Search Console's ordinary report is quietly logging your AI Mode conversations
Google confirmed that follow-up turns inside AI Mode get logged as brand new queries in the classic performance report. The dedicated AI report that went live for everyone on August 11 still won't show you the queries, the clicks, or give you API access.
Guilherme Hortinha · 18 August 2026
Can your business run ChatGPT Ads? Check in 10 seconds.
Google's Search Console has been logging queries that were never typed anywhere. John Mueller confirmed it in early August after SEO Anastasia Kourou spotted rows in her performance report that read like replies: "Yes," "Yes go on," "Yes, pricing." A follow-up turn inside AI Mode gets processed as a brand new Google search, and anything cited in the response is logged as an impression against that follow-up's text.

That leak matters more now because the report built to show AI visibility on purpose doesn't do this. Google's Generative AI performance report launched in June and went live for everyone on August 11. It has five data views: impressions, pages, countries, devices and dates. Queries and clicks aren't among them, and there's no API route around it either. SEO Suganthan Mohanadasan checked: the Search Analytics API's type parameter still stops at googleNews, the searchAppearance dimension returns nothing AI-related, and the BigQuery bulk export schema has no AI column. The export button in the UI is the only way the data leaves Google, and even that has no query field.
So the classic performance report, the one nobody thought to filter, is currently the only place to see the actual language people use inside AI Mode. Mohanadasan sorted 16 months of his own Search Console history and found these fragments split into seven kinds: reply artifacts ("yes," "sure"), pivot follow-ups ("what about resend?"), conversational questions ("is it free"), tracker probes (scheduled prompts from AI visibility tools), agent harnesses (full instruction sets logged as one string), pasted strings, and long queries with no other marker.
Across his site that came to 1,127 queries and 20,300 impressions over 16 months, against millions of ordinary impressions. Individual rows carry real signal: "yes" alone shows 110 impressions, six clicks and an average position of 4.5, a ranking that only makes sense inside an AI answer block. His conversational-question bucket alone ran 559 queries, 8,834 impressions and 13 clicks. Tracker probes accounted for 124 queries and 2,902 impressions, with one facet-matrix prompt running daily for 59 consecutive days against a single review page.
None of this is first-party. It's one analyst's own property, not ours or a client's, and Google has not published aggregate numbers on how common the leak is across the wider web.
Frequently asked questions
Does Google's new Generative AI performance report show which queries triggered an AI citation?+
No. It has five data views, impressions, pages, countries, devices and dates, and leaves out queries and clicks entirely. There is no API workaround either, confirmed against the Search Analytics API, the searchAppearance dimension and the BigQuery bulk export schema.
Can I see AI Mode conversation fragments in my own Search Console account?+
Yes, inside the ordinary performance report. Mohanadasan's method: filter Search results by query with a custom regex, ^(yes|yeah|ok|okay|sure)[?!.,]*$, to isolate reply artifacts, then check for the other six kinds he defined: pivot follow-ups, conversational questions, tracker probes, agent harnesses, pasted strings and long uncategorized queries.
Which brands does ChatGPT name for your buyers' questions?
Explore with AI
[ Keep exploring ]
Terms in this article

Co-founder & Head of AI Innovation, Index Lab
Guilherme co-founded Index Lab, an AEO/GEO agency that makes brands the answer AI gives across ChatGPT, Gemini and Perplexity, taking clients from zero AI visibility to top recommendations.
[ Related reading ]
Listicles still dominate Google's top 10, but the winners keep turning over
New research on 3,000 ranking listicles finds the format never left. What changed is which URL wins, and what makes one listicle outrank another.
Read →NewsGoogle AI Mode picks different sellers than the shopping carousel
New tracking data finds AI Mode overlaps with Google's own product carousel only 1.28% of the time, and swaps in a different top seller for 49.6% of the matches it does find.
Read →NewsProduct pages beat Reddit for AI citations, but the data is B2B only
Ten Speed's study found product pages draw 24.1% of AI citations against 4.2% for Reddit, YouTube and forums combined, and the number survived six fact-check questions. The dataset behind it is B2B SaaS and professional services clients only.
Read →See where your brand stands in AI.
Free visibility check →