Step-by-step guide Get Recommended by ChatGPT: local service businesses →
Ask ChatGPT for a good physiotherapist nearby and you won't get ten blue links and a map. You'll get a short list of names, usually about four. If your business isn't on that list, it isn't on page two. It simply isn't there. And people are asking: in BrightLocal's consumer research, published on 10 March 2026, 45% of consumers had used AI tools for local business recommendations, up from 6% a year earlier. Most local businesses have never checked what those tools say about them. Here's how ChatGPT, Google's AI Mode and Perplexity choose which local businesses to recommend, what the largest study so far found, what we saw when we ran the test ourselves, and how to check and improve your own position.
How an AI assistant builds a local answer
There is no advert auction behind a local AI answer. The assistant runs a handful of searches in the background, reads what the web says about businesses near the place in the question, and writes a short list with reasons. BrightLocal calls those background searches fan-out queries. One question about a family lawyer becomes several narrower searches for specialists, reviews and locations, and the answer blends what comes back.
On 16 September 2026, BrightLocal published the largest study of local AI answers so far. It covers about 200,000 non-branded local searches in ChatGPT, Google AI Mode and Google's AI Overviews, repeated over 60 days for about 1,300 business locations, and 1.9 million citations: the links each assistant showed as its sources. ChatGPT named 4.1 businesses per answer on average, AI Mode 3.5 and AI Overviews 2.5. This is where the citations came from:
| Source | Share of all citations |
|---|---|
| Business websites | 42% |
| Google Business Profile | 28.6% |
| Yelp | 9.5% |
| 2.2% | |
| Tripadvisor | 1.8% |
| Industry directories and other sites | The rest |
Two things stand out. First, your own website is the biggest single source, and 93% of all the domains cited were business websites. Assistants quote sentences, so they need sentences that say what you do, for whom, where, and how to book. Second, the sources differ by assistant. Google Business Profile feeds Google's two AI surfaces heavily, but BrightLocal found that ChatGPT hardly uses it, and points to ChatGPT's ties with Yelp and Bing instead. A perfect Google profile on its own does not cover ChatGPT.
Three findings that change how you should measure
- Your Maps ranking doesn't carry over. A tracked business appeared in 66% of Google Maps results, but in only 32–38% of AI answers (38% in AI Overviews, 33% in ChatGPT, 32% in AI Mode). In BrightLocal's words, "Ranking on Google Maps doesn't translate directly to AI visibility."
- The answer changes every time you ask. Repeat runs of the same question overlapped by only 20–33% (23% in ChatGPT, 21% in AI Mode). A business that appeared once came back in about half of later runs.
- Close still counts. Between 53% and 59% of the businesses picked were within 5 km of the searcher. Storefronts had a 33.6% mention rate; service-area businesses with no public address had 29.4%.
The middle point is the one most people miss. If you ask ChatGPT once, see your name and relax, you've learned very little. If you ask once and don't see it, you've learned just as little. The useful number is a mention rate: the share of answers that name you, across several runs.
What we saw when we tested it in Amsterdam
To test the method behind our new guide, we ran it ourselves on 9 November 2026. We asked six real patient questions about physiotherapy in Amsterdam in ChatGPT (in a temporary chat) and Google AI Mode, plus Perplexity until it asked us to sign up after two questions. We repeated one question in ChatGPT and AI Mode to see how far the answer moved. As in the guide, the practices are anonymised as letters.
The answers moved exactly as the study predicts. ChatGPT changed its first choice between two identical runs, minutes apart. Google AI Mode kept only one of its four practices. Across 20 answers, about 30 different practices were named. What decided the winners was more interesting:
| Question | Who won | The reason the assistant gave |
|---|---|---|
| Knee pain after running | Practice A (ChatGPT, first run) | A therapist profile on its site says they specialise in running injuries |
| Desk-related back pain | Practice H | Its page says it specifically treats lower-back pain from prolonged sitting |
| English-speaking physio | Practice E, in both assistants | Google cited its dedicated Expats page |
| Evening appointments | Practice I, after the closest practice | The late evening hours stated on its website |
When we asked about dry needling, ChatGPT went further and sorted practices into those that explicitly confirm the treatment and those that merely list it.
The pattern: in every case, the winner had written down, in plain words, the exact thing the customer asked for, on a page the assistant could find and quote. Reviews came in as supporting reasons, with star ratings and review counts quoted back. Practical facts broke ties: insurer contracts, no GP referral needed, opening hours. And for one question, Perplexity's only pick came from a single Reddit comment.
How to check your own business in about an hour
You don't need a paid tool for this. You need twelve questions, three assistants and a spreadsheet, spread over one week.
- Write twelve customer questions. Four that describe a problem in the customer's own words, four that name a service, and four about place and practical needs: neighbourhood, opening hours, language, accessibility. Never put your business name in them.
- Ask them in clean sessions. In ChatGPT, switch on Temporary chat so your history doesn't colour the answers. Use Google's AI Mode in a private window. In our test, Perplexity stopped signed-out visitors after two questions, so use a free account there.
- Score every answer. R if you're recommended (named first, or with a reason to pick you), M if you're mentioned, a dash if you're absent. Note who came first.
- Ask why, and where from. On three answers a competitor won, ask the assistant why it recommended them, then which websites it used. The reasons show what you're missing. The websites become your source-gap list: the places assistants read where you're absent or out of date.
- Run it twice more. Two or three days apart, ask the same questions again and score them. Your mention rate per assistant is the answers that named you divided by all answers: out of 36 after three runs.
From then on, compare averages with averages. Your first run might show you in two answers and the second in six. Neither is the truth. The average is.
What to fix first
These fixes feed every assistant at once, which is why they come first. None needs a developer except schema markup, and even that is mostly copy and paste.
- One plain answer sentence on your homepage. What you do, for whom, where, and the practical facts people ask about. The guide's fictional example practice swaps a line about being passionate about movement for this: a physiotherapy practice in Amsterdam-Oost for running injuries, back and neck pain, and rehabilitation after knee surgery, English-speaking, with evening appointments until 21:00 and no GP referral needed.
- A page per service, in your customers' words. Title it the way people say it ("Running injuries", not "Lower-extremity rehabilitation"), add a short FAQ built from your audit questions, and state the area you serve. A page for a group you serve well, such as expats or athletes, can win a whole category of questions. Schema markup then states the same facts in a form machines read reliably.
- Three profiles with the same facts. Google Business Profile, Bing Places and Apple Business (the new name for Apple Business Connect). Make your name, address and phone number identical everywhere, down to the abbreviations. Assistants cross-check sources, and mismatches look like two businesses, or an unreliable one. Our local SEO checklist covers the groundwork.
- Reviews, inside the rules. Google's Maps content policy, as it reads in November 2026, lets you ask customers for honest reviews. It forbids incentives, asking only happy customers, pressing people to review while they're on your premises, and asking for specific content. So ask everyone, the same day, with one link.
- Replies that protect privacy. In health, legal and financial services, a warm reply that mentions someone's treatment confirms they are a client and what for. Thank people without confirming they are customers, and take complaints offline.
- The sites on your source-gap list. Claim or correct every listing the assistants cited for competitors, starting with your own industry's directories.
What doesn't work is just as clear. Fake or bought reviews break platform rules. Keyword-stuffed pages read as spam to people and to assistants. And treat any offer of a bought spot in AI answers with suspicion: the assistants build their lists from what they can find and check. Every fix above simply makes true information about your business easier to find, check and quote. After 30 days, run exactly the same audit again and compare the averages, then judge progress by the trend over two or three months rather than by one re-test.
The full method, in one guide
We've written all of this up as Get Recommended by ChatGPT, local service businesses edition, a 21-page fast-track guide for clinics, practices, salons, trades and studios. It walks you through the one-week audit with a printable score sheet, the full Amsterdam test, a 30-day plan for your website, profiles and reviews, privacy-safe reply templates and ten copy-paste prompts. It sits alongside our other practical titles in the Neurova AI guides.
AI answers will keep moving. What they reward has stayed steady in every test we've run: specific, consistent facts that a machine can find and check. Neurova AI builds AI systems to healthcare standards from Eindhoven, and visibility work like this is part of what we do for practices and service businesses. If you'd rather have it done for you, tell us what you need.
Frequently asked questions
How does ChatGPT decide which local businesses to recommend? It runs several web searches in the background, reads what it finds about businesses near the place in the question, and writes a short list with reasons. In BrightLocal's September 2026 study of about 200,000 local AI searches, business websites made up 42% of all citations, Google Business Profile 28.6% and Yelp 9.5%. ChatGPT relies far less on Google Business Profile than Google's own AI answers do.
Why does ChatGPT recommend different businesses each time I ask? Each answer is generated fresh. BrightLocal found only 20 to 33% overlap between repeat runs of the same local question, and a business that appeared once came back in about half of later runs. That is why you should measure a mention rate across several runs instead of judging a single answer.
If I rank well on Google Maps, will AI assistants recommend me? Not automatically. In the same study, a tracked business appeared in 66% of Google Maps results but in only 32 to 38% of AI answers. The assistants also read your website, review sites and directories, so a strong Maps position is a good start but not the whole picture.
Which AI assistants should a local business check? Start with the ones your customers use. In BrightLocal's March 2026 consumer research, 45% of consumers had used AI tools for local business recommendations, with ChatGPT used by 31% and Google's AI Mode by 23%. We test ChatGPT, Google's AI Mode and Perplexity, because each one draws on different sources.