Type your business category and city into ChatGPT — "best plumber in Leeds," "recommend a physiotherapist in Austin," "where should I get my car serviced in Manchester" — and one of two things happens. Either your business comes up, or it doesn't. Most business owners have never actually run this test. They assume ChatGPT works something like Google, where a business either "ranks" or it doesn't, and where checking is as simple as searching your own name.
It isn't that simple, and the difference matters. Here's what actually happens when someone asks ChatGPT to recommend a business like yours, how to check it properly, and what the answer tells you.
The short answer: in SOCi's tested sample, ChatGPT named 1.2% of locations
SOCi's 2026 Local Visibility Index tested about 350,000 locations of 2,751 multi-location brands. ChatGPT recommended just 1.2% of them.1 For comparison, the same brands appeared in Google's Local 3-Pack 35.9% of the time, Gemini recommended 11.0% and Perplexity 7.4%.1 In this tested sample, ChatGPT cited the fewest locations of the three AI platforms — and AI use for this purpose is growing: 45% of consumers now use AI for local business recommendations at all, up from 6% in 2025.2
That combination — rising usage, a low citation rate in this sample — is why it's worth checking rather than assuming. The 1.2% figure is SOCi's tested rate for its own dataset, not a probability for any specific business: your own result depends on the question asked, your location, your service category, and the exact test conditions. The only way to know where you stand is to run the check below and record what you find.
Yext's API sample showed different source mixes
In Yext's API-derived sample of 6.8 million citations, OpenAI model output drew 48.73% from controllable third-party listings, while Gemini drew 52.15% from first-party websites.3 Inspect sources per platform and prompt.
How to actually check what ChatGPT says
The test is free and takes a few minutes. The method matters more than most people expect, because getting it wrong produces a misleading answer in either direction — a false "I'm invisible" or a false "I'm fine."
Why running it once isn't enough
This is the part most people skip, and it's the part that most changes what the check actually tells you. ChatGPT doesn't retrieve a fixed answer the way a search engine returns a stored index entry. It generates a response by sampling from a distribution of likely next words, which means identical questions asked in separate sessions can surface different businesses, different phrasing, or a different level of detail — and retrieval, location signals, and session context can add further variation on top of the model's own sampling.
"One ChatGPT answer is one observation under specific conditions — not a 50/50 coin flip, and not proof either way. A few repeat runs can show you variability, but they still can't establish a stable recommendation rate or explain why it happens."
Practically: if you ask once and you're not mentioned, that's not proof of absence — run it again, and possibly a third time, before concluding anything. If you ask once and you are mentioned, that's encouraging but not a guarantee it happens reliably — the same variance cuts both ways. Three to five runs, done informally, will show you a pattern of variability; a full statistical measurement would use far more samples and report a confidence interval rather than a single yes or no, and even five runs don't add up to a reliable rate — this is a manual protocol you can run yourself today, and it's on the roadmap to become an automated feature of AEO Radar's monitoring.
What the answer tells you to fix
What to check for each kind of result:
| What you see | What to check |
|---|---|
| Never mentioned across 5 runs | Audit relevant directory accuracy and compare it with sources shown in repeated direct-app answers |
| Mentioned, but details are wrong | Inconsistent listing data across directories — ChatGPT may be citing a stale or duplicate profile |
| Competitors appear, you don't | Compare cited facts and relevant service fit. The cause of selection remains unknown; SOCi's 4.3★ average for ChatGPT-recommended locations is not a cutoff.1 |
| Mentioned inconsistently (2 of 5 runs) | Observed variability; cause unknown. Save the answers, source URLs, dates, location and settings, and investigate only verifiable fact differences. |
Notes and sources
1 SOCi, "2026 Local Visibility Index," January 2026. Dataset: 2,751 multi-location brands, about 350,000 US locations. Benchmarks for ChatGPT, Perplexity and Gemini: share of locations recommended 1.2%, 7.4% and 11.0% (Google 3-Pack: 35.9%); average rating of recommended locations 4.3, 4.2 and 3.9 stars; business-profile accuracy 68.3%, 68.0% and 100.0%, measured against Google Maps data.
2 BrightLocal, "Local Consumer Review Survey" 2026. Sample: n=1,002 US consumers. Finding cited: 45% of consumers used AI for local business recommendations in 2026, up from 6% in 2025.
3 Yext, "AI Citations, User Locations, & Query Context," Yext Research, Oct 9, 2025. 6.8 million citations from ~1.6 million questions across OpenAI, Gemini and Perplexity APIs, July-August 2025, four industries. In its API-derived sample, OpenAI model output drew 48.73% from controllable third-party listings and Gemini drew 52.15% from first-party websites. API citations can differ from direct user applications.
RaveHQ Insights share our views, based on our research and the sources cited. They are general information, not legal, financial or professional advice; check what applies to your business before acting.