ways before it gets stored or used to remove personally identifiable information.

For example, a user’s identity may get treated so that no personally identifiable information can get determined, or geographic location may get generalized where location information gets obtained so that a particular area of a user cannot get determined. Thus, the user may control what data gets collected about the user, how that information is used, and what information is provided.

The on-device platform does not need context, so the context provider collects context for client applications. For example, a client application may need “location” as a feature for its machine learning model. Such an application can explicitly inform the on-device platform that the “location” context gets required.

The on-device platform can first check whether the client application has permission to access the device location. If not, the platform does not give context to the client. Otherwise, if the application does have permission, the location context will get populated for the client once it sends the platform a training example.

At each instance in which a client sends the on-device platform an example, the platform examines their permissions and decides whether the client has the permissions to access the context they claimed.

The actual context content is not provided to the application. Instead, the context is simply populated with the training example for the client. The training examples will get kept within the on-device platform database, so the client cannot access the accurate context content.

Certain types of context need a user account to access, like place alias and calendar. The context provider itself does not state which account to use for context. In such cases, the client application should specify the account. If no account gets set by the client, only context that does not need a bill will get provided for the client.

Context Manager

A context manager generally provides the context the on-device platform is using. The context manager can get located in many places, including, for example, within the forum, within an application that includes the platform, and within an operating system of the device. The context provider can register a listener to the context manager and always keeps the latest context updates in the on-device platform memory to improve performance.

The on-device platform can perform or include stale context end. If users turn off a context signal on the device (e.g., turn off location or activity recognition), the context manager does not inform the on-device platform that the context gets turned off. Instead, the context manager simply stops sending the on-device platform the context updates for these contexts. Thus, to avoid using stale context for future events, the platform can cause the stale context to expire. In particular, based on the context properties, different end-time periods can get defined for each context feature or context type. Once the context has reached its end time is can get deleted.

Another example context feature includes the place aliases context feature. In particular, whether a user is at home or work is an essential feature for many client applications. Considering that the user’s home/workplaces do not change frequently, the platform can ask for the current home/work alias once the context provider gets constructed. If the user has consented to such information, the users’ home/workplaces can contact cached. The context provider can use location context to determine whether a user is at home or work by comparing the location context to the cached locations. The place’s alias information can get received from the context manager or a place’s API.

The context manager can deliver contexts in at least the two following ways. In a first example, the on-device platform can register as a listener. In this case, the on-device platform can maintain a cache of updated contexts, which also means that the on-device platform is an always-on service. With a listener, the updated data can get kept in the cache.

The Benefits of Registering An Application as a Listener

1. Low latency. All contexts get cached within the on-device platform and get transformed to machine learning-friendly formats.

2. If IPCs (Inter-Process Call) to the on-device platform is at a high rate, caching contexts saves battery.

In a second example, the on-device platform can get current contexts in one shot. In this case, the context manager can provide another API to get all contemporary contexts in one go. If this API gets used, the on-device platform typically will not maintain a cache of contexts but instead gets current contexts on demand.

In this mode, the context manager gets asked to keep the updated contexts for the platform, so the platform typically obtains extra UDC permissions, which are not needed in the first option.

The benefits of the one-shot mode include that if IPCs to the on-device platform is at a low rate, it may save battery.

The on-device platform may get user permissions to all contexts listed above. Specific clients may not have the same permissions as the on-device platform has. So the platform can control the permissions for clients. This can get done by using a package manager to extract the permissions associated with a client application. The platform can maintain a mapping between contexts to permissions. Thus, a client will typically explicitly claim what contexts it wants to use when it registers itself with the on-device platform. The platform checks whether the client has the permissions to access the requested contexts. Only contexts to which the application has permission will get used to train and inference the corresponding model.

Besides, the permissions of a client may get changed on the fly. For example, a user may approve an application to use his location when he feels the application may be useful but then revoke the permission. To handle this case, each time a new event comes from the client, the platform can check their existing permissions and involve the corresponding contexts with the possibility to train.

During the inference, the platform can also analyze current client permissions to allow the context features for the prediction. The machine-learned ****** can accept missing parts for training and inference.

To use the on-device platform API, clients can add an API key to their application. The application can get the key by registering with a central authority that manages the API. For example, applications can get signed with a digital certificate for which clients hold the private key.

An Example Machine Learning Authorization

One example authorization procedure can include:

1. The on-device platform gets a package name and API key pair when a client registers to the on-device platform. The on-device platform sends the package name and API key pair to a central authority for verification.

2. Once the client gets verified, the on-device platform will generate a platform key for the client to use on that device.

3. For future API calls in that device, the client should provide the platform key to the on-device platform.

The on-device platform can check whether the platform key matches the package name and API key. But, the on-device platform typically does not send them to the central authority to verify again.

In some instances, a device can get shared by more than one user. As such, the on-device platform can extract a primary account from a list when an API gets used. The on-device platform can tie the training data to the primary account and update the related model. Suppose the model for the performance is not available on the device. In that case, the on-device platform can download the model for the client application from the cloud or use a base model (e.g., average model in Federated Learning).

A user can clear his location history or account history. In such instances, the on-device platform can remove all the corresponding contexts. The platform can re-train the model using the remaining contexts for this user in this case.

Machine Learning Platform Injecting Context Features

In particular, the patent depicts the context provider injecting context features before a training example. An example gets stored in an example collection. The patent describes the context provider injecting context features before an example gets passed to a prediction engine implementing a machine-learned model.

Example Devices Performing Machine Learning Model Training

In particular, the patent shows a background training process.

An application process can feed training data into a training data store. A training process can get scheduled in the background. Suppose allowed by certain device conditions such as, for example, idle and plugged in.

The training process can pick up a model state and training plan. It can repeatedly pick data from the cache to train the model. It can also eventually publish statistics and model update messages to a cloud server. The training phase may belong, in minutes, so the training process can suspend and resume based on changing device conditions.

It includes a federated learning service that enables federated learning. The training plan and model can get distributed by the federated learning service. The training process can perform training in the background to generate a model update. The model update can be uploaded to the federated learning service (e.g., aggregation). Besides, a quality assurance (e.g., a semi-automated quality assurance) can extract learned ****** and distribute the ****** back to the devices.

An Example Method to Generate Inferences Using Machine-Learned ******

A computing system can receive input data from a first application via a prediction application programming interface.

The computing system can supplement the input data with context features requested by the first application and which the first application has permission to access. Next, the computing system can determine a permission status for the first application relative to each context type. The input data gets supplemented only with context features included in context types that the first application has permission to access.

This computing system can use a first machine-learned model to generate at least one inference based at least in part on the input data and based at least in part on the supplemental context features. And, the computing system can provide at least one inference to the first application via the prediction application programming interface.

An Example Method to Collect Training Examples for Performing Machine Learning

prediction application

A computing system can receive a new training example via a collection application programming interface from a first application. A computing system can supplement the recent training example with context features requested by the first application and which the first application has permission to access. Also, the computing system can determine a permission status for the first application relative to each of the context types. The new training example gets supplemented only with context features included in context types that the first application has permission to access.

The example in the centralized example database according to options parameters that have been previously defined for the first application via the collection application programming interface. As one example, the options parameters can include a time-to-live parameter that defines a period for which training examples get stored.

Storing the new training example can include assigning an end period to at least a first context feature of the context features. The method can further include deleting the first context feature or the entire new training example from the centralized example database at the end period assigned to the first context feature.

The method can further include receiving a sign of a change to a permission status for the first application relative to at least one context type. In response to the transition to the permission status, deleting from the centralized example database any context features of at least one context type that gets associated with training examples related to the first application. After deleting the context features, the method can further include re-training machine-learned ****** associated with the first application using the training examples associated with the first application in the centralized example database.

A Method to Train Machine-Learned ******

retrain on device machine Learned ******

A computing system can receive an instruction from a first application via a training application programming interface to re-train a first machine-learned model based at least in part on training examples stored by a centralized example database.

The computing system can re-train the first machine-learned model based at least in part on the training examples stored by the centralized example database.

Also, the computing system can determine an update that describes parameters of the re-trained model or changes to the parameters that occurred during the re-training of the model.

And, the computing system can send the update to a central server computing device for aggregation with other updates provided by other computers.

Sharing is caring!



Source link

What happens to crawling and Google search rankings when 67% of a site’s indexed urls are pagination? [SEO Case Study]

By | October 12, 2021


This website uses Google Analytics to collect anonymous information such as the number of visitors to the site, and the most popular pages.

Keeping this cookie enabled helps us to improve our website.

This site also uses pixels from Facebook, Twitter, and LinkedIn so we publish content that reaches you on those social networks.



Source link

How to Speed up Your Website: 15 Quick and Easy Fixes

By | October 12, 2021


How to Improve Page SpeedIn today’s post, I’m going to show you how to quickly and easily decrease your website load time (and why it’s important).

In fact:

The techniques I’m about to reveal are the very same strategies I used to increase my page speed from a fully loaded time of 3+ seconds to 1.2 seconds sharp. 

But here’s the best part:

None of these website loading improvements require developer know-how and can be implemented by even a novice SEO super fast.

If you want to know increase your site speed, rank higher, boost engagement and convert more traffic, this post is for you. But first, why is a fast webpage load time important?

Website Load Time Statistics

Website Load Time Statistics

  • 59.2% of users will leave a website in the first 6 seconds if it is slow to load – Unbounce
  • As a page load time goes from 1 seconds the probability of a bounce increases 32% – Think With Google
  • Decreasing load time by 0.1 seconds leads to the bounce rate of lead generation pages improving by 8.3% – Google, Fifty-Five, and Deloitte
  • When load time is reduced by just 0.1 seconds:
    • Retail customer engagement increases by 5.2%.
    • Bounce rates on product listing pages in the retail and travel categories improve by 5.7% and 5.4% – Google, Fifty-Five, and Deloitte
  • Website conversion rates drop by an average of 4.42% with each additional second of load time between zero and five seconds – Portent
  • Increasing site load time by one-tenth of a second results in a 7% decrease in conversion rates – Akamai
  • Decreasing site load time by one-tenth of a second results in an 8.4% increase in conversion rates for retail sites and a 10.1% increase in conversions for travel sites – Google, Fifty-Five, and Deloitte
  • If an eCommerce site loads slower than expected, over 45% of people admit they are less likely to make a purchase – Unbounce
  • Around 70% of people say that the speed of a page affects their willingness to buy from an online retailer – Unbounce
  • 77% of smartphone shoppers say they are more likely to purchase from mobile sites that offer them the ability to make purchases quickly – Think with Google
  • Retail sites see a 9.2% increase in average order value when load time decreases by one-tenth of a second – Deloitte
  • The mean time it takes a mobile webpage to visually complete loading is 21.6 seconds – Backlinko
  • The mean time it takes a desktop webpage to visually complete loading is 8.2 seconds – Backlinko
  • The average page loading speed for a fully loaded web page is 27.3 seconds on mobile – Backlinko
  • The average page loading speed for a fully loaded web page is 10.3 seconds on desktop – Backlinko
  • The average web page takes 87.84% longer to fully load on mobile vs. desktop – Backlinko
  • Slow-loading mobile pages result in two times higher bounce rates and 27% fewer page views when compared to desktop pages – Perficient

Now you understand the importance of a fast-loading website, let’s cover what page speed is (precisely) and how you can improve it.

What is Page Speed?

Page speed is the amount of time it takes for a webpage to load.

Page Speed Definition

Otherwise known as loading time, page speed is measured by the interval between a browser requesting a webpage and when the browser renders the page.

A page’s loading speed is influenced by various factors such as the distance data travels, the website’s server speed, the complexity of its code, the page file size, and the connection type.

Not to mention, file compression, caching, render-blocking, and dozens of other factors that affect page speed and SEO.

As you are beginning to tell, improving page speed isn’t as simple as installing a page speed plugin.

There are numerous, sometimes complex factors that contribute to how fast – or slow – your website loads.

Not just that, there’s also no “one way” to measure the load time of a page, there are many.

Here’s the three most common:

Page Load Time

This is how long it takes for the page to load in full.

Time to fully loaded measures the loading of all the visible parts of a page, as well as the hidden elements like scripts and code – basically it measures the load time of 100% of all resources on the page.

It’s the simplest page speed metric to understand, but not necessarily the most useful, since the page may appear loaded to the user (and actually work) long before the page loads in full.

Largest Contentful Paint

Largest Contentful Paint (LCP) is a more valuable metric to determine the load time of a page from the point of view of a user.

It measures the time a webpage takes to show the user the largest content on the screen.

Typically that’s the page headline, the featured image, or main text content.

In short, LCP measures the time for the important above-the-fold content to be fully visible to the user.

Largest Contentful Paint

For example, let’s say a page takes 10 seconds to fully load.

The page wouldn’t “appear” to load slowly as long as the first contentful paint occurs within a couple of seconds.

In short, you can get away with a slow fully loaded time as long as your largest contentful paint is fast.

Time to First Byte

Simply put, this is a measurement of how long a web browser has to wait before receiving its first byte of data from the website’s server.

TTFB doesn’t really measure how long a page takes to load, it does however influence page load time.

The longer it takes for a browser to receive data from the server, the longer it will take to display a page.

How to Calculate Page Load Time?

There are lots of different methods to measure page speed:

  • Time to First Byte (TTFB)
  • Largest Contentful Paint (LCP)
  • First Meaningful Paint (FMP)
  • Time to Interactive (TTI)
  • First Input Delay (FID)

The list goes on…

And because not one of these metrics is any better than another.

I recommend you work on speeding up your website across all of them.

Don’t worry. In a moment I’m going to show you how, but first:

Why Is Page Speed Important for SEO?

There is no denying it, Google is obsessed with page load speed.

Page Seed as Ranking Signal

Ever since 2010, Google has used site speed (and therefore page speed) as a ranking factor in its search engine algorithm.

Then in 2018, Google rolled out the “Speed Update” which further impacted the search engine rankings of slow loading pages.

And most recently in 2021, Google introduced Page Experience of which a major component was core web vitals and loading speed.

Page Experience Ranking Signals

Google cares a lot about page speed – and why wouldn’t they?

If Google is going to send a boatload of free traffic to your website, they want users to enjoy a GREAT experience. A huge part of that is a fast-loading webpage.

Simply put:

A slow-loading website will damage your Google rankings.

A fast-loading website will help your Google rankings.

The only question is, how do you go about improving your page load times? We’ll cover that next.

How to Speed up Your Website: 15 Best Practices for Decreasing Web Page Load Times

There is no escaping it, a lot of what contributes to page speed is deep-rooted in code and highly technical.

Creating AMP Pages, minimizing HTTP requests, and using asynchronous loading – all require developer-level know-how. And, because of that, I won’t be discussing those page speed optimizations in this post.

Instead, I’m sharing 15 easy-to-implement (yet often forgotten) ways to boost your site speed like never before.

Let’s jump in:

(1). Perform a Page Speed Test So You Can Benchmark Your Loading Time (and Know What to Improve)

Before you attempt to improve your page speed, it’s helpful to know how fast – or slow – your website loads.

After all, if you don’t know your current page speed, how will you know if your speed optimizations are successful?

You can use several tools to assess your page speed improvements, and I recommend running your site through all.

But a quick word of warning:

Your scores will vary wildly from tool to tool.

That’s because each web page speed test tool uses different methods to assess performance. Not only that, the locations of their servers are in varying cities – and as you’ll find out later – server location has a HUGE impact on website load times.

There’s no need to get caught up on the specifics. For now, just run a test with each tool and note down your scores.

Pingdom

An easy-to-use site speed test that mimics the way a page is loaded in a web browser. Pingdom currently offers seven server locations from where you can test your site.

Unlike other tools that measure the load time according to when the page is fully loaded, Pingdom uses On Load Time which records when all resources on the page have been downloaded. At this point, the page would be interactive, but scripts may still be running in the background.

Pingdom Page Speed Testing Tool

GT Metrix

A page speed testing tool that offers both desktop and mobile measurement and the option to test across multiple browsers. GT Metrix provides various server locations, including 7 on the free plan and a further 15 on the pro plan.

The default measurement is fully loaded time, with on-load time optional.

GT Metrix Website Speed Testing Tool

WebPageTest

WebPageTest is another great tool that shows you the speed and performance of your website in different browsers.

The WebPageTest has the most extensive number of test locations, with 38 in total. Like GT Metrix, WebPage Test uses fully loaded time by default, with on-load time (also known as Document Complete time) optional.

WebPageTest

PageSpeed Insights

This is Google’s tool for assessing page speed performance. Unlike the three tools mentioned above, Google uses field data instead of lab data in its scores – assuming field data is available.

What this means:

Instead of running a test that simulates the load time for a real user, Google PageSpeed Insights collects real user data from Chrome browser users. This data is aggregated over 30-days, which means that you may not see an improvement in your score until 30-days after you optimize your load time.

PageSpeed Insights

It’s important to know Google’s assessment of your load time. It’s this score they use for Core Web Vitals – a key part of the Page Experience signals that contribute towards search engine ranking performance.

That said, the other three tools (Pingdom, GT Metrix, and WebPageTest) are more useful for a quick before and after measurement of your page speed performance.

A couple of pointers:

Firstly, select a test location near your target audience. If your test is run from the United States, but your users are in Australia, you won’t get a realistic measure of page speed performance.

Test Location Webpagetest

Secondly, don’t make the mistake of testing ONLY your homepage.

Pagespeed is unique to each page.

I suggest you test all your most important URLs and use Google Analytics to identify your slowest loading pages quickly.

Google Analytics Page Timings Report

Tip – you can find ‘Page Timings’ under ‘Site Speed’ in GA’s ‘Behaviour’ menu.

Got that? Cool.

Let’s begin to optimize our page speed.

(2). Switch Out Your Free Domain Name System for a Lightning Fast DNS Host

Think of a Domain Name System (DNS) host, like a phone book.

It maps your human-readable domain name to an IP address (where your website is actually located).

Domain Name System Host

In short:

Each time a user types your web address into their browser’s URL bar, the browser does a DNS lookup that converts the web address into an IP address, and by doing so, locates the server where your website is hosted.

By default, you’ll be using the free DNS service offered by your domain registrar.

NameCheap, GoDaddy, Google Domains (and almost every other domain registrar out there) provide you with the capability to set up your nameservers and route your domain to the IP address of your web host.

The trouble with these free DNS services?

They’re really slow.

If you want a rocket-fast website, I recommend switching your DNS to Cloudflare.

I shifted from Namecheap’s DNS service and went from a query time of 26 ms:

Namecheap DNS performance

To a staggering 13 ms.

Cloudflare DNS performance

In other words, I cut 50% from the DNS processing time.

Best of all, it didn’t cost me a dime.

Once you’ve signed up for a free Cloudflare account, you simply login into your domain registrar and replace the default nameservers with Cloudflare’s.

Change Nameservers to Cloudflare

And with that, you are done.

(3). Implement This Caching Double Whammy and Accelerate Load Times

Here’s what (typically) happens when a user lands on your website:

  • Their browser contacts your web server
  • Your content management system pulls the latest data from your database (like your recent posts and pages)
  • Your web server compiles the data in to an HTML page and serves it to the visitor
Browser and Database Connection

A caching plugin removes the first two steps entirely.

To avoid a new request being made to your server every time a visitor views a page, a caching plugin will save a prebuilt version of your web page and serve it to your visitors.

How caching works

And in doing so, it speeds up the load time for users.

My go-to recommendations for WordPress caching plugins are:

Since a caching plugin removes database lookups, it can cut 1-3 seconds off your website load times.

Not bad.

But if you want your pages to load crazy fast, I recommend this:

Edge Caching (CDN)

Unlike page caching, which stores a cached version of your website on your central web server…

Edge caching stores entire HTML pages on “edge nodes” located on a server network spanning the globe.

What this means:

Instead of a request being made to a server on the other side of the planet…

When a visitor requests to open a page, it gets delivered from a server located close by.

How Edge Caching Works

Like page caching, a prebuilt version of the web page is saved for immediate access, and a call to the database is not required.

As you can imagine, if data only needs to travel a few hundred kilometers versus thousands – and the page is prebuilt ready for users – it’s going to result in faster load times.

Not only that, since edge caching does a tonne of work, it also takes the pressure off your web hosting, which can improve your site speed further.

The Edge Caching service I recommend you use is Cloudflare’s CDN:

With 200+ servers around the world:

Cloudflare Edge Caching Network

And, at only $20 per month, it’s an absolute steal.

(4). Optimize Your Images to Reduce Huge Payloads

Nothing slows a website down like bloated images.

And, it’s hardly surprising.

Images make up a huge chunk of the page size.

Images Contribution to Page Weight

(Nearly 40% of the total page weight for our SEO services page).

So the more that you can optimize your images, the faster your page will load.

The only question is how?

Firstly you want to resize all your images to the largest size they need to be – and no larger

For instance:

The images on this page have a maximum width of 640px.

If I were to load an image 650px wide, I’d have added unnecessary weight to the page.

Not only that, since my CMS would need to resize the image to 640px wide to fit the dimensions of the page, load time would be slowed down even further.

With that in mind, I recommend you resize your images to the max they’ll need to be BEFORE uploading them to your website.

How to compress image files

You can use an online image editing tool like Pixlr or a tool native to your computer. The mechanism doesn’t matter – you need your images the same size as the maximum viewable area.

The second thing you want to do is choose the right file type.

Selecting the wrong file format can add extra weight or result in images being low quality.

Unless your images are animated, PNG or JPEG is the lightest – and therefore fastest to load.

Here’s how to know which file type to choose:

How To Choose The Right Image File Type

Thirdly, you want to strip out all the unnecessary bloat from your image file while maintaining as much quality as possible.

Don’t worry about the technicalities.

Just install Imageoptim and let the tool do the rest.

Image optim

It can help reduce your file sizes by about half.

But wait!

What if your images are already loaded?

Good news. If you are on WordPress, I have a solution:

WP Smush!

WP Smush automatically compresses any image you upload to the WordPress media library.

WP Smush Image Optimization plugin

And reduces the file size by more than 2 times, without any noticeable drop in visual quality.

Pretty cool, right?

Once you’ve compressed your image file sizes, there is one more image optimization trick you can use to improve your webpage load times.

(5). Implement Lazy Loading (But Not on Your Logo)!

Even if your images are light as a feather, your page weight can still be heavy if you use a tonne of graphics in your posts (like me).

Not only that, if every image needs to load for the page to render – your page loading time will be prolonged.

Enter lazy loading.

Lazy loading is where the browser delays the loading of images until they need to be shown on screen.

In other words, not all images are loaded at once.

Some images (and other assets like videos and embeds) are postponed and loaded ONLY when needed—i.e., as users scroll down to that section of the page.

If you watch my blog posts closely, you’ll see lazy loading in action:

Lazy loading is powerful. It can really speed up a site. But, there’s one BIG issue with lazy loading:

When it’s turned on, it lazy loads ALL the images.

Since most sites – like mine – have images that should display on load, using lazy loading can actually hurt the largest contentful paint.

The point here is that default lazy loading settings will apply lazy loading tags to all of the images on the page, including those at the top that should immediately display to the user.

Which means:

Images “above the fold” have to wait for the lazy load JavaScript library to execute before the images at the top of the page can load. In other words, the lazyload JavaScript blocks the rendering of assets needed for a fast largest contentful paint (LCP).

To address this problem, grab the URL of your logo and any other above-the-fold images (Hint! Your featured image) and add it as an exclusion in your lazy loading plugin.

I use WP Rocket, and here are the settings:

Lazy Load Logo Exclusion

Whether you use WP Rocket, or Autoptimize (another great plugin for lazy loading), be sure to defer the loading of images and set your logo as an exclusion.

With that, on to another thing which should be set to “defer loading.”

(6) Defer JavaScript Files and Increase Your Page Speed

If you’ve ever run your site through PageSpeed Insights or GT Metrix, you’ll be familiar with this warning:

Defer JavaScript Passing Warning GT Metrix

There is no escaping it. JavaScript is one of the biggest culprits affecting website load time.

But why?

When someone visits your site, your web server delivers your website’s HTML content to the user’s browser.

The user’s browser then works through the HTML from top to bottom to render the page.

If it finds any JavaScript during that process, it stops rendering the rest of the page until it has fetched and parsed the complete JavaScript file.

Since script files are big and can take some time to load, they can really affect loading times.

And here’s the deal:

Most JavaScript files like Live Chat, popups, widgets, or even tracking scripts are not needed for the page to properly render.

And that’s why I recommend you delay the parsing of JavaScript files until after the page has loaded.

This way, all the essential content on the page gets loaded, and then scripts execute later.

If you use WordPress, both WP Rocket and Perfmatters allow you to delay JavaScript rendering.

In case you are curious, here are my Perfmatters settings:

Perfmatters Delay JavaScript

The scripts you delay on your site will be unique to you.

A quick review of GT Metrix recommendations will show you the culprits slowing your load times.

(7). Upgrade Your Hosting

Not all web hosts are made equal. Far from it.

The difference in response time between a fast and slow hosting provider can easily be 800 milliseconds.

It might not sound much, but that’s a 4X difference:

Hosting Speed Comparison

I’m not going to call out individual hosting providers because I’ve only tested a handful of them, but the chances are if you spend $4.99 per month on hosting, your site isn’t going to load quickly.

That’s because most budget hosting plans put your site on the same (overburdened) server as hundreds of other websites – leading to slowdowns, lags, and even downtimes.

I can’t speak to the performance of most hosts out there, but as a general rule-of-thumb, you get what you pay for.

Still, if you are looking for hosting suggestions, here are two premium hosts I use and can personally recommend:

  • Cloudways – Managed cloud hosting on a speed-optimized server stack to provide with 60+ locations around the globe. My go-to recommendation for clients.
  • WP Engine – Managed hosting for WordPress websites with page caching at the server level. The same webhost this site is hosted on.

Whether you go with my recommendations or another, you should look for a couple of factors.

Firstly, you want a hosting provider that uses a solid-state drive (SSD).

Solid-state drives are way faster than standard hard drives, which means shorter page load times.

Secondly, go with a host with a server located in your country – or at least your continent.

How server location impacts page speed

Websites typically perform 1+ seconds slower outside of the country where the server is located.

With that stat in mind, if you’re currently on a cheap shared server located far, far away, it’s time to upgrade to a dedicated, local SSD server.

(8). Clean and Compress Your Code

In short, you want to minify the resources found on your page.

Minification (if that’s actually a word) is the process of removing unnecessary spaces, characters, and comments – in fact, any unneeded elements – to reduce the size of files.

Lighter files mean less data to transfer – and faster load times.

The process of minifying files is typically focused on JavaScript and CSS but can also include HTML.

But, before you freak out at the thought of combing through every line of code on your website to find redundant elements, don’t worry.

You can easily minify your code by installing WP Rocket and ticking a couple of boxes:

WP Rocket Minify settings

WP Rocket will minify your code automagically!

A second method to reduce files is GZIP compression.

GZIP works similarly to the ZIP and RAR compression you use on your computer but is handled by your server.

Basically, when a visitor lands on your website, their browser requests the site’s files from your web server. When GZIP is enabled, your server will compress those files before sending them to the user’s browser.

And, because those files are compressed and lighter – the transfer is faster.

Since I use WP Engine hosting, GZIP is already enabled.

If your hosting doesn’t support GZIP compression, you can enable it via WP Rocket or W3 Total Cache.

Or, do it manually if you’re skilled in this kind of thing.

(9). Ditch Your Outdated HTTP to Take Advantage of the Faster HTTP2 Protocol

I can’t believe I’m saying it in this day and age.

But if you are still running your website on HTTP (as opposed to HTTPS) protocol, make the shift now.

You probably know that HTTPS is a ranking signal.

HTTPS as ranking signal

Basically, if your website uses a secure and encrypted HTTPS protocol, Google will give your site a small ranking boost. On the contrary, if you use HTTP, your ranking performance gets dampened slightly.

But that’s not the only reason to get an SSL certificate and switch to HTTPS hosting.

Did you know that HTTPS is also faster?

When your site is running in HTTPS mode (encrypted), your web browser software will use the newer HTTP2 protocol (assuming your host supports it), which is significantly faster than the earlier HTTP 1.1 protocol.

In fact, using HTTPS will allow your site to download 14.3% more quickly:

HTTP Vs HTTP2 speeds

Not too shabby!

Before hastily switching to HTTPS with your current web host, consider that many cheap hosting providers still don’t support HTTP2.

If yours doesn’t, it’s a surefire indication their infrastructure is outdated and yet another reason to jump ship to a premium hosting company like Cloudways, WP Engine, or Kinsta.

PRO TIP: Subscribe to Cloudflare’s 0-RTT (zero round trip time) feature, which speeds up the HTTPS encryption negotiation even further!

(10). Use the Highest Version of PHP Your Website Will Allow (And Lock in Faster Processing Times)

PHP is the programming language used by popular CMS like Joomla, Drupal, and WordPress.

Believe it or not, it’s the codebase used by 4 out of every 5 websites:

Website Programming Language Usage

One good thing about PHP is that it is continually updated for enhanced security and improved performance.

Infact, new versions of PHP get released every 6-12 months or so – and each new PHP update is typically is 10-30% faster than the previous version. 

PHP Version Processing Speed

Each version operates faster than the last due to an increase in the number of database requests it can process at once.

For instance, comparing PHP 7.3 to PHP 5.6, the later version can handle 3 times more transactions per second.

If your site is on WordPress, it’s a no-brainer decision to update your site to the latest PHP version.

Specialized hosting providers like those already mentioned update your PHP version automatically.

Alternatively, you can check your Site Health settings under the Tools menu in WordPress. From there, you will see the PHP version you are running:

WordPress PHP Version

And can update it accordingly.

(11). Use This Service for Video – It’ll Speed Up Your Website Significantly

There’s no denying it; videos are an exceptionally good way to engage your audience.

But the trouble is, they take up a huge amount of server resources to load.

And, as you are beginning to learn, the less strain on your server, the better, because the faster your page speed will be.

So, in the case of video…

Instead of self-hosting videos on your site, use a standalone video hosting service like YouTube, Vimeo, or Wistia.

Then add the videos to your website using an embed code.

Wista embed code

This way, your server won’t get occupied loading heavy video files and can use the time to process other elements on the page.

But of each of these services, which one is better?

According to WP Rocket, Vimeo is marginally faster than YouTube and Wistia – and MUCH faster than Daily Motion, the slowest player.

Still, the difference between these top-3 is minimal, which is why I opt for Wistia for most of my videos.

It has outstanding analytics and supports video site maps – which is a great feature for SEO.

(12). Find (and Fix) 3XX Redirects

301 and 302 redirects are useful.

They’re used to redirect a redundant page to an alternate working URL.

301 and 302 redirects explained

The trouble is they are dastardly for page speed.

Why?

Because whenever a user attempts to access a 3XX status code page, two requests are made to the server.

One for the original (redirected) URL and another for the final destination URL.

That can mean a significant increase in website loading time.

But here’s the thing…

You cannot do away with 3XX redirects entirely. That would be massively detrimental to your SEO.

However, you can minimize unnecessary 3XX status codes:

(a). Internal Redirects

Internal redirects occur when a page with internal links pointing to it gets redirected to another page.

For instance:

On your homepage, you link to an article about the keto diet (e.g., healthyfitness.com/keto-diet/)

Later you decide to create a new article on that topic and redirect the old article to the new article (301 redirects from healthyfitness.com/keto-diet/ to healthyfitness.com/new-keto-diet/)

Great job – that’s good practice for SEO.

The trouble is, the internal link you placed on your homepage now goes through an unnecessary “hop” – and while it works for the user – it’s adding additional load time for those following that link.

Redirect Chain

Thankfully, there is an easy fix. Simply update the link at the origin to the final destination URL.

A simple way to find redirects on your site is to install Check My Links. This browser extension will crawl your page and highlight any links that are redirecting.

A quick scan of the SEO Sherpa homepage identifies multiple internal redirects in the footer:

Find and Fix Internal Redirects with Check My Links

All I need to do is log in to WordPress and update those footer links to the final destination.

(b). Redirect Chains

Redirects are bad for load time, but they are even worse when they are in a chain.

Chains of redirects do happen when one redirected page redirects to another redirected page – and so forth.

They are normally an outcome of changes made to a site over time.

For instance:

  • domain.com/about/ updated and 301 redirected to:
  • domain.com/about-us/ updated and 301 redirected to:
  • domain.com/about-our-company/ updated and 301 redirected to:
  • domain.com/company-history/

Now anyone trying to access the original page will get redirected not once but three times.

I like to use Ahrefs Site Audit to find redirect chains. After crawling your site, it will identify the number of redirect chain issues you have (along with other errors).

Redirect Chain in Ahrefs

You can drill into the list of redirect chains to see the chains that occur:

Ahrefs Redirect Chains

To fix each chain, update the 301 redirects at the origin URL with an absolute link to the final destination – thus removing the chain of redirects in the middle.

(13). Eliminate 404 Errors

It happens!

Files get moved (or go missing), and links go broken.

Broken Links Affect Page Speed

When this occurs, users will experience longer page loading times due to additional web server requests to find the missing files.

Eventually, leading to a 404 error.

The causes of these errors are numerous and inconsequential.

All you need to know is how to find them:

One way is to use Google Search Console to find and repair them.

Once logged in to your GSC account, click on Coverage and then Error.

Click on Submitted URL not found, and you will see a list of all the links resulting in the 404 error:

404 Page Not Found Google Search Console

Alternatively, you can use a site audit tool like Ahrefs or Screaming Frog.

Whatever tool you use, know that you need to identify links to 404 pages and media files.

Then replace or remove those broken links.

Doing so will speed up your site.

(14). Cutback, Update and Upgrade – Three Steps to Improve Your Page Speed with Plugins

When it comes to plugins and extensions, the fewer the better.

Most add-ons load JavaScript and CSS.

And, as I already pointed out these two things can put your website in go-slow mode.

So, wherever possible, I recommend you strip back your plugins and use ONLY what’s (absolutely) essential.

Anything, that isn’t necessary for your site’s functionality turn off and then delete.

Of course, some plugins affect site speed more than others.

So even if you absolutely need a plugin, there may be a faster alternative.

Slow WordPress Plugins
New Relic can be used to identify slow plugins on your site.

This post by ServeBolt lists the WordPress plugins most affecting load time performance – and lists better, faster options.

You should make the switch if you find quicker alternatives.

The third step in speeding up your website via plugins is to ensure you are running the latest plugin versions.

Very often, older plugins have incompatibilities with the current version of WordPress or PHP.

These incompatibilities almost always impact page speed performance.

Thankfully, the plugin menu in WordPress makes it easy to spot outdated plugins:

Perfmatters Plugin

From there, you can simply upgrade to the latest version.

(15). Use This Little Known Google Tag Manager Trick to Speed up Your Website

You’ve probably heard of Google Tag Manager? And, hopefully, you are using it on your website.

Google Tag Manager is a container for all your pixels and tracking scripts, which makes deploying codes simple – and bypasses the need for a developer.

Simply moving your tracking pixels from embedded in your website code to Google Tag Manager can shave up to 2 seconds off your load time.

This is a great page speed improvement, but it doesn’t stop there…

Did you know that Google Tag Manager, allows you to control when your tags are fired and thus speed up your load times even further?

As you’ve learned already, any code that’s not required for a fast largest contentful paint should be delayed until after the main content has loaded.

By default, Google Tag Manager will fire your tags on page view.

GTM Window Loaded Trigger

But, by selecting the ‘Window Loaded” trigger your tags are fired once the entire content of the page has been fully loaded, including any embedded resources such as images and scripts.

SEO Sherpa Tags

I set this up on the SEO Sherpa website, and because I have a bunch of tracking set up, it boosted load times significantly.

To implement this page speed trick on your own website, login to GTM and select ‘New Trigger’ and then ‘Window Loaded.’

And once you’ve done that, you’re done.

What Did You Think?

Now I’d like to hear from you:

Which technique from today’s post are you going to implement first?

Are you going to fix broken links? Or implement lazy loading?

Either way, let me know by leaving a comment below.



Source link

10 Reasons Why Your Content Doesn’t Attract Links – SEO Consultant

By | October 12, 2021


Photo by Judit Peter

I originally published this post on 08/15/2012 on my friends’ site CopyPress.com. They are no longer using it, so I have brought it back to life here. Enjoy!

So we have all heard time and time again, “to attract links you need to build great content“. But very few actually talk about what good content looks like. That’s because good content can come in many different forms. But bad content (that doesn’t attract links) usually follows some of the same patterns. Below are 10 reasons why your content might not be attracting links.

1) Bad Title

A bad title is often times the first barrier to attracting links. The title pulls the audience in and makes them want to learn more. Without a good title most users won’t even click. Also good titles can set the tone for the rest of the content and define what type of back links you want. For example if you are looking for the term “tennis shoes” to be used as anchor text in back links, you better make sure to include it in your title. People that link out pull from the original content for ideas when linking, so include all the right terms in a snappy title to attract the right links.

2) Bad Design

I know you are probably confused with this one, but the truth is, pages that have bad designs or poor user experiences attract less links. People usually only link to pages that have an established sense of authenticity or trust. Bad designs destroy a user’s trust. User interfaces that have clean white space and well-formed navigation tend to attract more links.

3) No Hook

How do you pull a fish out of the water? You use a hook. You can also use hooks to pull in links. A hook is anything that peaks a readers interested to learn more and share it with their audience. This can be talking about a new idea, taking a controversial stance, or telling a good story. Either way without a hook most won’t link out.

4) No Point of Difference

So you wrote a post about iPhone apps for real estate? Awesome! The only problem is that so did everyone else. If you want your content to attract links you need to make sure there is a clear point of difference. This means offer something that no one else does. An example would be for this iPhone real estate post would be to ask a handful of successful real estate agents their favorite iPhone apps, compile their responses into an interview style post. Now when someone is looking for a post about iPhone apps for real estate, they are likely to choose yours because it’s different.

5) Too Long

We now live in a world of 140 characters and text messaging. Unfortunately, our attention spans are dwindling and as a result it is incredibly hard to attract links to content that is overly long. Now, don’t get me wrong if it’s well organized and all really good quality, then long content can attract links. But if you have little to say then brevity is key.

6) No Social Traction

“If you build it, they will come.” No, you aren’t Kevin Costner standing in a corn field! And because of that you need to start driving traffic at your content if you want links. Having great content alone won’t build links, you need to also make sure it has traction in social media to get it in front of the right people. Otherwise it will sit on your blog unread, and invisible to those that link out.

7) No Unique Voice

If you have ever read anything by, or spoken to Lisa Barone, you know she has a unique voice in her writing, and her speaking. This unique voice is in part the secret to Lisa’s success. Her ability to write in a way like no one else captivates her audience, and builds a brand that people regularly link to. When creating content, if you can sustain a familiar “voice” coupled with high quality, readers are more likely to link to your content because you have built a familiar trust.

8) Bad Topic

Do you know how many people are talking about Doctor Who online? A Lot! And how many are talking about “Cat Organs“? Not very many people at all. Therefore, you would have a much better chance attracting links with a post about Doctor Who, then a post about a creepy instrument made out of cats.

9) Not Engaging

Which post do you think would get more links: “Joe Hall Eats 23 *** Dogs in One Day” or “How To Eat 23 *** Dogs In One Day”? Unless you know me personally, then you are likely not going to care very much about reading the first blog post. But the second post is engaging because it teaches the reader how to do something. Engagement is all about making your content personal. Talking about yourself or your company doesn’t engage anyone. Engagement = More Links.

10) Not Focused

Have you ever read a blog post title that reads like a random list of words? For example, “What Skittles, The IRS, Katy Perry, and Jazz Music, Taught Me About Breast Feeding”. Seriously? If your content reads like a free flow of your inner thoughts no one is going to link to it and very few are going to read it. Content that gets linked to is focused on one, or two topics.

One last final point that needs to be remembered: If you are writing content to attract links, focus on giving bloggers a reason to link to you. If you can zero-in on a specific reason, and leverage it correctly, then building links should be easy.

Joe Hall is an SEO consultant focused on analyzing and informing the digital marketing strategies of select clients through high level data analysis and SEO audits.





Source link

How User Experience (UX) Can Impact Your SEO

By | October 12, 2021


User Experience (UX) is being touted as one of the most critical factors for effective SEO in 2021. With the average attention span of internet users seemingly decreasing by the day, and a generation that expects instant results, a website’s speed and ease of navigation are of utmost importance.

In this post, we’ll be looking at UX, what it is, Search Engine Optimization, how to improve it, and why it’s so essential for every website owner. But, first, check out our passing Web Core Vitals Score.

UX and SEO

What Is User Experience?

The phrase is relatively self-explanatory as it’s all about the experience a website visitor receives when they view and navigate your site. A positive UX is made up of several factors including, but not limited to, the following:

  • Website loading and page loading speed
  • Easy and logical site navigation
  • Minimal ads and pop-ups
  • Quality content
  • Text with plenty of white space and broken up into subheadings
  • Content that can be quickly scanned
  • A website that’s easy on the eye and pleasant to view
  • Keeping it simple
  • Having a page of frequently asked questions (FAQ)
  • A clear call to action (CTA)
  • And more

 In essence, the goal is for the visitor to have an excellent visual experience and navigate your site easily. It sounds easy, but it’s not that simple as other criteria are needed to produce the maximum experience, results, and conversions.

How To Achieve Passing Scores for Web Core Vitals?

The numbers below are SEO Inc’s current Scores for Web Core Vitals. The scores encompass the elements of Page Experience. If you have issues reaching numbers like ours, you are not alone. SEO Inc prides itself on being the top technical SEO Company in the US. If you need assistance, don’t hesitate to reach out to us, you can schedule a call with us here.

Who Really Cares About User Experience?

Visitors care about excellent UX, so basically everybody. Dwell time will be improved on a website that offers a positive user experience. Present them with a page loaded up with flashing ads, loud colors, massive blocks of text, and the visitor is likely to click away and go somewhere else.

Site Speed is also part of tweaking your UX, as long load times need to be a focused effort to fix regarding SEO, and it is an ever more critical aspect in 2021. True, this will also be dependent on the visitor’s internet speed, but even a fast website load time won’t make that much difference if a site has errors, long load time, slow TTFB, and is sluggish when navigating. The aim is to make sure each page on the website opens quickly and seamlessly. It’ll keep visitors on your site and reduce your bounce rate, which is a good thing for your Google rankings.

Besides Google, Bing also cares about the UX of websites, and we’ll take a closer look at that now.

Does a Poor UX Lead To Lower Search Engine Rankings?

UX  User Experience and SEO

Google especially is placing more and more emphasis on website owners providing a positive user experience for visitors, and it can affect your rankings, positively or negatively. At SEO Inc, we confidently state UX will affect your rankings, conversion rates, and bounce rates. And that adds up to gaining or losing traffic, rankings, and new clients.

One area Google and Bing are focusing on is site speed. We talked about it earlier, and it’s worth mentioning again. Sites that are slow to load will be disadvantaged when getting higher placements in the SERP results. It’s not something that will bring a manual penalty or lead to your site being deindexed, but it can lower your rankings and increase your bounce rates.

Google aims to deliver the very best results for a search query. There are many factors involved in how Google arrives at these results. Quality content is always vital, as are natural backlinks. UX ranks right up there, as quality overall is becoming the big thing in the search engine world.

Suppose people are visiting a website and immediately bouncing out of it. In that case, this is an indicator to the search engines that the site is not offering a good user experience in some form or other. It could be something as simple as the content not being what the visitor expected to find, the site looking like a jumbled mess, and far too slow to open.

Every website owner needs to look at their website from a visitor’s point of view to tweak, test, and make design changes or adjustments that will improve the UX.

Why Is Technical SEO and UX Also Important?

Technical SEO and UX help with your site’s overall SEO and ranking performance. It also adds to the user experience because the website functions correctly and delivers the promised content discovered in search results.

  • Technical SEO can cover things like:
  • The site’s initial interaction with the SERPs
  • Overall quality and relevance of the landing page
  • Page loading speed
  • JavaScript frameworks
  • And so much more

The user experience can even go one step further by providing ways to interact with your site and your brand even after a visitor has left. This could involve inviting them your join your email list, re-marketing, drip campaigns, a social media presence for your brand, and so on.

How To Improve Your Website’s User Experience

In the modern world, every website owner should at least get familiar with white hat SEO and UX basics to remain competitive moving forward.

We mentioned earlier that one of the best ways to improve the user experience for your website is to view it from the angle of your visitor’s personas:

  • What are they hoping to find when they get there?
  • Is the content useful and easy to read or scan through?
  • If you were a visitor, would you like to spend time on your site?
  • Are the pages filled with too many distractions?
  • Is the menu and navigation logical, making it easy to locate information, products, or resources?

It’s questions like the above that site owners need to be asking themselves about their sites, as only then can they hope to improve their website and offer the best UX possible.

Enlisting the help of industry experts who fully understand the mechanics of SEO and high-quality user experience will also help no end. Unfortunately, not all webmasters and online business owners will have the time, knowledge, or inclination to improve the UX of their sites effectively, so you must call on someone who does.

The Takeaway

Moving forward, search engine giants like Google will be factoring in indicators of a positive user experience more and more every time they do an algorithm update. So if you want your website to stay ahead of the curve and competition instead of lagging behind, then focusing on providing an exceptional UX for all your visitors is of paramount importance.



Source link

Why Long-Term SEO is Worth It: A COVID-19 Case Study

By | October 12, 2021


When the pandemic first hit in early 2020, did your business pull back on marketing efforts? 

If so, you’re not alone; global ad spend fell by $63.4 billion last year, and digital marketers reported budget cuts to be their top challenge in a post-COVID-19 world.

But, like they say, hindsight is 20/20. With more than a year and half of COVID under our belts, this might not have been the best course of action for every business. 

In fact, when it comes to organic traffic results in 2021, we’re seeing top performance from our clients who stayed the course — continuing to invest heavily in search engine optimization efforts, despite the uncertainty around their industry.

In this case study, we’ll tell you about two of them. Both involve industries that were heavily impacted by the pandemic, shutting down for months and leaving owners frantic about the future. 

But, thanks to a steady SEO hand, they’re now reaping the rewards of their discipline and marketing decisions.

Client #1: The Local Lawyer

One of our clients — a criminal defense attorney — has employed our SEO team’s strategies since 2013. While unable to devote a huge budget to these efforts, the firm has nonetheless seen organic traffic improvements over the years of our partnership.

Unfortunately, that all changed in late 2019. The lawyer’s site was hit hard by a Google core algorithm update in September of that year, likely due to its positioning as a “Your Money, Your Life” legal website. Due to its thin, low-quality blog content, we concluded it was a victim of Google’s ongoing E.A.T. initiative.

In a double stroke of bad luck, COVID-19 emerged just a few months later. With courthouses and legal proceedings on pause, organic traffic to the client’s site dropped off and representation inquiries stopped coming in.

Google Analytics pageviews graph from Sept. 2019 to April 2020. Traffic declines in mid-Sept. 2019 and then again in March 2020.

When presented with these reports, many businesses would (understandably) look to cut back on their investment. However, because this client’s investment was rather small in comparison to our other clients, it was a simple decision to continue our efforts when the cost didn’t adversely impact the client’s bottom line.

The Strategy

At the same time that the pandemic shut down our client’s business, it also provided a silver lining. With more time available, the lawyer could focus more on growing and optimizing his website content. 

SEO strategies often take months to prove themselves. By putting in the effort to improve the firm’s site during this downturn, the business would be in a much better position when courts eventually opened back up a few months later.

The client came on board with little convincing, giving our team the freedom to move forward with our SEO approach as we best saw to.

1. Update Google My Business Listing

Even though our client’s business location was essentially shuttered during the early days of the pandemic, prospective customers would still be looking for lawyers — and our client’s Google My Business listing needed to be accurate for them.

After ensuring contact information was correct, we added new health and safety standards to reflect our client’s COVID-19 commitment, included new Q&As to address the most frequently asked questions, and posted new images to the account. We also used his GMB listing as another promotional avenue for re-optimized blog content (more on that below).

Google My Business Q&A: A family member was just charged with a *****, what should they do?

The goal: Show prospective clients that the firm was still active and paying attention to customers during this uncertain time.

2. Build Linking Authority

As part of the firm’s SEO content marketing strategy, we conducted trusty internet outreach. We added the client’s listing to authoritative lawyer directories, and we used HARO to build brand awareness through thought leadership.

3. Optimize Blog Content

The bulk of our SEO work during early COVID was to improve domain authority with optimized, trustworthy content (per E.A.T. best practices).

We began with the path of least resistance: blog content that was already performing well and creating inbound links. We updated meta data, images, header tags, and the on-page copy to be as thorough and informative as possible.

After updating one such blog (a piece on gun laws within the lawyer’s state), organic traffic grew by 143% year-over-year.

Google Analytics results, showing 143.3% increase in sessions, 2.35% increase in new sessions, and 149.01% increase in new users YOY.

For blogs that weren’t performing as well, we started with optimized meta data and H1 tags, to be revisited when the budget and capacity would be increased.

The Results

Even small SEO efforts can make a difference, especially when other marketing takes a backseat. Our client’s site hadn’t seen any significant SEO improvements since 2019, but our renewed focus on their blog finally brought the results the site had desperately needed.

Not only did informational calls to the firm increase, their site also saw:

Google Analytics traffic graph showing steady increase from August 2020 to August 2021. Red arrow indicates the upward growth of traffic numbers.
  • 147% increase in organic traffic YOY
  • 160% increase in clicks YOY
  • 282% increase in impressions YOY
  • 200+% increase to keyword footprint

Of course, it’s important to note relaxed COVID-19 restrictions in the legal industry starting in 2021 contributed to this growth. However, by optimizing website content during the pause in legal proceedings, we ensured the firm was in the position to capture that renewed interest.

Client #2: The Travel Tour Provider

Since partnering with Inflow in 2017, our cruise excursions client had seen significant organic traffic increases every year — that is, until 2020.

Google Analytics traffic map, showing steady traffic growth until 2020, where traffic drops off dramatically. Arrow pointing to drop-off point reads "COVID-19 Travel Lockdown."

With the abrupt shutdown of the travel industry at the onset of the pandemic, our client’s site performance dropped like a rock. But, while cruises wouldn’t sail for more than a year after the pandemic’s onset, that didn’t mean our client’s SEO efforts needed to take a pause.

On the contrary, like with our local lawyer, our strategists recommended using this downtime in business to undergo some serious SEO work (and prep for the eventual return of tourism).

The Strategy

Big site updates are understandably nerve-wracking for clients. The possibility of user experience and SEO performance tanking after changes like migrations is a legitimate concern.

But, with our client’s site traffic already in the hole (and unlikely to return for a few quarters), it was the optimal time to implement much-needed improvements.

1. Complete a Theme Migration

Our client had long wanted to redesign their site, especially as our SEO strategies brought in more and more traffic each year. The tourism slowdown gave them the perfect opportunity to get it done.

A site or theme migration could cause big drops in organic traffic and revenue as Google slowly recrawls every updated page. But, with those metrics already way below average for our client, any migration would have little to no impact at all. Because the cruising industry was still in strict lockdown at the end of 2020, Google had plenty of time to recrawl and reindex new pages without an impact. (Of course, we were there to monitor and resolve any tech SEO issues along the way.)

With their new 2020 design, our client’s site was officially ready to welcome back cruisers, whenever the travel industry came back to life.

2. Continue Optimizing Site Content

At the same time, our SEO team kept doing what they did best: optimizing existing content for future cruise vacationers.

Due to the historically low travel traffic volume, and a change in Google’s algorithm to show news-related “cruise” content in SERPs, our strategy had to change. Instead of using recent traffic trends, we looked back in order to move forward — that is, by using 2019 data. 

While older than the data from 2020, it was a far more accurate place to start when planning for the return of the cruising industry.

The Results

While the travel industry is still in a recovery phase and likely won’t be the same for years to come, there’s no denying that our client’s continued SEO investment paid off. While revenue is nowhere near pre-COVID levels, organic performance continues to make significant gains and set our client up for future success.

One of the best indicators is keyword rankings. Even with Google’s changing SERPs, our client maintained steady positioning in the #1–4 results (those most relevant to their products and most likely to convert) and continued to grow their keyword footprint in ranking positions of #1–10 (by 72% year-over-year).

With these keywords secured, their site is set to capture the most interested cruisers as they start coming back aboard.

Ahrefs organic keyword footprint map, showing steady growth in footprint for positions #1–3 and #4–10.

This focus on SEO content also has helped our client steadily regain whatever organic traffic there is currently available, even though the industry traffic remains historically low at this time. 

Google Analytics traffic map, showing steep drop before gradual traffic growth over 2021.

By maintaining ranking positions for high-value keywords and continuing to improve other keyword rankings, we’ve ensured our client will be in the best position possible when the industry does return to normal.

Bottom Line: Ongoing SEO Efforts Make a Difference

During downturns, it’s normal for businesses to jettison marketing efforts with less-than-immediate ROI. But, as these two examples show, SEO strategies should be the exception.

Search engine optimization and content marketing are long games, where ongoing efforts breed more substantial results than one-hit wonders. That’s why we recommend every business — even highly seasonal eCommerce brands — invest time and effort into long-lasting SEO strategies. With the right approach, you can set your business up for success and position yourself ahead of the competition for when it really matters.

Need help designing an SEO strategy for your business? Our team of strategists is always happy to help. Request a free, personalized proposal anytime to start boosting your organic performance today.



Source link

What does Yoast SEO do?

By | October 12, 2021


Yoast SEO is a WordPress plugin that helps your site perform better in search engines like Google. It also gives you the tools to bring your content to the highest standards of SEO and overall readability. Here, we’ll explain how our plugin helps you build the best website you possibly can!

What Yoast SEO does

Yoast SEO offers you loads of tools and features to boost your SEO. Some of these features influence the SEO of your whole site; other features help you to optimize individual posts and pages for search engines. At Yoast we really believe in our motto “SEO for everyone”, so you can access all the most essential SEO tools in our free Yoast SEO plugin. But if you really want to give your SEO a boost, upgrade to Yoast SEO Premium — it’s got even more amazing SEO features!

Keep reading to find out what Yoast SEO can do for your SEO!

SEO for your posts and pages

If you want your posts and pages to appear in the search results, you need to optimize them! So, when you use WordPress to create/edit posts, you’ll find a whole load of Yoast SEO tools to help you draft and optimize great content. And if you think SEO optimization is all about keywords, think again. The tools and tips in our Yoast SEO plugins put equal focus on quality content and user experience, too. Trust us — it will all help your rankings, whether directly or indirectly.

Here’s how the plugins will help you optimize your posts and pages:

Make sure you’re optimizing correctly (we’ll tell you if you aren’t)

After you’ve done your keyword research (you can use our integrated Semrush keyword data for that), you’ll have to start optimizing the pages and posts on your sites for the keywords and keyphrases you want to rank for. To do that, you can set a focus keyphrase for an article in Yoast SEO. Then, the plugin uses our content SEO analyses to determine how your content scores on different ranking factors. It checks things like how many times you use your keyphrase, the length of your text, or whether you used any internal links.

The results of these analyses guide you to optimize your post or page to rank with your chosen keyphrase. You’ll see red, orange and green bullets to indicate how every factor scores. This gives you an easy overview of the overall score and what you can still tackle to increase your rankings!

the SEO analysis in Yoast SEO
The content SEO analysis tells you how to optimize your text for a certain keyword with the use of red, orange and green bullets. This is a screenshot of our Premium plugin.

Get guidance for writing high quality, readable content — in 19 languages!

Optimizing your content to rank with the right keyphrase is important, but don’t forget your reader! Even if you write amazing content for search engines, your audience won’t benefit from it if they don’t understand it. When a person doesn’t understand your content, the chance of them buying something from you is close to zero. The same is true for the **** of them sharing one of your articles with their friends. So, you need to make sure your content is also easy to understand. And that’s where the readability features come in.

Our readability checks let you adopt the feedback in a way that suits you, without losing your personal touch. If you’re interested in all the factors that increase readability, you can read more about the Yoast SEO readability features.

The readability analysis tells you how to optimize your text to make it easy to read with the use of red, orange and green bullets. This is a screenshot of our Premium plugin.

The full set* of readability checks are available for English, German, Dutch, Farsi, French, Spanish, Italian, Portuguese, Czech, Russian, Polish, Swedish, Hungarian, Indonesian, Arabic, Hebrew, Turkish, Norwegian, and Slovak. Selected readability features are available for even more languages; check out the full overview of languages per feature if you want to know more!

* Unfortunately, it’s not possible to calculate the Flesch reading ease score for some of these languages. Check the overview below to see which languages.

Yoast SEO’s readability features are well-researched analyses that give you feedback on how to optimize your writing. Now, this may sound strange, because the way you write can be very personal. Let us explain how it works. The plugin uses an algorithm to check your content on different factors that are proven to increase readability. We look at the use of transition words, the use of passive voice, your sentence and paragraph lengths and more. But we carefully crafted this algorithm to make it as accurate as possible without being too strict.

Influence what Google shows in search results

Of course, you don’t just want your pages to show up in Google’s search results. You want your search results to look amazing, too! That’s why Yoast SEO comes with a set of tools to let you plan and preview how each page will (probably) look when it appears on Google. Probably is something we can’t really avoid here — Google will occasionally decide it knows better, and show something else instead. But by optimizing certain outputs on your page, you can indicate how Google should present your content to users. And that’s definitely still something worth doing.

With our plugin, you’ll be able to specify an SEO title (the ‘headline’ of your search result) and a meta description (a short piece of text underneath your search headline, describing what users can find on your page) for each new page you publish. We’ll let you know if these are too long, or if your keyword is missing. If you want to, you can also set defaults for all of your pages.

Google preview in Yoast SEO

You might have seen search results that contain extra parts before, beyond the usual headline-and-description format. The example below contains recipes with extra information like reviews, cooking time, ingredients and images, for instance. And that’s just one example. There are possibilities to add extra information for all kinds of results, including products!

The way to get results like this is by using Schema structured data. We won’t lie: it’s complex, technical stuff. Luckily for you, you won’t need to know a thing about the tech wizardry behind it. Just having Yoast SEO installed means you’ll automatically have structured data output for your pages. All you need to do is select a few options to make sure it suits your needs.

Manage social outputs

Now, social media isn’t strictly a part of SEO. But when you make great content, you often want to share that content on your social feeds, too. That’s why Yoast SEO also comes with Facebook and Twitter previews that you can adjust to make sure your content is always looking great, whoever is sharing it. You can set a specific title, description and OpenGraph image for each post. Again, if you prefer to set one standard structure for all posts, there’s an option to do that.

Technical SEO for your website

We’ve taken a look at what Yoast SEO can do for your posts and pages. But what can it do for your site overall? If technical SEO isn’t your strong suit, much of the following may not make sense to you. But, don’t worry! Yoast SEO exists to make sure you don’t have to know all of these things.

Set up your site for SEO

The plugin settings are very sensible by default, and our configuration wizard also guides you through the steps to get your technical SEO settings right. Once you’re done, you’ll have an XML sitemap, a robots.txt file, site-level Schema structured data, and more. That all sounds complicated, but the important part is this: the plugin will automatically ensure Google can find, read and index your content. If you don’t want some of your content to be found, you can prevent it from being indexed with just a couple of clicks, too.

Manage your content

As you write more and more content for your site, you’ll be looking for easy ways to manage it! The Yoast plugin comes with a few features to help you manage your content well — and avoid common SEO issues. For instance, when you make changes like deleting a page or changing a URL, if you don’t know what you’re doing then things can get messy. And if you make a lot of similar pages that can be a problem too, as Google doesn’t know which one it should direct users towards. To help you deal with SEO issues like these, Yoast SEO comes with two unmissable tools: canonical URL tags, and the Redirects tool.

Canonical URLs are really helpful if you have a lot of similar content, such as a webshop with multiple variants of the same product, each having its own page. To make life easy for you, Yoast SEO automatically adds canonical tags to all content marked for indexing. All of the canonical tags will be taken care of in the background; in most cases you won’t need to change a thing. If you do need to adjust your canonical URL tags, it’s easy to do so.

Redirects are essential if you’re moving, or removing, content. The fact is, users will probably still find their way to the old URL — but the content they’re expecting won’t be there. That’s not only disappointing and frustrating for users, but it can also make it harder for Google to find and index your content, too. Redirects can be fiddly to set up on your own. But when you’re using Yoast SEO, it’s actually hard for you to get your redirects wrong! Any time you change an existing URL or delete a page from your site, you’ll get a notification prompting you to set up a redirect. It only takes a few clicks to make sure any visitors arriving at the old URL will be redirected to the right place. Need to check or adjust a redirect you’ve set up? You can do that easily from the Redirect manager overview.

Build your site structure and internal links

If you want findable content that really ranks, you need to take care of your site structure and internal linking. The Yoast SEO plugin comes with a few tools to help you manage how your content links together: there’s a text link counter which will tell you how many incoming and outgoing internal links there are on a page, as well an internal linking suggestions tool (in the editor view) which can help you add more if necessary. Those features are available in both Yoast SEO free and Yoast SEO Premium.

If you’re using Yoast SEO Premium, you can use our SEO workouts to make internal linking easier than ever:

By doing the workouts, you can easily and confidently set up a solid internal linking strategy, and it will only take a few clicks! When you’re done, you won’t have any unfindable content on your site (unless you don’t want it to be found) and your most important pages will get all the links they deserve.

Even more technical features of Yoast SEO

By simply installing the plugin and following the steps in our wizard, you’re already fixing a lot of important technical SEO things for your site! We do these steps for you, so you don’t have to know about every little technical detail.

If you really want to know everything Yoast SEO can do for you, then take a look at the complete list of features. Additionally, if you are (a bit more) familiar with technical SEO, you might enjoy reading more about Yoast SEO’s hidden features that secretly level up your SEO!

Learn SEO by doing SEO with Yoast

Still need to learn about SEO? One of the biggest benefits of using the Yoast plugins is that they make it really easy to get started, and learn as you go along! We’ll give you pointers to help you get everything right, as well as links to read more about how SEO works and how to do it.

If you want to become an SEO expert (or just improve your knowledge) you can learn even more by following our Yoast SEO academy training courses. Access to our academy is free for all Yoast SEO Premium users. Not sure if it’s right for you, yet? Have a go with one of our free courses first and get a taste of what’s on offer!

A quick recap

In this article, I’ve shown you what Yoast SEO can do for your site. Our plugin helps you improve your content SEO by helping you set a keyphrase and telling you exactly how you can optimize your content to rank with this keyphrase. The plugin also helps you improve the readability of your content by providing feedback that you can easily incorporate into your own writing style. And last but not least, the Yoast plugin improves your technical SEO by taking care of a lot of technical things in the background.

I hope this article gives you some insight into our plugin and what it can do for you! It’s also worth noting that we offer two versions of our plugin, so have a look to decide which one suits you best. We offer a free version of the plugin, that will definitely get you started with your SEO. But I advise you to also take a glance at our Premium plugin to make sure you’re not missing out on the features that will get you that top position in the rankings.



Source link : Yoast.com

How to Optimize the Website for Google Featured Snippets (“Zero Position”)

By | October 12, 2021


Read 5 minutes

Who doesn’t wish to jump to the first Google ranking without a ton of backlinks and significant content updates? Featured snippets are a miraculous option to achieve this! Use the information below to get up there and wave to the world from the top.

You might have come across different formats for display of information in Google SERP. The elaborated display of content that gives you a quick view of information is called “Zero Position”.

What is a Featured Snippet anyway?

A featured snippet is a column that appears at the top of the search result page. It provides you with the exact information to concisely and accurately answer your question, with a citation to the website. Refer to the example below.

There are various types of Google featured snippets (or ‘Answer Boxes’) that pop-up for your search queries:

  • Paragraph
  • Numbered list
  • Bullet list
  • Table
  • Video

Featured snippets are crucial to trickle in high-intent clicks, and you must leverage their power to your benefit. Featured snippets offer greater visibility to searchers and help boost brand recognition. The next important thing to learn is how to rank your featured snippet on the top to increase traffic.

How to rank in Featured Snippets? Do I stand a chance?

Research shows that 99% of the featured pages are already ranking in the top 10 of Google. So, if your website ranks high, your snippet stands a greater chance of getting featured on the top. Do not be surprised to know that pages like Wikihow and Wikipedia are the most featured sites. The competition is, however, fierce, but not tricky enough to give-up without trying!

Some search queries related to Images, Videos, Local information, and shopping don’t show Featured snippets. With already high rankings, you must focus on long-tail informational queries to improve overall rankings. If your website’s niche is health, DIY, or finance, your snippet certainly stands a good chance to get featured on the top.

What’s to do to get included in Featured Snippets?

If you are familiar with basic SEO tools, you can easily get included in Featured snippets for relevant queries. Research, identifying the target keywords, changing structuring of on-page content, and a few other tweaks go into the process. Let’s take a look into it: 

Research, research, and more research

Featured Snippets demand a colossal research input. A lot needs to be studied about the types of answers people look for, the type of questions you need to answer and make a query list that your content must contain to optimize the chances of getting featured. You can browse and learn what people look for;  take the help of Google’s own “People also ask” sections to check for new search results. It provides valuable insights into which questions Google deems related to each topic.

Identify keywords you want to target

You can start with big “head” keywords. The next step should include focusing on long-tail keywords when testing out a strategy. Take the help of keyword research tools like Google’s Keyword PlannerSEMrush, and WordStream. Question research is keyword research with other terms added in the beginning. For example, instead of researching “blog revenue,” you can try researching “how to increase blog revenue.” SEMrush gives you an Organic Search Results section where the results vary from that on Google. This is because they eliminate advertised links and give you a more accurate idea of your competition for a given organic keyword. Check websites and break down keyword performance in-depth. You can also look for related keywords. But, remember to keep the search on question-based keywords.

Change the structuring of on-page content

Closely work with the writing team to update the content layout of your blog articles. Include keywords, frame and structure the blogs in a way that they are relevant to the user’s questions and look closely into other related things.

Measure, test, and repeat

Once you have a plan, you can alter it accordingly, test it live, and make repetitive changes until the desired results are achieved. Remember, like SEO, it takes time and patience. Try and measure performance of your changes to have the best output. Keep a check on the keywords your competitors are targeting. Check for improvements in the design, details, and other nuances of the content. If you manage to create a new answer and drive traffic to it, you’ll replace your competition in the featured snippet.

How to optimize your website for featured snippets?

Start with on-page SEO! For Featured snippets SEO, there is no quick-fix trick to get featured. Remember, Google favors the user, and you have to match-up to their standard. Start with non-specific SEO best practices. Aim for structured markups. Sometimes, using Schema.org is suggested.

What is schema mark up? 

Schema markup is semantic vocabulary or code that you tag to your website to assist the search engines to give more informative results for users. Schema explains to the search engines what your data means, rather than what it just says. It deploys a unique semantic vocabulary in a microdata formatSchema featured snippets help provide the most precise results of the user queries. 

By now, we have learned that the best way to get featured in snippets is to provide the best possible answers.

Below are some more tips to help you rank better in featured snippets;

  • Answer each question concisely: Google prefers to feature an answer which is given within one paragraph. Let this be a guideline. Ascertain how long each answer should be to get featured. Remember, Google does give preferences to long-form content divided into logical sections and structures of relevant attractive images. So, adjust your blogging style, ask questions in your article, follow up with a short-paragraph answer, and elaborate the details further. This trick will also gain higher user-retention as better structured articles enhance readability.
  • Be factual and organize well: Google appreciates steps, numbers, and lists. Featured snippets, for e.g., list the actual ingredients, number of steps, time to cook, flavor and nutrition for a recipe. Stay factual, although Google will **** to feature your well-structured and number-driven content.
  • Ensure one article contains answers to various related questions: Once a page gets featured, it is likely to get featured for a lot of similar queries. The requirement is being structured and worded in a way to address multiple-related questions. Google very well determines synonymic and closely related questions. So, mould your content this way.
  • Organize your questions accurately: To associate closely related questions in one article, organize the queries properly to craft a well-shaped, well-structured content. Use this keyword organization strategy for the same:

1. The generic keyword makes one section/category of blog

2. More specific search query forms a title

3. Other specific queries go into the subheadings to define its structure

  • Update the images: This is a trick as old as the internet. Update your content images with relevant alt-text and let Google pick your page for featuring the snippet. 

*Note

Google recently announced an update to the Google featured snippets. For some featured snippets, Google will now take the users to the relevant text on the web page and highlight the text. Clicking a Google featured snippet in this new feature leads to a page with the relevant text highlighted. 

In closing

Landing on a featured snippet position is not just possible, but also probable. Remember, as it happens in SEO and marketing, there isn’t a given method to achieve this. You are free to experiment and optimize content, but nothing comes guaranteed! Make sure you include content optimization as a regular process of your SEO strategy and you may expect to see one of your pages being featured in Google Featured Snippets.

Comment and share your experiences of Featured Snippets!

Get in touch with us to talk about SEO!

Get Started Now





Source link

What is Search Engine Optimization (SEO)? Executive’s Guide

By | October 12, 2021


Google owns 87.8 percent of the search engine market share. And, in 2021, an estimated 90.8 percent of the total U.S. population accessed the internet. In other words, the smart marketing money invests in Google search.

However, unlike traditional advertising and PPC, you can’t win just by throwing the most money at it. In order to beat the competition at search engine optimization (SEO), you need to know what SEO is, and how to strategically invest in it.

There are still too many CMOs who don’t care how the soup is made as long as it tastes good.

But, you’re smarter than that. You recognize that consumer behavior has permanently changed, and marketing must evolve in order to remain profitable. You also know that it’s impossible to make the right investments in SEO if you don’t know what it is.

I wrote this overview to introduce marketing executives like you to the basic concepts of SEO.

I won’t discuss the detailed nuts and bolts of how to do SEO. Instead, I’ll explain the fundamental elements of SEO that will help you make smarter investments in the organic search channel for greater marketing and business results. 

I’ll briefly summarize the following topics:

  • What is SEO?
  • Business benefits of organic search
  • Key elements of search engine optimization
  • Considerations before you implement an SEO strategy
  • Should you outsource SEO?

What is SEO in marketing?

Search Engine Optimization (SEO)

Definition

SEO is an acronym that stands for search engine optimization. It refers to the process of increasing organic (non-paid) search engine visibility in order to grow brand awareness, drive qualified website traffic, and protect brand reputation.

SEO is a multifaceted process that involves optimizing both your website and off-site elements to improve SEO rankings and generate high-quality organic traffic.

When marketers discuss SEO, the conversation is often focused on things like keywords, Google’s algorithms, and backlinks. However, at its heart SEO is about aligning with your audience, the way they think, and the intent underlying their Google searches. It’s about knowing what your audience is searching for, why they’re searching for it, and then providing them with effective solutions.

Ultimately, SEO marketing helps you attract your target audience at each stage of the customer journey, which results in more brand touch points, leads, conversions, and revenue.

What’s the difference between SEO and PPC?

There are a number of key differences between SEO and PPC. The more you understand those differences, the more you can strategize how it fits into your overall marketing strategy and the more you can benefit from SEO.

Both search engine optimization and PPC are strategies for driving traffic to your website. The difference is in how that traffic is acquired.

How SEO works

The internet gives humanity access to nearly unlimited information. Yet, few people actually know how search engines work in order to deliver just the right answer in the blink of an eye. If you haven’t seen it yet, I recommend watching Google’s video about how search works for beginners.

But, if you don’t have time, I’ll summarize it as it pertains to search engine marketing below.

Crawlers visit your website in order to understand what it’s about. These bots look at the types of content that users see, such as text and pictures, as well as things they can’t, like structured data, and image alt text.

They do this across billions of pages on the internet to understand every page on each website, and how they all relate to one another.

Then, search engine algorithms rank your content based on a specific set of criteria. For instance, do you have relevant content to satisfy a user’s search terms? Does your website perform well in mobile search? Are you an expert on the topic?

How well your site ranks in the search engine results pages (SERPs) depends on several factors, including how well you optimize your content to rank for relevant keywords and the overall quality of the user experience you provide.

Google ranking factors

How does Google decide what to include at the top of its search results? 

It relies on approximately 200 ranking factors, with some factors being more important than others. For example, it matters how quickly a page loads, but content quality matters more. Furthermore, some meta tags are critical to how well your site performs in search, while others, such as meta descriptions, are not factors at all.

One thing that’s consistent across the board for Google is that it’s always looking for high quality. To that end, you can find Google’s Quality Guidelines here.

SEO ranking factors related to the entire website include:

  • Website traffic volume
  • Domain trust
  • XML sitemap
  • SSL certificate
  • Streamlined HTML
  • Site architecture
  • The quality of content in the website
  • User experience
  • Internal linking
  • Backlinks from authoritative domains

SEO ranking factors related to the individual web page (on-page SEO) include:

  • Keywords
  • URL
  • Title tag
  • Headings on the page
  • Content quality
  • Schema markup
  • Internal links pointing at the page
  • Backlinks from authoritative pages
  • Anchor text of inbound links

Ultimately, the goal of SEO is to optimize pages of your website so that they each align with what users want when they search for something. Google encourages designing pages primarily for users, not search engines, giving users the best content relative to their search query.

Enterprise SEO PlaybookBuild a more efficient SEO strategy and leave your competitors in the dust.Download the Study

How PPC works

With PPC (Pay-Per-Click) advertising, you pay to show up in the advertising areas of the search engine results pages for particular keywords. When someone clicks on your ad, you pay a fee, whether or not the user intended to click.

The amount you’re charged per click depends upon a number of factors, including:

  • How many organizations are bidding for the same keyword
  • The amount you’re willing to pay versus your competitors
  • The quality of your ad
  • Day of the week and time of day relative to demand

The advantage of PPC is that it allows you to quickly drive targeted traffic to specific landing pages of your website. This can be particularly useful for time sensitive things like testing, promotions, sales, events, etc.

One big downside of PPC is that it can be an expensive acquisition channel. In addition, the moment you stop investing in PPC ads, the traffic literally stops. There’s no residual effect. The only way to sustain PPC traffic is by pouring a steady stream of cash into it.

This is one reason why SEO is a better investment over the long run. While you don’t get the immediate results of PPC, you reap the benefits of SEO for a much longer time with the added benefit of compound results. A page that ranks on the first page of organic search results can generate traffic for years after you initially create it.

Business benefits of SEO

5.3x

SEO delivers 5.3x ROI compared to 2x from paid advertising

10x

SEO drives 10x more traffic than social media on average

70%

Consumers do 70% of their research online before buying

There are significant business benefits to SEO, which underscore why it should be an integral part of your digital marketing efforts.

At a high level, search engine optimization amplifies growth and reduces risk, driving more revenue and increasing shareholder value.

More specifically, investments in SEO return numerous business benefits, including increased brand awareness, more website traffic, stronger brand loyalty, lower acquisition costs, and improved brand reputation.

Increase sales revenue

When done effectively, SEO drives tremendous sales revenue.

Google is an integral part of our personal and professional lives. We use it to guide buying decisions, research products, and find answers to our problems. We use it to find local businesses and decide which pair of running shoes we should buy.

Search engine optimization allows you to be present at every step of the customer journey, from the moment they first become aware of a problem, to when they make a purchase, and beyond. By creating high value content that aligns with what users are searching for, you can engage prospects at each stage of your marketing funnel.

Businesses that don’t leverage SEO as part of their digital marketing toolkit miss out on a significant amount of potential revenue. Whether your goal is to sell more ecommerce products, generate leads, increase brand visibility, or even capture traffic from TV ads, SEO can help you achieve your goal.

Increased Google Market Share by 265%Terakeet Increased a Financial Services Company’s Google Organic Market Share by 265% within 22 Months.See How

Improve margins

Search engine optimization reduces customer acquisition costs and increases the lifetime value of those customers. In fact, Terakeet delivers customers at 25 cents on the dollar compared to paid channels.

Consider the difference between SEO and other forms of digital marketing, such as PPC, social advertising, or display advertising. Most digital marketing interrupts a person’s online experience. In other words, it creates friction. They’re scrolling through Facebook and an ad pops up in their feed. Or they’re reading an article and the sidebar starts flashing ads for products.

With SEO, there’s no interruption to the customer experience. When a person searches for something, they’re actively seeking answers. By providing valuable content that matches what they’re looking for, you facilitate the path to their objective.

Not only do you avoid potential customer interruptions, SEO has a higher ROI than many other forms of digital marketing. For example, with PPC ads, you have to pay for every visitor to your website, which significantly drives up customer acquisition costs.

Although SEO requires an initial upfront investment to create and optimize pages, a single piece of content can drive organic search traffic for months, or even years after its creation. As you build an ecosystem of content and your website builds authoritativeness, it becomes easier and easier for your site to gain additional related Google rankings. It’s a case of a rising tide lifting all boats. The end result is a lower acquisition cost over time.

Reduce brand risk

SEO also allows you to protect your brand’s reputation. When someone does a Google search for your company’s name or product, you want to own the first page of the search results. You don’t want searchers to see a negative review from a random disgruntled customer or a scathing news article because your stock has fallen. SEO allows you to take control of the search results and ensure that only the content that reflects best on your brand shows up on the first page. This helps you balance the coverage of your brand across the major search engines.

Beyond protecting your brand identity, SEO plays an important role in increasing customer trust. When you provide the type of information customers are looking for (in their words), they begin to associate you with helpful insights and knowledge.

By spending the time and energy necessary to create great content that answers user questions, you can build customer trust with each piece of new content published. As a result, visitors are more likely to buy from you since they value your brand as an authority in the industry. They’re also more likely to dismiss the occasional negative review they come across since the overwhelming majority of what they see is positive.

Forrester Study – SEO Paradigm ShiftAccelerate growth and protect your brand with fully integrated SEO.Download the Study

Key elements of an SEO strategy

It’s one thing to know that search optimization is important and has tremendous benefits for your business. Actually executing a strategy based on SEO best practices that gets results is something else altogether. Let’s talk about the key elements involved in an effective SEO strategy.

Technical SEO

Technical SEO is foundational to optimization of your website. If you focus on optimizing your content yet your technical infrastructure is weak, you’ll hurt your ability to rank in Google.

Technical SEO is the process of optimizing a website to meet the technical requirements of Google. The main categories of technical SEO include:

  • Crawling
  • Indexing of pages
  • Site architecture
  • URLs
  • Page rendering

Crawling

First, make sure that Google can crawl the pages of your website. You’d be surprised how many webmasters accidentally block Google from accessing their website by adding a few simple words to the header of a webpage. These problems aren’t limited to small businesses either. Even Fortune 500 brands make mistakes.

Indexing

To ensure indexing of your pages, make sure that you’re using an XML Sitemap correctly. Make sure that the pages you want indexed by Google are included in their proper child sitemaps, and that all extraneous files are removed.

Site architecture

Next, focus on your website architecture. The structure of your site plays a key role in both helping Google understand how all the pages relate to each other and also makes it easy for users to find exactly what they’re looking for on your site. Make your architecture simple and intuitive, and highly organized.

URLs

The same can be said for the URL format you use. Short, descriptive, easy-to-read URLs that contain the primary keyword for a page help Google and users better understand the content of the page. For example, if you sell coffee grinders on your website, the URL www.yoursite.com/coffee-grinders is much better than something like www.yoursite.com/19j5xr39ab56.

Page Experience

The technical performance of your website is a critical part of providing site visitors a good experience. Google’s Page Experience Signals are a collection of metrics used to measure website page performance, and in turn, user experience.

We won’t get into the technical details of each one, but there are six page experience signals that Google takes into consideration. The first three are called “Core Web Vitals“, and include:

  • Largest Contentful Paint (LCP) – Measures how fast a page loads in the browser
  • First Input Delay (FID) – Measures how quickly a user can interact with a page
  • Cumulative Layout Shift (CLS) – Measures visual stability of a web page (how much elements of a page move around when the page is loading)

The other three page experience signals Google prioritizes are:

  • Mobile-friendly – Whether a site can be easily used on mobile devices.
  • HTTPS – Whether the site connection is secure via SSL.
  • No Intrusive Interstitials – Ensuring that excessive ads or pop-ups don’t disrupt the browsing experience.

Content quality and alignment

person's hands writing SEO content on a laptop

While the technical performance of your website is very important, the content of your web pages matters even more. As Google has stated:

Great page experience doesn’t override having great page content. However, in cases where there are many pages that may be similar in relevance, page experience can be much more important for visibility in Search.

All things equal, Google favors pages with the best content. What constitutes the “best” content? The following are two important elements:

  • Quality
  • Alignment with search intent

These are two sides of the same coin. Quality content marketing fully addresses a topic so that a person can find the information they need without going back to the search results and digging through additional websites. (Blogging is often an effective way to consistently generate high-quality content.)

This doesn’t necessarily mean that the content needs to be lengthy, although it often requires significant length in order to sufficiently cover a topic. For example, if you browse Google page one results, some are quite short because that’s all that’s necessary for the topic, while others go on for thousands of words. 

Although this may sound simplistic, a good rule of thumb is that quality content says all that needs to be said, taking into account the topic, audience, purpose of the page, etc.

With this in mind, it’s important to conduct a content audit to prune non-performing content, eliminate any duplicate content, and ensure that your content library is strong.

Search intent alignment

google search on a mobile device

Search intent is the goal that a user has in mind when they type a query into Google. For example, when people search for “men’s soccer cleats” they’re probably wanting to look at pages where they can buy cleats. However, if they search for “best men’s soccer cleats”, they’re most likely wanting to read reviews and comparisons of various brands of cleats.

To rank in Google search results, you not only need to know what your audience is searching for, but also why they’re searching for it. Keywords are important, but they don’t tell the full story. Apply a bit of user psychology and voice of the customer (VoC) to to better understand the intent underlying the different Google queries. Another factor is the point in the customer journey at which they are most likely to make the query.

An effective SEO strategy involves creating high quality content that is closely aligned with search intent through the customer journey. When a person arrives at a page on your website via search results, it should be immediately clear to them that the information on the page is exactly what they had in mind when they entered a particular search query. Do this, and watch your conversion rates increase.

Popularity and trust (E-A-T)

An additional element of an effective SEO strategy is building the popularity and trustworthiness of your website. As a general rule, Google prefers to rank well-known, reputable, authoritative websites that have higher PageRank.

How does Google determine whether a website is reputable? A few important factors include:

  • Links and mentions from high-quality websites
  • Quality of the content on the site

In addition, Google is increasingly measuring the E-A-T of a website in its ranking determinations. E-A-T stands for:

  • Expertise
  • Authority
  • Trustworthiness

In Google’s quality guidelines, Google includes Expertise, Authoritativeness, and Trustworthiness in its stated most important factors used to determine a web page’s overall quality.

Google further clarifies that they determine the E-A-T of a page through the following:

  • The expertise of the creator of the Main Content.
  • The authoritativeness of the creator of the Main Content, the Main Content itself, and the website.
  • The trustworthiness of the creator of the Main Content, the Main Content itself, and the website.

Google E-A-T is especially important when it comes to websites impacting your finances, health, safety, and/or happiness. “Your Money or Your Life” (YMYL) sites are held to the highest possible E-A-T standards due to the subject matter and what it means for a user if that information is misrepresented.

So, if your website deals with financial services or medical advice, for example, be sure to pay extra special attention that your website is conveying substantial E-A-T signals to Google.

Who is responsible for SEO?

marketing team discussing who is responsible for seo

At a high level, search engine optimization falls under the responsibility of the head of marketing since it is a marketing channel. However, depending on the size of the organization and the skills needed to execute an effective SEO strategy, there’s a good chance that the head of marketing isn’t directly involved in the day-to-day SEO activities.

That responsibility typically falls to the head of digital marketing or the SEO director. Additional team members would be responsible for various aspects of the SEO program, such as project management, topic and keyword research, content creation, page analysis, technical optimization, link building, etc.

At the executive level, SEO should be the center of audience insight helping to drive all marketing efforts. However, many executives fail to capitalize on the full potential of organic search to create value for the enterprise beyond traffic.

According to Forrester Consulting’s thought leadership paper “The New SEO Paradigm Shift” commissioned by Terakeet:

Business leaders are thinking too small about SEO. Companies need leadership that embraces a new way of thinking about SEO to overcome top barriers. Executives who want to truly dominate the landscape need to directly link new SEO investments to clear and measurable brand improvements that drive business growth and improve their reputation… This will require changing the culture around SEO and aligning others to think of SEO as a strategy to meet long-term goals, not just drive site traffic.

Should you outsource SEO, or bring it in-house?

The question of whether to outsource SEO or bring it in-house comes down to three primary factors:

  • Budget
  • Expertise
  • Speed of results

To do SEO in-house, you need the budget to hire, train, and maintain a department. Search engine optimization is a labor intensive process that requires a dedicated team to drive meaningful results.

Teams need a thorough understanding of technical optimization, SEO tools like Google Analytics, Moz, and Ahrefs, as well as keyword research, on-page optimization, content strategy, link building, and more.

If time to results is important, you should look to outsource SEO services.

Outsourcing SEO allows your brand to focus on what you do best while gaining the benefit of SEO experts. As a result, you’ll drive your organic results further than you could on your own.

When outsourcing SEO services, you have three options:

  • Hire a search engine optimization company
  • Hire a digital marketing agency (with SEO as one of their services)
  • Work with a freelancer or consultant

If you’re an enterprise business, outsourcing to an SEO company is the most reliable way to scale your efforts. You’ll be able to hire a smaller, more nimble in-house team focused on implementation while your SEO partner does all the heavy lifting.

However, to be successful and safeguard against risk, it’s crucial to ask the right questions before signing any contracts. Blackhat SEO agencies may seem like a bargain until they land you in the Google penalty box for tactics like keyword stuffing or buying backlinks.

Limitless SEO ResourcesWhat would you do with unlimited SEO resources? Learn about outsourcing SEO.Read the Post

If one thing is clear, it’s that SEO is only going to increase in importance. As the amount of information available to consumers continues to grow at an explosive pace, their reliance on search will continue to grow.

Brands that double down on SEO now will reap the rewards long into the future. They’ll be able to connect with a much wider audience, generate more leads and revenue, and create deeper customer loyalty than brands that neglect SEO.

NOTE: Although Bing and Yahoo are part of the search engine landscape, this overview focuses on Google given Google’s overwhelming market share.

SEO FAQs

What is SEO?

SEO is an acronym that stands for search engine optimization. It refers to the process of increasing organic (non-paid) search engine visibility in order to grow brand awareness, drive qualified website traffic, and protect brand reputation.

What are the benefits of SEO?

At a high level, search engine optimization amplifies growth and reduces risk, driving more revenue and increasing shareholder value.

More specifically, investments in SEO return numerous business benefits, including increased brand awareness, more website traffic, stronger brand loyalty, lower acquisition costs, and improved brand reputation.



Source link