Google can isolate generative-AI visibility in one report and expose query text in another.
Google Search Console is beginning to expose fragments of the conversations people have inside AI Mode. The problem is that Google still does not label the fragments as AI Mode activity, leaving publishers with a view of conversational demand that is real, useful and incomplete.
The reporting gap is now unusually precise. Google's dedicated Search Generative AI report can isolate impressions from AI Overviews and AI Mode, but it does not expose query text. The general Search Performance report exposes query text, but it does not identify whether a row came from traditional search, an AI Overview or AI Mode.[1]
Google can therefore show publishers where generative visibility occurred or what people typed. It still cannot show both in the same measurement view.
<h2>Why yes can become a search query</h2>
Google's counting documentation explains the underlying behaviour. When someone asks a follow-up question inside AI Mode, Google treats that interaction as a new query. Impressions, position and clicks in the resulting response are attributed to that new query.[2]
That makes conversational fragments plausible Search Console rows. Search Engine Roundtable documented examples including ‘yes’, ‘yes go on’ and ‘yes, pricing’ from a practitioner account. Google's John Mueller pointed back to the documentation confirming that AI Mode and AI Overviews contribute to Search Console's general Performance report.[3]
The examples are strongly consistent with follow-up questions, but Search Console does not attach an AI Mode label to an individual row. A publisher can observe the fragment without receiving direct confirmation of the experience that produced it.
<h2>The measurement split is now the story</h2>
Google introduced dedicated Search Generative AI performance reporting on June 3. The report separates AI Overviews and AI Mode impressions from the rest of Search and provides page, country, device and date dimensions. Google said it was starting with impressions and working with site owners to determine which additional metrics would be useful.[1]
The separation is valuable because publishers can finally identify which pages are appearing in Google's generative experiences. It is still not enough to reconstruct demand. There is no query dimension, so the report cannot connect an impression to the exact question or follow-up that triggered the response.
The general Performance report creates the opposite problem. It contains query text and the associated clicks, impressions and positions, but the rows remain blended across Google's search experiences. Publishers may see a sentence fragment that looks like a reply in an AI conversation, yet they cannot filter the report to prove that origin.
<h2>Conversational fragments are not all the same</h2>
Suganthan Mohanadasan tested the scale of this phenomenon across 16 months of his own Search Console data. He reported identifying 1,127 candidate conversational or machine-generated queries representing about 20,300 impressions. His rule-based classifier grouped the strings into seven categories, including very short affirmative replies, sentence-like phrasing, first-person prompts, long-tail conversational searches and possible internal query fragments.[4]
Those categories are useful analytical hypotheses, not Google labels. A first-person query can originate in a conventional search box. A long question can be typed directly into Google. Very short words can be navigation, voice input, noise or a genuine AI Mode reply. Mohanadasan is explicit that the work surfaces candidates rather than proving provenance.[4]
The ambiguity is not a reason to discard the data. It is a reason to classify it conservatively. Short replies such as ‘yes go on’ are more suggestive of a conversational context than a complete research question, while ordinary long-tail searches should carry much lower confidence.
<h2>Search Console may now contain the user's second question</h2>
Traditional search measurement has been organised around the initial query. AI Mode changes the unit. The user can refine the request, reject an answer, ask for pricing, introduce a constraint and request a comparison without beginning a visibly separate search session. Google nevertheless counts each follow-up as a new query for Search Console purposes.[2]
This gives publishers a limited view into the second and third questions people ask after the initial answer. Those questions can reveal where confidence breaks down, which details remain unresolved and what information the first response did not supply.
A pricing follow-up suggests that the user has moved from discovery toward evaluation. A request for proof can expose a credibility gap. A correction can reveal ambiguous terminology or weak entity resolution. A short acceptance followed by a new constraint can show the decision criteria introduced later in the journey.
<h2>The data should change content strategy, not become a new vanity metric</h2>
The practical use is qualitative before it is quantitative. Publishers should review conversational-looking rows for repeated themes, then compare those themes with the pages receiving generative impressions. The goal is to identify missing evidence, unclear explanations and decision-stage questions that deserve stronger coverage.
A page that appears regularly in the generative report but is associated with fragments about pricing or proof may need clearer commercial details, dates, methodology or citations. A cluster of correction-like queries may indicate that the brand's terminology is ambiguous. Repeated comparison language may justify a transparent decision framework rather than another generic ‘best’ list.
The denominator still matters. Search Console exposes only the queries associated with the publisher's own visibility, subject to Google's reporting thresholds and interface limits. It is not a census of everything people ask in AI Mode. It also does not reveal prompts that produced no link to the site.
<h2>What Google still needs to provide</h2>
The next useful step is not another total. Publishers need a privacy-preserving way to connect query themes with generative visibility. That could mean aggregated query clusters, a high-confidence conversational-query flag or a report that joins query categories to AI-visible pages without exposing individual conversations.
Google should also provide enough metadata to distinguish AI Overviews from AI Mode and to separate the opening query from later follow-ups. Without that structure, publishers must infer intent from linguistic fragments and combine reports manually.
The current design is therefore informative but structurally incomplete. Google has separated the new answer surface from conventional search, while leaving the corresponding demand signals in a different report.
<h2>AEO Updates Takeaway</h2>
AEO practitioners should treat conversational-looking Search Console rows as evidence candidates, not confirmed AI Mode prompts. Use Google's counting rules to understand why follow-ups can appear, classify fragments by confidence, compare the themes with generative-AI page impressions and look for unmet questions that deserve clearer evidence. Do not report the resulting count as total AI demand. The important development is not that Search Console has become a clean AI query tool. It is that fragments of the AI conversation are becoming measurable before Google has provided the labels required to interpret them safely.
<h3>References</h3>
[1] <a href="https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports" target="_blank" rel="noopener noreferrer">Google Search Central: ‘Introducing Search Generative AI performance reports in Search Console’</a>, June 3, 2026.
[2] <a href="https://support.google.com/webmasters/answer/7042828?hl=en" target="_blank" rel="noopener noreferrer">Google Search Console Help: ‘What are impressions, position, and clicks?’</a>, accessed August 24, 2026.
[3] <a href="https://www.seroundtable.com/google-search-console-ai-mode-queries-41821.html" target="_blank" rel="noopener noreferrer">Search Engine Roundtable: ‘Google Search Console Shows AI Mode Queries, Not In Generative AI Reports’</a>, August 6, 2026.
[4] <a href="https://suganthan.com/blog/ai-mode-queries-search-console/" target="_blank" rel="noopener noreferrer">Suganthan Mohanadasan: ‘Yes, Go On: The AI Conversations Leaking Into Your Search Console’</a>, August 13, 2026.
[5] <a href="https://www.searchenginejournal.com/google-reports-ai-search-impressions-how-to-read-them/582824/" target="_blank" rel="noopener noreferrer">Search Engine Journal: ‘Google Now Reports AI Search Impressions. Here's How To Read Them’</a>, August 4, 2026.