AI visibility tracking means measuring whether AI answer engines mention and cite your brand when buyers ask questions in your category. Most tools report a mention count. That number alone is not actionable. In our own measurement of 23 buyer prompts, 91% triggered a Google AI Overview and we were cited in zero of the non-branded ones — and the reason turned out to be something a mention counter would never have shown us.
This is not a roundup of tools. We built a tracker, pointed it at our own category, and are publishing what came back, including the parts that make us look bad. If you are evaluating AI visibility tracking, the useful question is not which dashboard is prettiest. It is which number tells you what to do next.
On 26 July 2026 we ran 23 prompts through two surfaces, in the US market, in English. The prompt set was built the way a buyer actually searches: four branded prompts about our own company, and 19 non-branded prompts covering the questions our category gets asked — pricing, vendor selection, industry-specific automation, build-versus-buy.
Two signals, both machine-readable: Google AI Overview. For each query: does an AI Overview trigger at all, is our domain among its cited references, and which domains are cited instead. We also captured our classic organic position for the same query, which turned out to matter more than we expected. Embeddings retrieval. Whether our domain surfaces in the vector index layer that many AI applications query before generating an answer.
What we deliberately did not claim to measure: the answer text inside ChatGPT, Claude, or Perplexity. No API exposes what those systems said about your brand in a real user session. Any tool claiming to track this is running its own prompts through the API and inferring, which is a reasonable proxy but is not the same thing. We think vendors should say so plainly, so we are saying it about our own work first.
21 of 23 prompts (91%) returned a Google AI Overview. Not a handful. Not the informational ones only. Nearly every commercial question our buyers ask now gets answered above the organic results, by a system that cites a handful of sources and moves on.
That reframes what a ranking is worth. Position 4 on a query where an AI Overview occupies the first screen is not position 4 in any meaningful sense. If you are still reporting average position without segmenting by whether an AI Overview fires, your report is describing a page layout that most of your buyers are no longer seeing first.
Here is the uncomfortable table. Of our four branded prompts, all four triggered an AI Overview and three cited us. Of our 19 non-branded prompts, 17 triggered an AI Overview and zero cited us.
A mention-counting tool would have reported "3 citations, 13% visibility" and drawn a line on a chart. That framing would have been actively misleading. The three citations were on queries containing our own company name — queries where a buyer already knows who we are. On every question where someone was actually shopping, we did not exist.
Branded visibility and non-branded visibility are different metrics measuring different business realities, and averaging them together produces a number that flatters you. If your AI visibility report shows a single blended percentage, split it. The branded number tells you whether your entity is understood. The non-branded number tells you whether you are in the consideration set. Only the second one is a growth metric.
This is the finding that changed how we plan, and it comes from a second, smaller run rather than the study above. Two days later, on 28 July 2026, we re-tested six of the 19 non-branded queries. Four of them returned an AI Overview. For those four we captured every domain cited, then checked whether those same domains also ranked in the classic organic results for the identical query. So the overlap figures below describe four AI Overviews from 28 July, not the 21 from 26 July.
Across the 31 citations those four answers issued: 74% also ranked in the classic organic top 20 48% ranked in the top 10 Only 25% were cited without ranking anywhere in the first two pages
The sample is small and we would not present it as a universal law. But the direction is clear enough to act on: for most queries, Google is drawing its AI Overview citations from a pool it has already decided to rank. There is no separate side door. The bulk of "AI visibility optimization" is ordinary ranking work, and a tracker that reports citations without reporting your organic position for the same query has hidden the causal variable.
That 25% residual is real and worth chasing — it is where format, structure, and being the definitive source on a narrow question earn a citation without a ranking. But it is the exception, not the strategy. Anyone selling you an AEO methodology that bypasses ranking should be asked to show the overlap data.
One prompt in our set asked for the top AI automation agencies in a specific city. The AI Overview cited exactly four domains, and every one of them was a directory: Clutch, DesignRush, GoodFirms, and The Manifest. No agency websites at all. Not the market leaders, not the best-optimized pages. Directories.
For location-shaped queries, Google appears to treat curated third-party lists as the authoritative source and does not read vendor sites for the answer. If your visibility strategy for local intent is on-page content, you are optimizing an asset the system is not consulting. The lever is a complete, claimed directory profile.
We found this because we measured. We would not have found it by reading a best-practices article, and a mention-count dashboard would have shown a zero with no explanation attached.
Back to the 26 July study. Across its 21 AI Overviews, 149 citations — counting each domain once per answer, out of 174 raw references — went to 116 unique domains. The concentration is lower than we expected — this is not a cartel of five big publishers.