All guides · Reviewed 2026-10-18

How to Check If ChatGPT Recommends Your Business (and Why the Answer Keeps Changing)

You type "best dentist in Charlotte" into ChatGPT, read the list, and your practice isn't on it. Or it is, and the next time you ask, it isn't. Both experiences are normal, and both show why a one-off check misleads.

This guide gives you a free, repeatable method you can run in about an hour. It works for ChatGPT, Perplexity, Gemini, and Google's AI Mode. It also explains how to read the results so you fix the right things.

The short version: Write 5–10 questions your customers really ask, with your city in each one. Ask each in a fresh chat on each AI engine, at least three times. Record who gets named and which sources are cited. Your score is the share of runs that name you, not whether one answer did.

Why one answer isn't enough

AI answers are generated fresh each time, and several things change them:

Randomness. The same question can produce a different list on the next run. In a small study of 10 local searches run three times each in ChatGPT, only 40% of the businesses recommended appeared in all three runs, and one search "changed almost completely every time" (Sonick Marketing).
Location. ChatGPT may estimate your location from your IP address. Device location sharing is optional and off by default. OpenAI recommends including your city, neighborhood, or postal code in the question to get the right area (OpenAI Help Center). If you test from your office, "near me" questions are skewed toward you.
Memory and personalization. If memory is on, ChatGPT may use saved details when it rewrites a search query (OpenAI Help Center). If you have asked about your own business before, your results may not match a stranger's.
Search vs no search. An answer without web sources comes from the model's built-in knowledge and can be out of date. For local questions, you want to test answers that used web search, which show citations or a Sources panel.
Model and product changes. Engines update often, so last quarter's result may not hold today.

That is also why people distrust some tools. One user on X described cancelling a $125/month AI visibility tool after getting "3 different scores" for the same site within an hour (X). Different scores aren't a bug in that sense, because the answers really do vary. The fix is to sample several times and report a rate.

Step 1: Write the questions your customers actually ask

Start with 5–10 questions. Cover a few types:

Type	Example (swap in your trade and city)
Best-of	"Who is the best family dentist in Charlotte, NC?"
Specific service	"Which Charlotte dentist does same-day crowns?"
Constraint or situation	"Recommend a dentist in Charlotte that's good with anxious patients and takes Delta Dental."
Neighborhood	"Best dentist in South End, Charlotte?"
Comparison (branded)	"Compare [Your Practice] and [Competitor] for dental implants in Charlotte."
Facts check (branded)	"What are the hours and services of [Your Practice] in Charlotte?"

Rules:

Put the city (and state) in every question. Don't rely on "near me."
Don't use your business name in the discovery questions (the first four types). You're testing whether the AI finds you unprompted.
Write the questions down and never change the wording, so future tests are comparable.

Step 2: Set up a clean test

ChatGPT: use a temporary chat or a logged-out window, so memory and past chats don't color the answer. Confirm that the answer used web search: look for inline citations or a Sources button. If it didn't search, pick Search from the tools menu or type / and choose Search, then ask again (OpenAI Help Center).
Perplexity: open a new thread for each question.
Gemini: open a new chat. Gemini Apps can draw on public Google Maps information for place questions (Gemini Apps Help).
Google AI Mode / AI Overviews: search in a private window. AI Overviews only appear when Google decides they add value, so many local searches won't show one (Google Search Central).

If you can, also run a set from a phone on mobile data. It gives you a second location signal.

Step 3: Run each question three or more times and record everything

Use a simple sheet with one row per run:

Date	Engine	Question	Run #	Named? (Y/N)	Position	Businesses named	What it said about you	Sources cited

Run each question at least 3 times per engine, ideally spread over a few days.
Copy the businesses named exactly as written.
Copy the sources cited (domain and URL). They are the most useful column in the sheet.
Note any wrong facts about you: old hours, a service you don't offer, a wrong area.

Ten questions × three runs × three engines is 90 answers. That sounds like a lot, but it takes about an hour by hand, and it gives you a baseline you can trust.

Step 4: Read the results

Your recommendation share

Count the runs that name you and divide by total runs, per engine and overall. For example:

ChatGPT: named in 4 of 30 runs (13%)
Perplexity: 9 of 30 (30%)
Gemini: 0 of 30 (0%)

Compare engines against each other and questions against each other. Being strong for "best dentist" but missing from "same-day crowns" points to a content gap on your site. Being strong in Perplexity but absent from Gemini points to your Google Business Profile.

Who wins instead

List the businesses that appear most often. Then look at what they have that you don't: review count and recency on multiple platforms, specific service pages, mentions in local "best of" lists, trade directory profiles.

Which sources the engines trust

Group the cited sources:

Your own site: good. Note which pages get cited.
Directories and review sites (Yelp, trade directories, BBB, and so on): check that you're listed and your details match.
Editorial lists and local media: these are your outreach targets.
Competitors' sites: see what those pages say that yours doesn't.

BrightLocal's earlier study of ChatGPT's local sources found business websites, "business mentions" (publishers, guides, Wikipedia), and directories in that order. Directories and mentions made up a bigger share for "best ___" questions (BrightLocal). Your own data for your city and trade beats any general study.

Accuracy

If an engine describes you wrongly, follow its citation to the source and fix it there, whether that's your website, a listing, or a directory. Then re-test.

Why your business might not be showing up

If you're rarely or never named, these are the usual causes, roughly in order of how easy they are to check:

Crawlers are blocked. ChatGPT search requires that OAI-SearchBot can crawl your site. Sites opted out "will not be shown in ChatGPT search answers" (OpenAI). Perplexity uses PerplexityBot (Perplexity). Check robots.txt and your CDN or firewall settings.
Your listings disagree. Different names, phones, hours, or categories across Google, Bing Places, Yelp, and Apple make it harder to match you.
Missing from the platforms AI uses. For example, Yelp content is licensed to ChatGPT (Axios), and ChatGPT's search partners include Microsoft (OpenAI Help), so an unclaimed Yelp or Bing listing is a gap.
Thin reviews outside Google. BrightLocal's CEO warned that LLMs "can't see inside Google's walled garden of reviews" (BrightLocal).
No page that answers the question. If you offer same-day crowns but no page says so, it's hard to cite you for it.
No independent mentions. Competitors appear in local lists, news, and association directories, and you don't.
It's just variance. If you're named in 2 of 30 runs, you're on the edge. The work above is what moves you from occasionally to consistently.

Our full playbook, How to Get Your Local Business Recommended by ChatGPT, covers each fix step by step.

How often should you re-check?

After a fix: wait for re-crawling. OpenAI notes robots.txt changes take about 24 hours to be reflected in search (OpenAI). Listing and review changes take longer. Re-test in 2–4 weeks.
Ongoing: weekly or every other week, with the same questions, so you can see a trend instead of noise.
When a competitor jumps in: look at the sources they're cited from. That's often a new list, article, or directory you can get onto too.

Free tools vs doing it by hand

There are free one-off checkers from local-focused startups, and Semrush and Ahrefs each offer a free brand-level AI visibility checker (Semrush, Ahrefs). Before you trust any score, ask four questions:

Does it show the actual answers, or only a number?
How many times did it run each question? One run is a sample of one.
Which engines and settings did it use? API results can differ from what consumers see in the apps, and some vendors say so openly (MentionedOn).
Does it use local, city-specific questions? Brand tools often track national prompts, not "best plumber in Tulsa."

FAQ

Is there a way to see exactly what my customers see? No. Answers depend on each person's location, settings, history, and chance. A good test samples many runs with fixed questions and reports a rate.

Should I test logged in or logged out? Logged out, or in a temporary chat, for a clean baseline. Your logged-in account may carry memory and past conversations that change results.

My business shows up when I ask by name. Does that count? That tests whether the AI knows you, which is useful for checking accuracy. Customers who don't know you yet ask unbranded questions, so measure those separately.

ChatGPT recommends me but Gemini doesn't. Why? They use different data. Gemini can draw on Google Maps information (Gemini Apps Help), so start with your Google Business Profile. ChatGPT relies on web search partners and licensed data such as Yelp.

Skip the spreadsheet

RecommendRadar runs this method for you. It asks your customers' local questions on ChatGPT with web search, Perplexity, and Gemini: 10 editable questions per location, each run twice per engine (16 questions run 3 times on the Agency plan). It reports the share of answers that name you and who's named instead, with the raw answers on hand so you can verify every result. A Monday email lists the sources AI cited and a fix list. It costs $19/month for one location, or $79/month for 10 locations (pricing). The free check asks 3 buyer questions on Perplexity with no account. Run the free check →

Sources: OpenAI Help Center (ChatGPT search, location, memory); OpenAI crawler documentation; Perplexity crawler documentation; Gemini Apps Help; Google Search Central "AI features and your website"; Sonick Marketing ChatGPT local business study; BrightLocal "Uncovering ChatGPT Search Sources" (Dec 2024) and LCRS 2026 AI report; Axios (Yelp × OpenAI, Jul 23, 2026); X post by @Jackrel67; Semrush, Ahrefs, and MentionedOn tool pages. All links accessed Oct 18, 2026.