Does ChatGPT recommend your business?
There's a good chance a customer asked ChatGPT for a business like yours this week and acted on the answer — without ever seeing your name. Here's how to find out whether you're in that answer, using nothing but a browser and twenty minutes.
How to check, in one paragraph.
Open ChatGPT in a fresh, logged-out session and ask it the way a customer would — "best emergency plumber in Leeds", "who should I call for a blocked drain near me" — then note whether you're named and who's named ahead of you. Repeat with five or six phrasings, then do the same on Gemini and Claude. One answer proves nothing; a pattern across many phrasings and all three models is the real signal.
That's the method in miniature. The rest of this guide turns it into a repeatable audit you can run properly, explains why the answers move around, and covers the only thing that actually changes whether the assistant names you.
The buyer's-seat audit.
Anyone can type their business name into ChatGPT and see a flattering summary — that tells you nothing, because customers don't search for you by name; they search for the job. The buyer's-seat audit forces you to ask the way a stranger who's never heard of you would. It has four steps.
1. Sit in the buyer's seat
Forget your business name entirely. A customer types the problem, not the solution: "my tooth is killing me and my dentist is closed", "cleaner for a end-of-lease in Brunswick", "good family lawyer near me for a custody question". Start every query from the customer's problem and location, never from your brand.
2. Build prompt coverage
The single most common mistake is asking once. Real demand is expressed a dozen ways, and each phrasing can return a different shortlist. For a physio, "best physio near me", "who should I see for a running injury in Brunswick" and "sports physio that takes weekend appointments" are three different prompts. Write down five or six genuine variations per service and ask all of them. This is prompt coverage. It is different from a fan-out search, which is an additional query the answer engine itself may create while researching one prompt.
3. Cross the models and sessions
ChatGPT is not the whole picture. Gemini sits inside Google's ecosystem, Claude draws on its own sources, and buyers use all three. Run your phrasings on each, in fresh, logged-out sessions so your own history doesn't quietly bias the answer toward businesses you've clicked before. If you're logged in, the model may flatter you with your own past behaviour — exactly the blind spot this audit exists to remove.
4. Log the answer and its evidence
For every answer, record whether you were named, the tone and position of the mention, and who appeared instead. Also capture any cited pages, search terms or result links the engine chose to disclose. A simple grid — prompts down the side, models across the top — reveals recurring competitors and sources. Do not treat an undisclosed trail as zero retrieval; it is simply unknown.
Why one check is worthless.
If you run the audit twice a week apart, you'll notice the answers aren't identical — and that's the most important thing to understand about AI recommendations. They vary by phrasing (different wording, different shortlist), by session (the same question can resolve differently minutes apart), and over time (models update; the sources they lean on change; competitors do work that moves them). An HVAC company named in "best HVAC in Austin" might vanish from "who fixes air conditioning fast in Austin" — same firm, same city, different words.
This is why a single lucky check breeds false confidence and a single unlucky one breeds needless panic. Measure the pattern: how often you appear, how the model describes you, where you are positioned and whether those signals move over time. Then use recurring citations and disclosed retrieval searches to investigate the movement. That is the logic behind formal AI visibility tracking: measure the distribution, never the coincidence.
What actually changes whether AI names you.
There is no phrase or schema tag that guarantees a recommendation. The practical job is to make accurate evidence easy to retrieve and easy to corroborate. Four areas deserve attention:
- Reviews — volume and sentiment. A steady flow of recent, positive, specific reviews is the single strongest signal that you're a safe recommendation. A med spa with two hundred warm reviews will be named over one with nine, almost regardless of anything else.
- Consistent business information. Your name, address, phone, hours and category should match everywhere they appear. Contradictory details make a model less confident naming you, because it can't tell which "you" is real.
- Clear, useful pages. Service, location and evidence pages should answer the buyer's actual question with concrete details. When your site is surfaced but a rival is cited, compare what the cited page makes easier to verify.
- Genuine community mentions. When real people recommend you in Reddit threads, Facebook groups and forums, that's third-party evidence the model trusts. This is why showing up helpfully in those conversations compounds — it improves both the lead you win today and the recommendation you earn tomorrow.
Your own site and third-party reputation work together. A clear page can enter the retrieval pool; independent reviews and mentions help corroborate it. Healthy Google rankings alone still do not guarantee a spot in an AI answer, so prioritise the source patterns you can observe over generic optimisation checklists.
Run it by hand — then decide.
Do the buyer's-seat audit yourself this week. It costs nothing, takes twenty minutes, and almost everyone learns something uncomfortable and useful. But the audit is only valuable if you keep running it, and that's where hands break down: thirty phrasings across three models, every day, for four services and three locations, is a job no team should do manually. And because answers vary, occasional manual spot-checks can genuinely mislead you.
That's the honest case for a tool — not that you can't do it by hand, but that consistency and evidence capture are hard to sustain. Reputably repeats buyer-style prompts across ChatGPT, Gemini and Claude, records visibility, sentiment, position and competitors, maps cited sources and disclosed fan-out searches, and can connect the result to AI crawler and referral traffic. Its review and community workflows then give the team somewhere practical to act. If you're weighing options, compare the best AI visibility tools; if you run an agency, see how to report this to clients.
Checking AI recommendations, answered.
How do I check if ChatGPT recommends my business?
Open ChatGPT in a fresh, logged-out session and ask it the way a customer would — 'best emergency plumber in Leeds', 'who should I call for a blocked drain near me' — then note whether your business is named and who's named ahead of you. Repeat with five or six different phrasings, then do the same on Gemini and Claude. One answer proves nothing because answers vary by wording and session; a pattern across many phrasings and all three models is the real signal.
Why does ChatGPT name my competitor but not me?
The answer engine may find clearer, more corroborated evidence for the competitor: useful pages that match the question, consistent business details, strong reviews, authoritative citations and genuine community discussion. When retrieval is disclosed, inspect which searches ran and which sources were surfaced or cited rather than guessing from the final answer alone.
How often do AI recommendations change?
Often enough that a one-time check is unreliable. Models get updated, the sources they draw on shift, and competitors do reputation work that moves their standing. A practice named this month might not be next month. That's why checking is a habit, not a one-off — you're tracking a direction of travel, not taking a single photograph.
What actually moves whether AI recommends me?
There is no guaranteed prompt trick. Improve the evidence an answer engine can retrieve and corroborate: clear service and location pages, consistent business details, strong recent reviews, authoritative third-party citations and authentic community mentions. Use repeated citation and retrieval patterns to prioritise the gaps you can actually address.
Can I automate checking whether AI recommends me?
Yes. Reputably repeats buyer-style prompts across ChatGPT, Gemini and Claude, tracks visibility, sentiment, position and competitors over time, and records cited sources and disclosed fan-out searches. With site tracking installed, it also separates AI crawler activity from detectable referral visits.
Related guides.
Facts checked 11 August 2026. ChatGPT, Gemini and Claude are trademarks of their respective owners; Reputably is not affiliated with or endorsed by them. AI answers and disclosed retrieval trails vary by prompt, model, session and time.
Someone is asking for a recommendation right now. Make sure it's you.
Start your 3-day free trial and see real conversations about what you sell — in minutes, no card, no accounts to connect.