There is no AI Overviews row: designing GEO measurement around what you can see
Published · Updated
The first GEO question is usually a citation count, and the available reporting does not produce one. Here is what you can measure instead, and how to date the windows so the numbers mean something.
The first GEO question from a stakeholder is almost always a count: how many times did an AI answer cite us last month? The reporting does not produce that number. Google documents that appearances in AI Overviews and AI Mode are reported in the Performance report, within the "Web" search type. There is no separate row to isolate, so the measurement plan has to be built around what is actually visible.
No separate row means no clean before and after
Because AI feature impressions and clicks sit inside the same "Web" search type as ordinary results, a rise after a GEO change cannot be attributed to that change by the report alone. Fan-out makes attribution harder again: both AI Overviews and AI Mode may issue multiple related searches across subtopics and data sources, so the page surfaced may not be the page you edited. Plan for population-level comparisons, not per-answer accounting.
Eligibility is the one binary you control
Google states that to be eligible as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. That is checkable per URL, on demand, without depending on any answer engine's behaviour. Track it as a count of pages failing the precondition, and treat that count as the first metric in the plan.
Platform-level numbers behave the same way. The Abbys Consult case study records technical SEO issues at launch: 0, and time to first byte under 200 ms. Those describe the state of the platform rather than anyone's citation behaviour, which is exactly why they are stable enough to measure repeatedly. A regression in either is a real, attributable finding; a quiet week in AI answers is not.
Date the change window, don't read day-of movement
Opt-in and opt-out decisions take time to propagate. OpenAI documents that for search results it can take around 24 hours from a site's robots.txt update for their systems to adjust, and the two tags carry different meanings: disallowing GPTBot indicates content should not be used in training generative AI foundation models, while sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though they can still appear as navigational links.
For measurement, the consequence is procedural. Record the exact date and time of any robots.txt change, and discard the first day of data after it. Without a dated annotation, a lagged propagation and a routine traffic fluctuation look identical in the chart, and someone will draw a conclusion from the overlap.
Proxies that already come with numbers
Structured data gives a worked example of the shape of measurement available. Google reports that Rotten Tomatoes added structured data to 100,000 unique pages and measured a 25% higher click-through rate for pages enhanced with structured data, compared to pages without. That is a comparison between two sets of pages, not a count of individual placements, and it is the comparison a site can reproduce.
Pick the metric that matches the eligibility rule. All required properties must be included for an object to be eligible for enhanced display, and it is more important to supply fewer but complete and accurate recommended properties than every possible one with badly formed data. So measure the share of pages whose required properties are complete, not the number of schema types deployed.
Audit scores measure your implementation, not your citations
The Matthieu Pesesse platform case study records a GEO audit score of 66 to 90 or above after overhaul. Be precise about what such a score is: an assessment of the site's own characteristics, repeatable on demand, independent of whether any model cited a page this week. That makes it a good regression check and a poor stand-in for visibility. Report it as implementation quality, labelled as such.
Don't track files that aren't signals
Google is explicit that you don't need to create new machine-readable files, AI text files, or markup to appear in these features. The llms.txt file is a proposal, with an H1 naming the project as its only required section and an Optional section used by convention for links an agent can skip when a shorter context is needed. Shipping it is cheap; putting its existence on a GEO dashboard reports nothing. Measurement pressure also invites shortcuts worth refusing: don't create blank or empty pages just to hold structured data, and don't add structured data about information that is not visible to the user, even when it is accurate.
Write the measurement plan before the GEO work starts, and write down what it cannot answer. What you can report: the count of pages failing the indexed-and-snippet-eligible precondition, the share of pages with required properties complete, audit score deltas, and click-through comparisons across page sets, each tied to a dated change window. If someone asks for a citation count, the honest answer is that the report does not contain one.
Sources
Google Search Central — AI features and your website — https://developers.google.com/search/docs/appearance/ai-features
Google Search Central — Introduction to structured data markup in Google Search — https://developers.google.com/search/docs/appearance/structured-data/intro-structured-data
OpenAI — Overview of OpenAI Crawlers — https://developers.openai.com/api/docs/bots
Neurolinks case study — A personal brand that publishes itself — https://neurolinks.be/work/matthieu-pesesse-media
Neurolinks case study — A consulting firm, dressed for trust — https://neurolinks.be/work/abbys-consult
Working on a project where these methods apply?