If your business has fewer than 25 Google reviews, ChatGPT names you in roughly 3.4% of relevant local searches. Above 1,000 reviews, that rises to 22.2%. Those figures come from a study of 2,975 businesses across 39 trades and 49 cities, run in September 2026. The same study found that star rating made no difference whatsoever, and that ChatGPT and Claude agreed on which business to recommend only 28% of the time.
The number that should worry most local businesses
The study, by 8ight.AI, asked one question through each engine’s API with web search enabled: “Who is the best [trade] in [city]? Name specific businesses.” It ran across 516 trade-and-city pairs, pulling businesses from Google Maps data.
Across every business tested, ChatGPT named 12.4% of them. Broken down by review count, the pattern is stark.
Google reviews | Businesses tested | Named by ChatGPT | Rate |
|---|---|---|---|
Under 25 | 435 | 15 | 3.4% |
25–99 | 903 | 90 | 10.0% |
100–249 | 775 | 108 | 13.9% |
250–999 | 527 | 110 | 20.9% |
1,000+ | 135 | 30 | 22.2% |
Read the first row carefully. Fewer than 25 reviews means a 3.4% chance of being named, and that figure rests on just 15 businesses out of 435. For a local service business with a handful of reviews, AI search is currently close to a closed door.
Rating does not matter. Review count does.
This is the finding most likely to change what you do on Monday.
The median star rating was 4.9 for businesses ChatGPT named and 4.9 for businesses it did not. No pattern at all. The median review count, meanwhile, was 175 for named businesses and 97 for unnamed ones.
Almost every local SEO conversation focuses on protecting a rating. The data says rating is table stakes and volume is the differentiator. Twenty reviews at 5.0 performs worse than two hundred at 4.6.
If you take one action from this post, it is to build a systematic process for asking satisfied customers for reviews. Not a campaign. A process that runs every month.
Even the market leader gets left out
Before anyone concludes that reviews alone solve this, one more figure.
In the 285 city-and-trade pairs where at least four reviewed businesses existed, the single most-reviewed business in the market was named 22.8% of the time. The other businesses were named 10.3% of the time. So being the clear leader roughly doubles your odds, and still leaves you out of three answers in four.
This is a probabilistic system, not a ranking. There is no position one to win. The realistic goal is to move from rarely named to sometimes named, and the honest framing with any client is exactly that.
The engines disagree with each other profoundly
Of 1,775 businesses tested across both ChatGPT and Claude, 294 were named by at least one engine. Only 81 of those, 28%, were named by both. Ninety-six were named by ChatGPT alone and 117 by Claude alone.
Their sources diverge even more sharply.
ChatGPT cited Yelp in zero of 1,412 answers. Claude cited Yelp in 202 of 218, or 92.7%. ChatGPT cited the Better Business Bureau in 38.7% of answers and a government or licensing site in 25.1%, roughly seven in ten of those being licensing or registry pages. Claude cited no government sites at all. Reddit appeared in six ChatGPT answers and none from Claude.
There is no single thing called “optimising for AI”. Getting recommended by Claude means Yelp. Getting recommended by ChatGPT means business licensing records, the BBB, and your own website.
Which trades get named, and which do not
Recommendation rates varied enormously by trade. Among trades tested in twelve or more cities, Pilates studios were named 90.5% of the time and women’s fitness gyms 80%. Towing reached 64.3%, roofing 50%, plumbing 47.4%, HVAC 45%. House cleaning managed 26.7%, barre studios 16.7%, and boutique fitness studios just 6.3%.
The pattern appears to track how many credible businesses exist in a category per city. Where a city has three or four Pilates studios, most get named. Where it has fifty cleaning companies, almost none do.
That has a practical consequence. If you are in a crowded category, review volume matters more, not less.
What ChatGPT actually reads for local queries
A separate BrightLocal study of 800 ChatGPT local searches across 20 verticals and 20 US cities found business websites accounted for 58% of cited sources, business mentions 27%, and directories just 15%. Forums appeared zero times.
Wikipedia alone made up 39% of all mention sources. Among the directories that did appear, Three Best Rated led at 24%, followed by Expertise at 18% and Tripadvisor at 8%. Yelp, Facebook and Google Maps did not appear as directory sources at all.
Worth noting that study ran in November 2024, which makes it nearly two years old in a field that changes quarterly. Treat the direction as useful and the specifics as dated.
A word on the conflicting numbers you will see
You will encounter a widely shared report claiming Yelp is the single largest source of AI local citations, ahead of Google. It is a vendor case study drawn from their own client campaigns, it states no time period or method, and its top-fifteen source counts sum to exactly its headline total, which is not how real citation data behaves.
Set against 8ight’s finding that ChatGPT cited Yelp zero times in 1,412 answers, the discrepancy is not a small one. When studies disagree this violently, check who ran them, over what period, and whether the method is published.
What to actually do
Four things, in order of how much the evidence supports them.
- Build review volume systematically. The 3.4% to 22.2% gradient is the clearest signal in the data. Ask every satisfied customer, every month, through a process that does not depend on anyone remembering.
- Get your licensing and registry records accurate. ChatGPT cited government and licensing sites in a quarter of its answers. State licence boards, trade registries and the BBB are cheap to fix and almost nobody thinks to check them.
- Claim and complete Yelp if you care about Claude. It does nothing for ChatGPT and almost everything for Claude.
- Make your own website the best source about you. It was the single largest cited category at 58%. Clear service pages, clear locations, clear contact details, answers written in plain language.
Where you stand right now
The most useful thing you can do this week is find out whether you appear at all. Ask ChatGPT, Claude, Gemini and Perplexity the question your customers would ask, several times, and write down who gets named instead of you.
Run it more than once. The same question asked 39 minutes apart produced only about 29% overlap in named businesses, so a single check tells you very little.
We will run it for you
We check twenty buyer-intent prompts across four AI assistants, record where you appear and who appears instead, and audit the signals the research says actually matter: review volume, licensing records, directory presence and your own site.
No charge, and the report is yours either way.
Frequently Asked Questions
Does ChatGPT recommend local businesses?
Yes, but selectively. In a study of 2,975 businesses across 39 trades and 49 cities, ChatGPT named 12.4% of them when asked for the best provider in a city. The rate ranged from 3.4% for businesses with under 25 Google reviews to 22.2% for those with over 1,000.
How many Google reviews do I need to be recommended by ChatGPT?
More is consistently better, with the steepest gains early. Businesses under 25 reviews were named 3.4% of the time, rising to 10% at 25–99 reviews, 13.9% at 100–249, 20.9% at 250–999 and 22.2% above 1,000. Median review count was 175 for named businesses against 97 for unnamed.
Does my star rating affect whether AI recommends me?
Apparently not. The median rating was 4.9 for businesses ChatGPT named and 4.9 for those it did not — no measurable difference. Review count showed a clear gradient; rating showed none. Rating appears to be table stakes rather than a differentiator.
Do ChatGPT and Claude recommend the same businesses?
Rarely. Of 294 businesses named by at least one engine, only 81 — 28% — were named by both. Their sources differ even more: ChatGPT cited Yelp in zero of 1,412 answers while Claude cited it in 92.7% of its answers. ChatGPT leaned on licensing records and the BBB, which Claude never cited.
Does Yelp help you get recommended by AI?
It depends entirely on the engine. Claude cited Yelp in 202 of 218 answers. ChatGPT cited it zero times across 1,412 answers. So Yelp matters a great deal for Claude and appears to do nothing for ChatGPT, which favours business websites, licensing records and the BBB.
Why does AI recommend different businesses each time I ask?
Because it is probabilistic rather than a ranking. The same question asked 39 minutes apart produced only about 29% overlap in the businesses named. Even the most-reviewed business in a market was named just 22.8% of the time. A single check tells you very little — run it repeatedly.