How I used Fable 5.1 to spot likely automated queries in Google Search Console

Conceptual illustration of an AI agent using a magnifying glass to analyze a floating table of Search Console CSV data, categorizing search queries into human and automated patterns with glowing cyan and purple metrics

Recently Google Search Console is showing some weird queries, which are highly unlikely to be something which a real human user is going to do like a Google Search. And it might be possible that those queries are from some AI Search systems (like ChatGPT, Perplexity, Claude) and while doing SEO Analysis it’s important to have some clarity around if a query is being done by actual human user or it’s something which an AI Search system is looking for.

First export data from search console (ideally 6 to 12 months)

Google Search Console Performance on Search results report for gaganghotra.com showing 1.27k total clicks and 154k total impressions over 12 months, with the Export menu open and an arrow pointing to the Download CSV option.

Then unzip the file which search console provided you, after that upload all the CSV files to Claude while Fable 5.1 Extra is selected.
And use this simple prompt – “need to do analysis of all the files in this folder and figure out which queries are likely done by some real human user and which ones are being done by some AI Search system, it’s like a fan out query or some other automated query”

Screenshot of an AI chat interface with four uploaded CSV files (Chart.csv, Countries.csv, Devices.csv, Filters.csv) and a typed prompt asking it to analyze the files to determine which search queries were made by real human users versus AI search systems, fan-out queries, or other automated queries.

And press Enter – then it’s going to take sometime (probably 5 or more minutes).

Then it’s going to show you a table view showing classification of queries like following.

Screenshot of an AI chat report titled 'Distinguishing human queries from automated search...' next to a linked GSC Query Classification spreadsheet, with the report describing LLM-generated retrieval queries, fixed prompts, SERP scrapers, and monitoring tools, and the spreadsheet's Queries tab showing rows labeled Automated or Likely automated with columns for impressions, CTR, position, and query source classification

Screenshot of an AI chat report titled 'Distinguishing human queries from automated search...' next to the GSC Query Classification spreadsheet, showing the Queries tab with rows highlighted green and labeled 'Likely human,' including searches like 'claude fable 5 system prompt leak,' 'google gemini cancel subscription,' and 'advance gsc visualizer,' classified as classic human web searches.

This classification is mostly directionally correct and isn’t alway absolute, but definitely is really useful to understand and get some ideas around if some queries in search console are actual searches being done by real human user or not.

More in SEO

View more

Do you agree with the comment below?

Join the conversation

Posting publicly · your email is never shown


  1. Gideon

    Rather than using GSC directly which limits the data to 1,000 lines I’d recommend using Search Analytics for Sheets which allows up to 20,000 lines.
    With many sites you’ll want to split the data into two groups:
    · branded inc phone number etc
    · non-branded
    That way you get a full 20,000 view of the competitive landscape.
    If you use the full 16 months view you’ll be able to use your favourite LLM to answer whether the brand has strengthened or weakened over time.
    And if you ask AI to tell you which pages are between 10 and 15 in rankings for useful terms you’ve got an effective hit list of pages to improve.

    Reply