Code Interpreter in ChatGPT has been a game-changer for many search marketers since its launch in March.

John Mcalpin has recently covered how the native OpenAI plugin can be used for several SEO use cases.

Code Interpreter also holds significant potential for aiding PPC marketers in swiftly revealing actionable insights. Let’s explore how.

Getting started with Code Interpreter

Code Interpreter is “an experimental ChatGPT model that can use Python, handle uploads and downloads,” according to OpenAI. 

Essentially, you can upload a file of various formats:

• Text (.txt, .csv, .json, .xml, etc.) 
• Image (.jpg, .png, .gif, etc.)
• Document (.pdf, .docx, .xlsx, .pptx, etc.)
• Code (.py, .js, .html, .css, etc.)
• Data (.csv, .xlsx, .tsv, .json, etc.)
• Audio (.mp3, .wav, etc.) 
• Video (.mp4, .avi, .mov, etc.)

Once you’ve uploaded one of these file types, you can perform various tasks. 

You can merge and change file types, ask questions about the data and, most importantly, perform analytics and visualization tasks. 

That’s where the magic happens for paid search marketers.

Too often, we spend our time combing through mounds of data, ad copy, landing pages, and search query reports. 

We are looking for insights and opportunities to improve our performance. 

We spend time trying to put these into graph form to communicate what we’ve found to a client or our leadership teams.

All these can be done very quickly and efficiently without much knowledge of math or Excel. 

Note: As of writing, it’s available to ChatGPT Plus users as a beta feature.

Code Interpreter for PPC

You can use Code Interpreter to analyze and visualize your campaign datasets.

Run any report from any dataset you like (Google Ads, Google Analytics, etc.). Upload that file in the message box and request the action you would like ChatGPT to take. 

If you upload a Google Ads report, the data doesn’t start until line 3. 

I found that ChatGPT sometimes could recognize this, and sometimes it did not. 

If it does not recognize the data starting on row 3, you can include this in the prompt, “Start the dataset on row 3.” (This worked every time!) 

Here are some basic examples from keyword reports. 

Prompt: “Can you graph the spend by day for the top 5 keywords?”

You can also be more specific in the type of graph. Below are examples of requests for different data views but with different graph types.

This example requests a bar chart for conversion rates by day of the week. 

You can have it organize the data from highest to lowest or request that the data stay in order of the week. 

In this case, I can see that our weekend conversion rates are ~50% lower than Thursday and Wednesday. I can use this data to help me determine my budget allocation and inventory needs.

I also pulled a conversion-by-hour report and asked Code Interpreter to create a heatmap of the data. This allows me to quickly understand the best-performing days of the week. 

Previously, I would have to pull the data down and know how to conditionally format the data. 

Let’s do a bit more analysis along with data visualization. 

Below, I uploaded a file with over 1 million rows and asked Code Interpreter to create a scatter plot with a trendline showing the relationship between the two metrics. 

It noted the dataset’s large size and created a way to simplify the results through sampling. 

This would have taken a long time with this much data, and I’m unsure how I would have created a “random sample” of 5% of the data. 

With Code Interpreter, the output was all done in less than a minute.

Another example is correlation analysis. I wanted to understand if there was a correlation between conversion and cost or conversions and CTR by day. You can see the output of this request below. 

For this data set, there wasn’t much of a correlation (1 = perfectly correlated). However, it shows you the data quickly and helps you interpret it and understand that this correlation isn’t very high.  

The final example I’ll share is a request from a search terms report with over a million rows about what keywords I should add to the negative list because they had high CPA and low CTR. 

One of this plugin’s most impressive features is how quickly and well it understands what’s in the dataset. 

For example, in this dataset, we have two custom columns for different conversion events.

I asked ChatGPT to show me the highest CPA. It understood and added the two conversion columns together to give total conversions.

This is exactly what I wanted it to do, but it did it without me having to prompt it or help explain the data. 

I then asked it to show me the bottom 10% based on that criteria. It provided a bulleted list to extract and upload into Google Ads. 

Finding ways to analyze your PPC data quickly and efficiently allows you to get insights and optimization faster. 

These insights drive the optimizations that differentiate the winners and losers in the auctions. 

As this tool evolves and more data gets migrated in and connected, you can see the overall impact for search marketers and businesses.

I’m excited about the opportunities this provides and look forward to how this evolves over the next few months.