How to check if ChatGPT recommends your business (and what to fix if it doesn’t)

To check if ChatGPT recommends your business, open a Temporary Chat with personalization off, ask the questions a new customer would type (“best emergency plumber in Mesa”, not your business name), confirm the answer shows web sources, and repeat each question three times. Score every answer for four things: were you named, linked, recommended, and described accurately. Do it monthly with the same questions so you’re comparing like with like.

There is no ChatGPT equivalent of Google Search Console. OpenAI doesn’t publish a ranking report or tell you how often your business shows up. So the only honest way to know is to sample the answers yourself, carefully enough that the results mean something. This guide gives you the exact setup, a prompt set you can copy, a scoring sheet, and a table that turns what you see into a fix.

Why a quick “ask ChatGPT about me” check misleads you

Most owners type their own business name into ChatGPT, see a decent description, and assume they’re covered. That only proves ChatGPT can find you once you’ve named yourself. Customers don’t search that way. They ask for a type of business in a place, and that is the answer you need to be in.

Three other things skew a casual check:

• Your own history. If memory is on, ChatGPT may lean on what it knows about you from past chats. OpenAI’s Temporary Chat FAQ says an unpersonalized temporary chat doesn’t use memory or custom instructions, which is what you want for a clean test.
• No web search. ChatGPT doesn’t search the web for every reply. An answer with no sources listed came from the model’s training data, which can be months or years old. It tells you nothing about whether your current website and listings are being picked up.
• One run. Ask the same question twice and you’ll often get a different list. A single answer is an anecdote, not a result.

How do you check if ChatGPT recommends your business?

Run a fixed set of six to ten customer-style questions in a clean, unpersonalized chat, make sure each answer used web search, and repeat every question three times. Record whether your business was named, linked, recommended and described correctly. The pattern across all runs, not any single answer, is your result.

1. Start clean. In ChatGPT, start a Temporary Chat and choose the unpersonalized option before your first message (OpenAI says you can’t switch it after the chat begins). A logged-out browser window works too. Close the chat and open a fresh one for every run.
2. Name the place in every prompt. Answers can shift with where you’re asking from, so don’t rely on it guessing. Use the city or neighborhood your customers actually type.
3. Ask the question, then check for sources. Look for the source links or citation chips under the answer. If there are none, follow up with “search the web and show your sources” and score that answer instead.
4. Repeat three times. Same prompt, three fresh chats. Three runs is enough to tell “always there” from “sometimes” from “never” without eating your afternoon.
5. Log it immediately. Write down the date, prompt, run number, the businesses listed in order, and the sources cited. Screenshots help when you want to show a web designer or listing provider what went wrong.
6. Finish with branded prompts. Only after the customer-style questions, ask about your business by name to check the facts it has on you.

The whole test takes about 20 to 30 minutes for one location. Repeat it in Perplexity and in Google’s AI Mode if you want the wider picture. The method is the same; only the sources each tool reads differ.

The prompt set: what to ask (and what not to)

Build your list from four kinds of question. Each one catches a different way customers ask, and a business can show up for one and vanish for another. Swap in your own service and area.

Two rules keep the test honest. First, never put your business name in the customer-style prompts; that turns a recommendation test into a lookup. Second, freeze the list. Save it in a note and use the exact same wording next month, or you won’t know whether a change came from your work or from rewording the question.

How to score the answers

Score each answer on four separate questions, because they fail for different reasons. A business can be mentioned without being linked, and linked without being recommended. Lumping them into one “yes” hides the actual problem.

• NamedYour business appears anywhere in the answer, even in passing.
• LinkedYour own website, or your profile on a listing site, is among the cited sources.
• RecommendedYou’re in the main shortlist, not tucked into a “you could also try” line at the end.
• AccurateName, area, services, hours and phone number are all correct. One wrong detail is a fail.

Then add it up. If you ran eight prompts three times, that’s 24 answers. “Named in 9 of 24” is a number you can track month to month. Do the same for your two or three closest competitors in the same answers; it costs nothing extra and tells you whether you’re losing ground or everyone is equally invisible.

Pay most attention to the cited sources. They show you which websites ChatGPT read to build the answer, and they are the closest thing you’ll get to a map of where to improve.

What your results mean and what to fix

Match the pattern you saw to a row below. Fix the top problem first; there’s no point chasing more mentions while ChatGPT is giving people your old phone number.

Check you haven’t blocked ChatGPT by accident

Visit yourdomain.com/robots.txt in a browser. OpenAI’s crawler documentation says sites that block OAI-SearchBot “will not be shown in ChatGPT search answers”, though they may still appear as a plain link. If you see User-agent: OAI-SearchBot followed by Disallow: /, or a blanket rule blocking all bots, that’s your problem. Blocking GPTBot is different: it only covers training data, not search. OpenAI says robots.txt changes take about 24 hours to register. Some security plugins and hosting firewalls add these blocks without telling you, so check even if you never touched the file.

Where ChatGPT’s local answers tend to come from

OpenAI doesn’t publish a list. The best public evidence is a BrightLocal study of 800 local searches run in November 2024, which found business websites made up 58% of the sources ChatGPT cited, with mentions on other sites and directories making up the rest. That data is getting old and the product has changed since, so treat it as a direction, not a rule. The practical takeaway still holds: your own website is the source you control most, and it has to say plainly who you are, what you do and where. Listings matter too, so keep your Bing Places profile as current as your Google one.

Does checking ChatGPT tell you anything about Google’s AI answers?

Only partly. Google’s AI Overviews and AI Mode are built on Google’s own index and Business Profile data, while ChatGPT draws on its own search and sources. A business can do well in one and be missing from the other, so test both if Google is where most of your customers come from.

Google is unusually direct about this. Its AI features documentation says there are no extra requirements or special optimizations needed: a page has to be indexed and eligible to show with a snippet in normal Search. In other words, the work that helps you in Google’s map pack and organic results is the same work that helps in its AI answers. If you’re weak there, start with our local SEO basics before anything labeled “AI optimization”.

The overlap with ChatGPT is your own content. Clear service pages, accurate structured data and consistent details help every tool that reads your site. If you serve customers at their location, our guide to service area business schema shows how to mark that up without exposing a home address.

Mistakes that waste your time

• Testing once and panicking. One missing answer means little. Three runs across eight prompts means a lot more.
• Buying a tracking tool before you’ve tested by hand. Paid AI-visibility tools save time across many locations, but for one shop a monthly manual test is enough to see the trend.
• Rewriting your prompts every month. New wording, new results. Keep the list frozen and add new prompts as extras, not replacements.
• Chasing fake signals. Planting reviews, posting fake “best of” lists or spamming forums can backfire, and Google’s review policies ban incentivized and fake reviews outright. Real reviews, replied to properly, are worth more; see how we handle replying to a 1-star review with no comment.
• Ignoring the wrong facts. Being named with the wrong phone number is worse than not being named. Accuracy fixes come first.

Where to start this week

Block 30 minutes. Write down eight customer-style questions using the table above, run each three times in clean Temporary Chats, and score every answer as named, linked, recommended and accurate. Check your robots.txt while you’re at it. Then pick the single biggest gap from the results table and fix only that before next month’s test.

If the cited sources keep pointing to pages you don’t control, or your own site isn’t clear about what you do, that’s usually a content problem rather than an AI problem. Our content marketing work starts there. Or ask us for a free visibility audit and we’ll run the test across ChatGPT and Google’s AI answers for you and show you what to fix first.