Original data, updated continuously
How often do AI assistants actually name a local business?
We ask web-grounded AI assistants the questions real customers ask, like “who's the best plumber near me?”, about the local businesses we measure, every week. This page publishes the aggregate results: 1771 engine answers about 4 US local businesses over the trailing 90 days.
of answers named the business they were asked about (914 of 1771)
US local businesses measured, 48 weekly measurement runs
engines probed, reported per engine, never blended into one score
Per-engine naming rate
Cross-engine overlap in AI answers is known to be low, so a blended score would mislead. Each row is one engine's own rate.
| Engine | Answers | Named the business | Rate | Answered with no sources |
|---|---|---|---|---|
| Perplexity | 514 | 316 | 61% | not recorded |
| ChatGPT (OpenAI) | 452 | 259 | 57% | not recorded |
| Google Gemini | 438 | 274 | 63% | 0% of 209 |
| Grok (xAI) | 367 | 65 | 18% | 86% of 177 |
Where the answers came from
When an engine cites its sources, we record them. These are the domains behind the cited answers in this dataset, by number of answers citing them. Answers that came back with no sources contribute nothing here.
How often is a single check wrong about a market?
Anyone can ask an assistant once, for free. We ask the same questions about the same markets every week, so we can say something a single check cannot: how much one reading disagrees with a sustained one. Across 697 market-days in 61 markets since 2026-07-26:
- Days the business leading a market was not the one leading it over the whole period
- 46.5%
- Days the top three on that day differed from the top three over the whole period
- 73.9%
Read plainly: on roughly half of the days we measured, a business checking once would have seen a different market leader than the sustained answer. That is not a flaw in anyone's method. It is what one sample from a noisy process does.
Where this differs from the rest of this page. Everything above measures real businesses we work with. This section measures markets with no business of ours in them, which is why it can describe a market rather than a customer. The two are never mixed into one number.
What we are measuring, honestly. A market needs four or more dated readings before it counts here, and one of the four prompts rotated between two wordings during this window, so a small part of the movement is our own probe rather than the market. It is far too small to explain a figure near half, and it is not zero.
Method and caveats
Each measured business gets customer-style recommendation prompts (category + city, plus its own tracked phrases) run through every enabled web-grounded engine weekly. An answer counts as naming the business only when the business is identifiably present in the answer text. An engine that fails to answer is omitted, never counted as a miss.
Not every answer is grounded in a live search. Some engines answer from their own recollection of the market and return no sources at all, and an answer like that still names a business (or fails to) exactly like a cited one, so it is counted the same way in the rate. The last column above says how often each engine did that, over the answers where we recorded it. We started recording it on 17 August 2026, so earlier answers in this window read as not recorded rather than as zero.
Who is in it: a business is included only while it has a city on record with us and is actively being measured. Businesses without a city are not local businesses and are excluded, which is why our own products and software customers are not in these figures.
Caveats we want you to know: this is a small, growing network of real businesses, not a random sample of all US businesses; businesses that sign up for visibility monitoring may differ from average; AI answers vary run to run. Rates here describe this dataset over this window, nothing more. Figures update as the network grows.
Cite freely with attribution (CC BY 4.0): GBPmonster, Local AI Visibility dataset, 2026-09-15.
Want to know what AI says about your business? Run the free check, no signup.