319 AI answers. Zero citations.
Our first study, and the one that started all of this. We ran 80 buyer-intent queries through ChatGPT, Claude, Gemini and Perplexity for three real local service businesses, including our own. Across 319 responses, not one of them was cited as a source. Including us, while we were selling this exact service.
This page previously described nine businesses. It measured three.
An earlier version of this page said the study covered nine businesses across seven verticals with 119 responses. Going back to the raw measurement files in July 2026, that is not what the data shows.
Three businesses were measured, and 119 was the response count for one of them rather than the total. The actual total across all three is 319 responses, which is larger than what we claimed, from a smaller sample than we claimed.
We are fixing it in public rather than quietly, because this site argues that most marketing statistics do not survive being looked up, and that argument is worthless if we do not apply it to ourselves. The corrected numbers are below.
Three businesses, four models, 80 queries.
| Business | Queries | Responses | Mentioned | Cited |
|---|---|---|---|---|
| Watersports and docks, Middle TN | 30 | 120 | 9 | 0 |
| Tree service, Middle TN | 20 | 80 | 62 | 0 |
| Tristar Marketing Solutions | 30 | 119 | 7 | 0 |
| Total | 80 | 319 | 78 | 0 |
Being mentioned and being cited are not the same thing.
The tree service is the interesting row. It was mentioned 62 times across 80 responses, more than any other business in the set, and cited as a source zero times.
That gap is the whole lesson. The models knew the business existed and would name it in passing, but never treated it as a source worth pointing to. Mentions come from a business being present in the world. Citations come from a business being structured, verifiable and quotable, and that second thing is what nobody had built.
The uncomfortable part is the third row. We sold answer engine optimization at the time, and across 119 responses about our own market we were cited zero times. That is what made this worth publishing rather than quietly fixing.
What this study does not support.
Three businesses is a very small sample. It establishes that a zero-citation baseline was possible and common among the businesses we could measure. It does not establish a rate for the market.
Two of the three were our clients. That is a selection problem we cannot argue away. The 2026 Local Visibility Study was designed specifically to fix it, using 64 businesses with no relationship to us.
May 2026 is old for this subject. AI tools have changed since. Citation behavior in particular has moved, and our later work found the four engines now name local businesses at similar rates while largely disagreeing about which ones.
One run per query. These systems are non-deterministic, so a single run is a sample rather than a verdict.
Read the newer study instead.
This one measured a single surface: AI answers. That turned out to be the wrong shape for the question, because a business can be invisible to AI while ranking perfectly well on Google, and the reverse.
The 2026 Local Visibility Study measures 64 businesses across all three surfaces customers actually use, excludes our own clients entirely, and publishes its method and limitations alongside the findings. It supersedes this page. This one stays up because deleting a corrected study would be its own kind of dishonesty.
See whether AI names your business.
Free 30-second check across four AI tools plus Google and the map pack. See who gets the call instead of you, and a link to your result you can keep. No sales call.
Check My Business Free →