How to Check Whether Your Business Appears in ChatGPT, Gemini and Perplexity

The first thing most business owners do when they hear about AI search is type their company name into ChatGPT. It is a natural instinct and it tells you almost nothing. A brand search checks whether the system has heard of you. The question that matters is whether it names you when someone who has never heard of you asks for help.

This is the audit I run before any AI visibility work, adapted so you can do it yourself in an afternoon. It is deliberately low-tech, because the tools sold for this are useful but they are no substitute for understanding what you are measuring.

Why does a single check tell me nothing?

Because AI answers are probabilistic, so the same question can return different businesses on different runs. Google says AI Overviews and AI Mode may use different models and techniques, which means the responses and links they show will vary. The same applies to every assistant. One run is one sample from a distribution.

That has two consequences. A single appearance does not mean you are reliably visible, and a single absence does not mean you are invisible. What you are after is a rate: out of a set of prompts run several times, how often do you appear, and where? I treat anything else as an anecdote.

Which prompts should I use?

Use the questions a customer who does not know you would ask, phrased the way people actually speak. Do not use your brand name. Write ten to fifteen prompts across the stages of a purchase.

Write the prompts down once and reuse them exactly. Changing the wording between checks makes any comparison meaningless. Include some that name your area and some that do not, and some phrased as a question and some as a plain request.

How do I run the check fairly?

1. Use a clean session. Log out, or use a private window, so earlier conversations and saved preferences do not shape the answer.
2. Set the location honestly. Local answers depend on where the system thinks you are, so run from the area you serve, or state the place in the prompt.
3. Use UK English. British spellings and place names affect what is retrieved.
4. Run each prompt three times, on different days if you can. Record every run, not the best one.
5. Capture the answer and the sources. Screenshots or pasted text, with the date and the platform.
6. Check each platform separately. They draw on different sources, so a result on one says little about another.

What should I record?

Record four things per run: whether you were named, where in the answer, how you were described, and which sources were cited. A simple spreadsheet is enough.

The sources column is the most valuable and the most often skipped. It tells you which websites are shaping the answer, and those are the places where being accurately described matters. Perplexity shows its sources alongside every answer, which LOCALiQ's UK guide notes makes it the most straightforward platform to audit, so it is a good place to learn what the pattern looks like.

How do the platforms differ?

They draw on different sources, so check them separately. SOCi's 2026 study described Gemini as grounded in Google Maps and found its business profile information 100% accurate in the sample, against about 68% for ChatGPT and Perplexity. That fits what you would expect: Gemini can read the Google Business Profile directly.

• Gemini and Google AI features: lean on your Google Business Profile and the Search index. Profile accuracy matters most.
• ChatGPT: depends on OpenAI's search crawler being allowed, and is widely reported to use Bing, which OpenAI's own crawler page does not state. Check your Bing presence as a precaution.
• Perplexity: cites sources visibly, so you can see exactly which pages it relied on. Its crawler is documented as being for search, not training.
• Google AI Overviews and AI Mode: measurable through Search Console, as below.

What does Search Console tell me?

Google's Generative AI performance report in Search Console shows how people discover your content through generative AI features on Search and Discover. Google's own guide points site owners to it as the way to measure visibility in AI Overviews and AI Mode, and says appearances in those features are also included in the overall Web search data.

This is the only measurement source that reflects what Google actually served, rather than what a tool sampled. Use it alongside your manual prompts, not instead of them. Look at the trend over months and ignore single-week swings, because those are mostly noise.

What can AI visibility tools do for me?

They automate the repetition, which is genuinely useful, but they sample a variable system and cannot tell you why an answer changed. Tools run your prompts on a schedule across several assistants and record mentions, citations and share of voice. For tracking a rate over time, that saves a great deal of effort.

• Good for: tracking your appearance rate, citation sources and competitors over months.
• Not good for: explaining a result, proving that an action caused a change, or telling you what a real customer saw.
• Worth checking: how the tool sets location, how often it re-runs prompts, and whether it shows raw answers so you can verify them.
• Cost versus value: for a single-location UK business, the manual method above is usually enough for the first quarter.

How do I interpret the results?

Read the pattern, not the headline. If you appear in discovery prompts about half the time, you are doing something right. If you appear nowhere and competitors with weaker websites do, the sources column usually shows why: they are cited in directories or articles you are absent from.

Then re-run the same prompts after a month of changes. Because of the variation, I would want to see a shift across several prompts and several runs before calling it real. That patience is the difference between measuring and seeing patterns in noise.

What if an assistant describes my business wrongly?

Find where the wrong fact came from and fix it there, rather than arguing with the assistant. A wrong address, an old service or a closed branch usually traces to a source the system read: an old directory listing, a previous website, a third-party article or a profile that was never updated. The sources column in your spreadsheet is how you find it.

1. Note the wrong claim exactly, with the platform and date.
2. Check the cited sources for the same claim, and search the claim itself on Google to find where it appears.
3. Correct the source if you control it, or ask the owner to correct it if you do not.
4. Update your own pages and profile so the right version is the most prominent one.
5. Use the assistant's feedback option if it has one, as a supplement and not a fix.
6. Re-run the prompt after a few weeks, remembering that answers can lag and vary.

How often should I repeat the audit?

Quarterly is enough for most UK local businesses, with a monthly look at Search Console. The systems change often, but your own footprint changes slowly, and checking weekly mostly measures randomness. Do a full run before you change anything so you have a baseline, one after the first round of fixes, and then settle into a rhythm.

If you would rather have somebody run it and tell you what the sources column means, that is what my AI search visibility service begins with. The method above is exactly what I use, and there is no reason you cannot do the first pass yourself.