If you work in Generative Engine Optimisation, you’ve probably noticed something unsettling in Google Search Console. Even now that GSC counts AI Overviews traffic and reports AI Mode surfaces alongside standard Search results, the question-based conversational queries that look most like real prompts are largely missing from the query report. And when GSC does surface them, they arrive without the context that would make them actionable.
What GSC actually shows
Google Search Console reports on the queries where your site earned an impression or a click in Google Search. As of early 2026, that includes traffic from AI Overviews and AI Mode alongside traditional results — Google reports these as unified “Search” data, which is good and bad. Good because you can see totals. Bad because you can’t separate them.
But even setting aside the aggregation problem, there’s a deeper issue: the queries GSC captures are not the same as the prompts real users are typing into ChatGPT, Claude, Gemini, or Perplexity. They’re the search queries that led to your Google impression — a fundamentally different data source.
The prompt-attribution gap
Here’s what happens in practice. A user types “which pension provider is best for someone self-employed under 50” into ChatGPT. ChatGPT synthesises a response, cites two or three sources, the user reads the answer. If they click through to your site, that click doesn’t come through Google — it comes through ChatGPT’s search grounding. GSC never sees the prompt that generated the click.
Meanwhile, GSC shows you that someone searched “self employed pensions comparison” on Google and clicked your result. Related query, different funnel. GSC gives you Google’s version of intent, not the prompt-based version.
For AI Search-native traffic arriving via Google’s own AI Mode, GSC does report it — but as an aggregated impression figure, without exposing the specific conversational query that surfaced you. The prompt itself remains invisible.
Why conversational queries are invisible
Multi-turn prompts are the most obvious blind spot. Someone might have a five-turn conversation with ChatGPT that ends with them clicking your pricing page. Which turn earned the click? What was the context? GSC will never tell you because GSC is a search log, not a conversation log.
Even in single-turn cases, the phrasing that surfaces citations in an LLM is often more structured or more colloquial than the equivalent Google query. “Best pension provider UK 2026” is what people type into Google. “Compare Aviva, Standard Life, and Legal & General for pension consolidation for a 42-year-old sole trader” is what people type into ChatGPT. Same underlying intent, wildly different query shapes — and only the first one appears in GSC.
The prompts that matter to your brand are being asked outside Google. Your measurement needs to move with them.
What this means for GEO practitioners
Three practical implications.
Reporting is systematically biased toward what GSC can see
GEO campaign reports built on GSC data alone will misrepresent your AI Search performance. You’ll see the Google-search side of the funnel but not the LLM side. This is why so many “AI SEO” reports over-index on Google AI Overviews specifically — that’s the one AI Search surface GSC does report, however imperfectly.
Content optimisation for LLM citation is optimisation for prompts, not queries
You need visibility into which prompts your content is being pulled for — not just which Google queries brought traffic. GSC can’t give you this, and no amount of pivoting Google Search Console data will substitute.
Executive reporting risks looking encouraging but hollow
Reports that treat GSC as the source of truth for “AI Search traffic” will produce numbers that look encouraging but don’t reflect what LLMs are actually doing with your content. When boards start asking specific questions about LLM performance, the gap becomes visible fast.
Where to look instead
The tooling has to change. AI Search monitoring means running prompts directly against ChatGPT, Claude, Gemini, and Perplexity, extracting the citations they surface, and mapping those to your content. That’s what GEO Jetpack does — but the important point isn’t the tool. It’s the methodology.
Prompt-level monitoring is a separate discipline from search query analysis. It requires a separate data source, a separate cadence, and a separate reporting layer. Trying to make GSC do this work will produce the illusion of measurement without the substance.
- GSC reports Google-search queries, not the prompts users type into LLMs
- AI Overviews and AI Mode traffic appear in GSC — but without prompt-level attribution
- Multi-turn conversations and conversational query phrasing are structurally invisible
- Prompt-based AI Search monitoring requires a fundamentally different data source
The takeaway
Google Search Console is a superb tool for what it does — Google search performance reporting. It’s not, and cannot be, an AI Search monitoring tool. Treating it as one will produce blind spots at exactly the moment your buyers are quietly shifting their information-gathering to LLMs.
The prompts that matter to your brand are being asked outside Google. Your measurement needs to move with them.