How to Check If a Brand Appears in ChatGPT (Manual Method + Free Tool)

If you want to know whether your brand shows up in ChatGPT, the instinct is to open ChatGPT and ask it. That single question will give you an answer, and the answer will be close to useless.

Here's why, and here's the protocol we actually run when we audit a business's AI-search visibility.

Why "just ask ChatGPT" gives you a false reading

A single prompt tests one phrasing, one session, one model, on one day. AI answer engines don't return a stable, canonical answer the way a database lookup does. They synthesize a response from whatever the model retrieves for that specific wording, and that response can change based on:

• Phrasing. A narrow, geo-qualified question and a broad, category-wide question pull from different sources, even when they're about the same buying decision.
• Time. The same question asked two weeks apart can return a different shortlist, because the underlying web content the model draws on has changed, or the model's own retrieval has.
• Web-search toggle. A model answering from training data alone and the same model with live web search turned on can name completely different companies.
• Which engine. ChatGPT, Gemini, Claude, and Perplexity don't share a source list.

We see all four of these in the audits we run. One case makes the phrasing problem concrete: a junk removal business we audited was named second of five companies, with its own service page cited as the source, on "best junk removal service San Fernando Valley." Ask a broader version of basically the same question, "junk removal Los Angeles same day," and it doesn't appear anywhere. The answer instead goes to national franchises and LA-wide platforms. Same business, same site, same week. The only thing that changed was how the question was worded. Overall the business scored 60 out of 100 on visibility across the three questions we tested it against.

A single AI-search prompt would have caught one of those two outcomes and told you nothing about the other.

The 15-minute manual protocol

Instead of one prompt, run a small matrix. You need three axes:

1. Query shape. For each core service or product, write three to five phrasings that a real buyer would actually type or say, covering:

• Category: "best [thing] in [city]"
• Comparison: "[your brand] vs [competitor]"
• Recommendation: "best [thing] for [specific use case]"
• Troubleshooting: "[thing] not working" or "how to fix [problem your product solves]"

2. Engine. Run the same phrasings in ChatGPT, and ideally also Gemini, Claude, and Perplexity (more on the differences below).

3. Web-search state. If the engine lets you toggle live web search, run the query both ways. A model answering purely from training data can behave very differently than the same model with search turned on.

That's a minimum of twelve data points per service line (three phrasings x four engines), more if you're including the on/off web-search variants. It takes about 15 minutes once you have the question list written down. Write down, for each answer: was your brand named, at what position, and what URL (if any) got cited.

Two things make the results more reliable. Run every query logged out or in a private/incognito window. A logged-in session carries your search and chat history, and a model that already knows you use your own product will sometimes surface it for reasons that have nothing to do with what a new prospect would see. And keep the raw answer text, not just a yes/no note. If your brand does show up, the exact wording tells you what the model believes about you, which is useful even when it's flattering, and especially useful when it isn't.

Reading the result: mentioned vs. cited vs. recommended

Once you have answers in hand, don't just mark them present or absent. There are at least three different outcomes, and they matter for different reasons:

• Mentioned. Your brand name appears in the answer, but there's no source link tied to it, or the citation goes to a directory or review site instead of your own domain. You exist in the model's world, but you aren't the source of information about yourself.
• Cited. The answer names your brand and links to a page on your own site as the source. This is the strongest outcome, because it means the model is treating your content as the evidence.
• Recommended vs. merely listed. Some answers rank or single out a top pick; others just enumerate a set of names with no ordering. Being third in a ranked "best X" answer is a different result than being one of nine names in an unordered list.

The same business can land in different buckets on different queries. That junk removal company was cited, not just mentioned, with its own page as the source, on "junk removal and demolition Ventura County," described in terms that matched its actual homepage copy. On the broad LA-wide query it was absent entirely. Uneven results across queries are the normal case, not the exception. A single overall "are we in ChatGPT, yes or no" question can't capture that.

It's also possible to score zero across the board. One business we audited, an online bank, tested at 0 out of 100 on visibility: absent from all three of the commercial questions we ran against it, even though its own products were priced competitively with several companies that did surface. Zero doesn't always mean something is broken technically. Its own branded FAQ content was reaching at least one provider fine; the gap was specifically on unbranded, comparison-style questions, where competitors with heavier citation footprints crowded it out.

The unevenness can also run the other direction, better on some queries than a business's own prior check. A cereal brand we track scored 68 out of 100 overall, but the three questions we tested it against told three different stories in the same audit pass. On a low-sugar query, it went from a brief, buried mention to the first-listed, cited answer, the strongest single-query result of the three. On a keto-specific query it stayed cited, but slipped from the top-named brand to second, behind Schoolyard Snacks. On a broader high-protein query it dropped further, out of the main list entirely and into a secondary "other notable mentions" tier, with its own domain no longer cited at all. Three queries, three different trajectories, one brand, one week. That is exactly the kind of result a single ChatGPT prompt cannot show you, because it only samples one of those three questions at a time.

Doing the same for Gemini, Claude, and Perplexity

The protocol is identical, but the results won't match, and that's useful information on its own.

Perplexity tends to lean harder on its live web index and shows its source list more explicitly in the UI, so it's often the easiest engine to audit by hand: you can see exactly which pages it drew from for a given answer.

Claude with web search turned on runs its own retrieval and links its sources, but don't expect that source list to match anyone else's. Across the four engines in our own audits, nearly nine out of ten of the individual pages cited were cited by only one engine.

Gemini grounds its answers in Google's own search index and ranking signals, which means a page's traditional SEO strength has more pull here than it does with ChatGPT or Perplexity. We've seen this gap directly: a business whose site is a client-rendered React application, serving an empty page body until JavaScript executes, scored 35 out of 100 on visibility and 33 out of 100 on a separate technical readiness check. Its raw HTML carried nothing but a title tag and meta description, no schema markup, no crawlable content. A crawler that doesn't execute JavaScript, which describes most AI-answer bots, sees almost nothing on that site regardless of how good the actual page content is.

Run the same query set across all four engines and you'll usually find your brand does better on one than the others. That's a signal about why, not just whether.

What the pattern of absence tells you

If you run this matrix and come back with a clean sheet of "absent" across most queries, the pattern of which queries you're absent from is the diagnostic. Broad, high-competition category queries going to national brands is a different problem than a technical crawlability failure, which is different again from having no third-party sources that mention you at all. Matching the pattern to the cause is what determines whether the fix is content, technical, or off-site authority work.

Automating this across every client

The manual protocol works for checking one brand once. It gets tedious fast once you're doing it for more than a handful of clients, or want to re-check monthly instead of once.

That's the entire job of the AEO audit we run: the same query-matrix method above, executed against live AI-search results for a real, current set of buyer questions, with every mention, citation, and absence logged with its source URL, plus the technical readiness checks (schema markup, crawlable content, entity clarity) scored separately from what the search results actually returned. It's the same protocol, just run at the scale an agency managing more than one or two clients actually needs.

FAQ

Does ChatGPT know my website? Almost certainly, if your site has been publicly crawlable for any length of time. Whether it surfaces your site in a given answer is a separate question from whether it has seen your site at all, and that's the distinction this whole protocol is built to test.

How often do AI answers change? Often enough that a single check isn't a permanent record. We re-audited a tactical holster manufacturer on the same three questions about a week after its first check. Two of the three questions returned the same result as before, still absent. The third, "best tactical holster brands," had flipped: the brand went from completely absent to directly named, the fifth of seven brands listed, described accurately by its actual product line. Nothing on the client's own site had changed in that window. The answer itself had simply changed. Treat any AI-visibility check as a snapshot, not a settled fact, and re-run it periodically, especially before deciding a gap is permanent.

Is being "mentioned" enough, or do I need to be "cited"? It depends on what you're trying to get out of it. A mention with no source link still puts your name in front of someone evaluating options. A citation, where the engine links to your own page, is the stronger and more durable outcome, because it means your content, not someone else's summary of you, is the thing the model is drawing from.

Want to see this full query matrix run against your own domain? Run a free AEO audit and get the complete breakdown, engine by engine, query by query. If you want to understand the methodology in more depth first, see how our AEO audit works.