Type “does ChatGPT recommend my business” into Google and you’ll find a wall of blog posts and a couple of Quora threads that all give the same advice: open ChatGPT, ask it for the best [your service] in [your city], and see if you show up. That test is close to worthless. I run AI visibility audits for local businesses every week, and the single-prompt check produces a wrong answer about half the time, in both directions. Here’s the method I actually use: a mention-rate test across 10–25 buyer prompts, run multiple times each, in a clean session.
Why asking ChatGPT once proves nothing
Three things break the “just ask it once” test.
1. The answers are non-deterministic
Large language models generate answers probabilistically. Ask the identical question five times and you’ll get five differently worded answers, and often a different set of recommended businesses. Similarweb’s analysis of ChatGPT answer consistency makes the point plainly: variability is a property of generative AI as a category, not a ChatGPT quirk. Gemini and Perplexity behave the same way. A business can appear in three out of five runs of the same prompt. If you ran it once and hit a miss, you’d conclude you’re invisible. Run it once and hit a mention, and you’d conclude you’re fine. Both conclusions are built on a coin flip.
2. Personalization and location skew the result
ChatGPT infers your location from your IP and adjusts local answers accordingly. If you’re testing from your clinic’s own building, you’re testing from a location signal your customers don’t share. Test from the neighborhoods your buyers actually live in when you can, or at minimum note the location the model assumed (it will usually tell you if you ask).
3. Memory contamination from your own chats
This is the one almost nobody warns you about. If you use ChatGPT while logged in, it has memory of your previous conversations. You’ve probably mentioned your own business name dozens of times — drafting Instagram captions, writing service descriptions, asking marketing questions. When you then ask “who’s the best med spa in Scottsdale,” the model already knows you own one, and it is measurably more likely to name you. That’s not visibility; that’s the model being polite to you.
The fix: always test logged out, or in a Temporary Chat with memory disabled. Every result gathered in your regular logged-in session is contaminated and should be thrown out.
The mention-rate method
Instead of asking “does ChatGPT recommend me?” (a yes/no question with no stable answer), ask “in what percentage of realistic buyer conversations does my business get named?” That’s a rate, and rates are measurable.
Here’s the setup I use in audits:
- Build a prompt bank of 10–25 buyer prompts. Not one prompt — a spread. Mix discovery prompts (“best lash studio in Plano”), problem prompts (“where can I get a chipped tooth fixed same-day near Frisco”), comparison prompts (“X vs Y, which is better for families”), and qualifier prompts (“affordable,” “open Saturday,” “that takes walk-ins”). My free GEO prompt generator builds this bank for you from your service list and city, so you don’t have to invent 25 prompts from scratch.
- Run each prompt 3–5 times in a fresh logged-out or temporary session each time. Yes, that’s 30–125 total runs. It takes an afternoon. It’s the price of an answer you can trust.
- Score every run on a three-point scale: mentioned (your business is named in the answer), cited (named and linked to your site or profile as a source), or absent. Log which competitors appear too; you’ll want that later.
- Compute two numbers: mention rate = mentioned-or-cited runs ÷ total runs; citation rate = cited runs ÷ total runs.
A basic spreadsheet works fine: one row per run, columns for prompt, engine, date, result (M/C/A), competitors named, and sources cited. If you’d rather not build it manually, my AI visibility auditor walks through the same scoring for clinics, and the AI citation readiness checker scores whether your site is even eligible to be cited in the first place.
Which engines to test (and how they differ)
ChatGPT is the biggest single engine (roughly 900 million weekly active users as of early 2026, per DemandSage’s tracking), but it’s not the only place buyers ask. Consumer behavior is split: Bain survey data shows 56% of consumers still mostly start with a search engine versus 16% who mostly start with a chatbot, which means you need visibility in both Google’s AI layer and the standalone assistants.
| Engine | How it answers local queries | Testing note |
|---|---|---|
| ChatGPT (search ON) | Pulls live web results via Bing’s index plus licensed data; cites sources | This is the default for local queries now; your main test surface |
| ChatGPT (search OFF) | Answers from training data only; reflects your reputation as of the model’s cutoff | Test it separately; it shows your “baked-in” reputation, which changes slowly |
| Perplexity | Always retrieval-based, heavy citation footprint, strong Reddit and review-site bias | Easiest to diagnose because every claim links to a source |
| Gemini | Leans on Google’s own index and Google Business Profiles | Notably, BrightLocal found Gemini was the only major LLM that didn’t directly cite Yelp |
| Google AI Overviews / AI Mode | Sits on top of normal Google results and local pack signals | Test in an incognito window with your target city set |
The engines genuinely disagree with each other. In Foundation’s analysis of 28 million small-business queries from late 2025, Yelp captured 72.5% of citations in its competitive set on Google AI Mode and 62.1% on Perplexity, while Gemini barely touched it. A business can score 60% on Perplexity and 10% on Gemini. That’s normal, and it tells you exactly where your gaps are.
Where AI assistants get local business answers: the citation supply chain
When ChatGPT with search enabled recommends a local business, it’s not consulting some private opinion; it’s synthesizing sources it just retrieved. BrightLocal’s study of ChatGPT’s local search sources found business websites make up 58% of sources, third-party mentions of the business 27%, and directories 15%. The supply chain looks like this:
- Review platforms. Yelp is disproportionately powerful (used as a source in 33% of local searches BrightLocal examined), along with Google reviews surfaced through Google’s own AI layer.
- Directories. Foursquare feeds a large share of ChatGPT’s local listings, and niche directories (legal, dental, medical) dominate in their verticals. Even MapQuest still gets cited.
- Reddit. An SE Ranking study of 129,000 domains found brands with heavy Reddit mention volume averaged 3.9x more ChatGPT citations than brands with minimal Reddit presence. “Best X in [city]” Reddit threads are read constantly by these engines.
- News and roundup posts. Local press features and “best of” listicles are exactly the passage format LLMs love to quote.
- Your own site. Only if crawlers can reach it. If your robots.txt or firewall blocks GPTBot, PerplexityBot, or Google-Extended, you’ve cut yourself out of the largest source category.
This is why I treat AI visibility as a supply-chain problem, not a trick. It’s the core of the generative engine optimization work I do: you can’t persuade the model directly, but you can stock every shelf it shops from.
Interpreting your score
Once you have a mention rate, what’s good? There’s no published industry benchmark for local businesses yet, so these bands come from the clinic audits I ran over the past year (est.):
- 0–10%: invisible. The engines either can’t find you or can’t verify you. Most local businesses I test for the first time land here; BrightLocal’s research suggests only a small minority of local businesses get recommended by AI at all.
- 10–30%: emerging. You appear when the prompt closely matches your strongest listing or review profile, but you lose most head-to-head prompts to competitors.
- 30–60%: competitive. You’re in the consideration set for most relevant prompts. This is a realistic 6–12 month target for an established local business (est.).
- 60%+: dominant. You’re the default answer in your category and city. Rare, and usually backed by a large review lead plus strong third-party coverage.
Citation rate will always run lower than mention rate. Being named without being linked usually means the engine learned about you from a directory or Reddit, not from your site, a signal your own pages aren’t citation-ready.
If you’re invisible: the fix priority order
Don’t fix things in random order. Work down this list:
- Crawler access first. Check robots.txt and your firewall/CDN for blocks on GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, and Google-Extended. This is a five-minute check that undoes everything else if it’s wrong.
- Review mass and freshness. Google and Yelp volume, recency, and reply rate. The engines cross-reference review platforms heavily; a thin or stale profile reads as an unverifiable business.
- Third-party mentions. Get into the directories your vertical’s engines actually cite, pursue local press, and earn honest Reddit presence (real participation; astroturfing gets flagged and backfires).
- Answer-ready pages. Service pages that state who you serve, where, at what price range, with FAQ blocks and schema, the passage format engines can quote directly. This is the on-site half of answer engine optimization.
Notice the order: two of the four biggest levers live entirely off your website. That surprises most owners, and it’s why pure on-page SEO effort often moves AI visibility very little.
Tracking over time
One test is a snapshot. The value compounds when you re-run the same prompt bank monthly:
- Keep the prompt bank frozen so month-over-month numbers are comparable; add new prompts in a separate tab.
- Log competitor mentions every run. Watching a rival’s mention rate climb tells you they’re building citation supply. Check what changed (new reviews, a press feature, a Reddit thread).
- Expect noise. A single month’s swing of ±10 points on a small run count is sampling variance, not a trend (est.). Three consecutive months moving the same direction is a trend.
- Re-test after model updates. When OpenAI or Google ships a major model or changes retrieval behavior, historical comparisons get shakier; note the date and keep going.
If you run a med spa or aesthetics clinic, I’ve written a companion piece that applies all of this to the specifics of aesthetic-treatment queries: how med spas show up in ChatGPT. And since roughly half of consumers now ask AI assistants for local business recommendations per BrightLocal’s Local Consumer Review Survey, this is no longer a curiosity metric — it’s a channel. Every AI recommendation you’re absent from is a call that goes to a competitor instead; my missed call calculator puts a dollar figure on exactly that kind of leak.
Want a second set of eyes on this for your clinic? Book a free strategy call or call/text me at +91 97297 12388.
Frequently asked questions
Can I pay OpenAI or Google to have ChatGPT recommend my business?
No. There is no paid placement inside ChatGPT’s organic answers, Perplexity’s answers, or Google AI Overviews’ recommendations as of mid-2026. Anyone selling “guaranteed ChatGPT rankings” is selling something they don’t control. What you can influence is the source material the engines retrieve: reviews, directories, mentions, and your own site.
How long does it take to go from invisible to mentioned?
Fresh content and new citations can enter retrieval-based answers within days to weeks, since ChatGPT search and Perplexity pull from live indexes. Building a consistent mention rate across many prompts typically takes two to four months of sustained work on reviews and third-party mentions (est.), and shifting the search-off “baked-in” answers takes until the next model training cycle.
Why does ChatGPT recommend me when I ask, but my customers say it doesn’t?
Almost certainly memory contamination. Your logged-in account knows you own the business from past conversations. Re-test logged out or in a Temporary Chat; that’s the answer your customers actually see.
Do I need to test all four engines every month?
Prioritize by where your buyers are: ChatGPT with search on and Google AI Overviews cover the majority of AI-assisted local discovery. I’d test those two monthly and Perplexity and Gemini quarterly, unless your audit shows a big gap on one of them worth tracking closely.
Is a 25% mention rate bad?
For a first test, it’s better than most. In my audit work, the majority of local businesses testing for the first time score under 10% (est.). At 25% you’re emerging; the priority is usually review velocity and third-party mentions rather than site changes.
Will blocking AI crawlers protect my content without hurting me?
It will remove you from the source pool. If GPTBot and PerplexityBot can’t read your site, the engines can only describe you through third parties — or not at all. For a local business trying to be recommended, blocking AI crawlers works directly against the goal.
