AI Visibility Checker

Check visibility across ChatGPT, Perplexity, and Google AI.

AI Visibility Checker that shows you the exact answer

Enter your category and your town. We put three questions a real customer would ask to an AI, and hand back the answer word for word, with the prompt we used and the timestamp. You see whether you were named, merely described, or absent — and which competitors got named instead. No email gate, because results you cannot show your boss are not results.

Full answer, quoted verbatimPrompt and timestamp shownThree buyer-intent questionsFree, no email required

Built and reviewed by Sayed Hasan, founder of SEOs Hut · Updated 28 September 2026 · Free, no signup

How do I check if ChatGPT mentions my brand?

Ask it directly, then ask again. Use the question a customer would type — your category plus your city — and read whether the answer names you, describes you without naming you, or omits you. Then run the same prompt at least five times, because LLM answers are not deterministic. One run is an anecdote; five dated runs are evidence.

One answer is not a ranking. It is a single sample.

The instinct after seeing four competitors and not yourself is to treat that answer as a readout of a fixed list. It is not. Anthropic states in its API documentation that even at a temperature of 0.0 "the results will not be fully deterministic". Google's Vertex AI documentation says the same, and adds that models default to a temperature of 1.0 — the setting that produces the most variation, and the one consumer chat products ship with. Google also states that AI Mode and AI Overviews "may use different models and techniques, so the set of responses and links they show will vary." So: run it again. Then three more times. The pattern is the finding; the single answer is a screenshot.

What the tool hands back

A real transcript, annotated — this is the whole product

Below is an illustrative run for a fictional business, Marston & Hale Roofing in Sheffield. It is written to show the shape and the annotations of a real result, not to report real data about any real company. Your own run returns your own answers, verbatim.

1Engine: ChatGPT · Logged out, memory off · 27 September 2026, 14:06 BST

2Prompt 1 — "Who are the best roofers in Sheffield?"

3"Here are some well-regarded roofing companies in Sheffield: Peak Roofing Services, Sheffield Roofing Co, A. Hartley & Sons and Rivelin Valley Roofing. Reviewers on Checkatrade and Google rate these highly, particularly for flat-roof work."

4Prompt 2 — "Who repairs flat roofs in Sheffield with a written guarantee?"

5"Several Sheffield firms offer written guarantees on flat roofing, including Peak Roofing Services and A. Hartley & Sons. Marston & Hale Roofing also advertises flat-roof work, though I don't have details of their guarantee terms."

6Prompt 3 — "Is Marston & Hale Roofing in Sheffield any good?"

7"I don't have enough reliable information about Marston & Hale Roofing to assess them. I'd suggest checking their Google Business Profile and Checkatrade listing for recent reviews."

8Re-run, 14:31 BST, same three prompts: named in Prompt 2 on three of five runs. Not named in Prompt 1 on any of the five.

1The header is the part every other tool omits. Engine, session state and timestamp are what make the result re-testable by somebody who does not trust you.

3Absent from the broadest, highest-intent question — and the answer tells you why it named the others: it leaned on Checkatrade and Google reviews. Those are third-party sources, not the business's own site. When you are missing from the sources, you are missing from the answer.

5Named, but described as merely "advertising" the service, with the guarantee unknown. That is a content gap with a precise shape: the guarantee terms are not stated anywhere an engine could lift them.

7The most diagnostic line in the transcript. This is not "you rank badly" — it is an entity resolution failure. The engine cannot assemble a confident picture of the business from anything it has. Our Entity Checker exists for exactly this sentence, because engines cannot cite a brand they cannot resolve to a single identity.

8Three of five. This is why we show the count. If we had run it once and got the miss, we would have told you that you were invisible for a question you are in fact named in most of the time.

Before you buy anything

Google's own answer to "what do I optimise for?"

“There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.”
<b>Google Search Central, AI features and your website</b> — Google Search Central

Reading the result

How many runs before "not mentioned" means anything

There is no published sample-size standard for this, so here is our reasoning and the evidence behind it, and you can argue with both. The short version: five runs of the identical prompt, and count.

Runs of the same prompt before you draw a conclusion

Good5 or more, identical wording, same day

Needs work3

Poor1

Thinking Machines Lab sampled 1,000 completions at temperature 0 from one model on one prompt and got 80 unique completions. All 1,000 were identical for the first 102 tokens and first diverged at token 103 — meaning the opening of an answer is stable and the list of names further down is exactly where the variance lives.

Runs in which you are named

Good4–5 of 5

Needs work1–3 of 5

Poor0 of 5

Zero of five is the only result that justifies the word "invisible". One to three of five is the interesting case: you exist in the model's candidate set but are not a default, which is a share-of-voice problem rather than an absence problem.

Named versus described

GoodNamed, spelled correctly, with your city

Needs workNamed but conflated with another business

PoorDescribed generically, no name

"A local firm offers this" and "Marston & Hale offers this" are different failures. The first is usually a source problem; the second is usually an entity problem.

Cited with a link to your own domain

GoodYour page is a linked source

Needs workYou are named, sources are third-party

PoorNeither named nor cited

Being named from a directory listing is fragile — the directory controls the facts. Being cited from your own page means your wording is the wording.

Days between a fix and the re-test

Good7 or more

Needs work2–3

PoorSame day

OpenAI documents that "it can take ~24 hours from a site's robots.txt update for our systems to adjust". Re-testing the same afternoon as a robots fix measures nothing but your own impatience.

Diagnosis

The tool says you are absent. Three very different reasons.

Work down this in order, because the cheapest cause is also the most common and the one people skip. Each branch ends in a different week's work.

First: is anything permitted to fetch your site at all?

Your robots.txt disallows OAI-SearchBot, PerplexityBot or Claude-SearchBot

Access problem — nothing else mattersFix this before writing a word of content. OpenAI documents that sites opted out of OAI-SearchBot "will not be shown in ChatGPT search answers". Run the AI Crawler Access Checker first: a single robots.txt rule explains an awful lot of zeros.

robots.txt is open, but your content only appears after JavaScript runs

Effectively empty on fetchMost AI fetchers do not execute JavaScript. Check what is in the server-rendered HTML, and check load behaviour with the Core Web Vitals Checker while you are in there.

robots.txt is open and the HTML is server-rendered

Not an access problemMove to the next question.

Second: does the answer describe your service but never name you?

The engine says "I don't have enough information about X"

Entity resolution failureThe engine cannot tie your name to one consistent business. Consistent name, address and phone across the web is the raw material for that — the Entity Checker shows what an engine can currently resolve about you.

Your service is described accurately but attributed to nobody

Content is unquotable, not missingThe facts exist on your site but not in liftable sentences. That is what the AI Content Scorer measures, and the AEO Readiness Checker scores the page structure around it.

Third: which sources did the answer actually lean on?

It cited directories, review sites and local press — none of which mention you

Third-party citation gapThis is the most actionable finding on the page and the one nobody reports. The AI Citation Gap Finder lists the sources that got quoted instead of you. That list is your outreach target list, in priority order.

It cited your Google Business Profile, with thin or wrong details

Profile is the weak linkFor "best X near me" style answers the profile is still the raw material. Fix it with the Google Business Profile Audit before touching your website copy.

It cited competitors' own websites

You are losing on your own turfTheir pages are being retrieved and yours is not. That is an indexing and structure problem, not a reputation one.

First-party evidence

What you can measure in a tool you already own

Since 3 June 2026 Google Search Console has reported on Search generative AI performance, and Google confirmed that "As of August 31, 2026, we've rolled out these insights to all websites worldwide." It is genuinely useful and considerably narrower than the coverage of it suggested.

What you want to know	What the report actually gives you	What it cannot tell you
How often you surfaced in Google's AI answers	Impressions. That is the one metric named in the announcement, broken out across four dimensions: Pages, Countries, Devices and Dates.	Clicks, average position, or the query. If you need query-level insight, that comes from the ordinary Performance report, not this one.
Whether it was AI Overviews or AI Mode	Nothing. The two are named together inside a single bucket.	Which of the two surfaced you. Google's post never says they can be told apart, and plenty of write-ups claim it does. They are wrong.
Whether ChatGPT or Perplexity surfaced you	Nothing — this is Google's report about Google's surfaces.	Anything about any other engine. No equivalent first-party report exists, which is the entire reason this tool does prompt testing instead.
Whether being fetched means you are included	Your server log shows the fetch and the user agent.	Inclusion. OpenAI states plainly that ChatGPT-User "is not used to determine whether content may appear in Search". Being fetched by it is not evidence of anything except a fetch.
Whether your pages are even eligible	Google's stated bar: a page "must be indexed and eligible to be shown in Google Search with a snippet".	Nothing extra. Check indexation with the Google Index Checker — an unindexed page is ineligible before any of this begins.

Scroll the table sideways to see every column.

Scale, for context, from Google's own announcement: AI Overviews "has over 2.5 billion monthly active users" and AI Mode "has surpassed one billion monthly users". Big surfaces, one metric.

Taking it upstairs

The same finding, told two ways

Your boss will ask "how do you know?" The difference between these two versions is whether you have an answer.

What most people report

I checked ChatGPT and we don't come up. Our competitors do. We need to start doing AI SEO — apparently AI search traffic converts much better than organic.

Why: Three problems. One test is one sample from a non-deterministic system. "Don't come up" does not say for which question. And the conversion claim has no primary dataset behind it — it circulates because it is quotable, not because anybody measured it. Stating it invites the one follow-up question you cannot answer, and it costs you the rest of the meeting.

What the tool gives you to report

Across five runs of "who are the best roofers in Sheffield?" on 27 September, logged out, we were named in zero. On the narrower flat-roof question we were named in three of five, but with "I don't have details of their guarantee terms". Both answers cited Checkatrade and Google reviews, where we have 11 reviews against Peak Roofing's 180. Google's own documentation says there is no special AI optimisation to buy — the gap is third-party sources and one missing page. Re-test booked for 11 October.

Why: Dated, counted, quoted, and it names the next action. It also concedes the good news, which is what makes the bad news credible. Nothing in it can be contradicted by re-running the test.

What to do in the next hour

Run the check, then do one thing: save the output somewhere dated. A single check is a snapshot, and snapshots only become evidence in a series — the AI Visibility Tracker exists to baseline the same prompts and watch the trend rather than the week. If you are already named some of the time, the useful next question is not "am I visible" but "how often, against whom" — that is AI Share of Voice. Both are more informative than running this page again tomorrow and reading the variance as progress.

Questions

AI Visibility Checker FAQ

Does ChatGPT recommend my business?

Run the check and read the answer. But treat the result as one sample: Anthropic and Google both document that model output is not fully deterministic even at temperature 0, and consumer products default to temperature 1.0. If you are named in four of five runs, it recommends you often. If zero of five, that is a real finding worth acting on.

Why doesn't ChatGPT know my company?

Usually one of three things, in this order of likelihood: nothing is permitted to fetch your site, no third-party source names you, or your facts are not written in sentences that can be lifted. The decision tree above separates them. Skip the diagnosis and you will spend a month on the wrong one.

Can you track ChatGPT traffic in Google Analytics?

You can see referrals from chat products in your analytics as ordinary referral traffic, but that only counts people who clicked through. It tells you nothing about answers where you were named without a link, which is most of them. Server logs show which AI fetchers hit you; neither source shows the answer text.

How do I see if my website appears in AI Overviews?

Search Console's Search generative AI performance reports, live for all sites worldwide as of 31 August 2026. They report Impressions across Pages, Countries, Devices and Dates. They do not separate AI Overviews from AI Mode — the two sit in one bucket, whatever you have read elsewhere.

ChatGPT recommends my competitor and not me. What do I change first?

Read which sources the answer cited. If it leaned on directories and review sites, your problem is third-party coverage, not your website. If it cited competitors' own pages, your problem is retrieval — indexing and structure. Those two findings lead to completely different work, and the answer text usually tells you which.

Why do I get a different answer every time I ask?

Because that is how these systems work. Thinking Machines Lab sampled 1,000 completions from one model on one prompt at temperature 0 and got 80 distinct completions; all 1,000 matched for the first 102 tokens then diverged at token 103. Variation in the middle of an answer is expected behaviour, not a malfunction.

Do I need to add anything special to my site to appear in AI Overviews?

Google says no, in those words: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." The stated bar is that the page is indexed and eligible to be shown with a snippet. Anybody selling you an AI-specific markup package is arguing with that sentence.

Do you store my results or email them to me?

No email gate and no account. The result renders on the page with the prompt and the timestamp so you can copy it, screenshot it, or paste it into a document. If you want the same prompts re-run on a schedule so you have a series rather than a snapshot, that is what the tracker does.

Primary sources used on this page

Google: no additional requirements or special optimisations for AI Overviews and AI Mode; query fan-out; AI Mode and AI Overviews may use different models — developers.google.com
Google: Search generative AI performance reports — one metric, four dimensions, worldwide rollout dates — developers.google.com
OpenAI: the ~24-hour robots.txt lag, and that ChatGPT-User is not used to determine search appearance — developers.openai.com
Anthropic: even at temperature 0.0, results will not be fully deterministic — docs.claude.com
Google Vertex AI: temperature 0 is only mostly deterministic; models default to 1.0 — cloud.google.com
Thinking Machines Lab: 1,000 completions at temperature 0 produced 80 unique outputs, first diverging at token 103 — thinkingmachines.ai
Google: AI Overviews at over 2.5 billion monthly active users, AI Mode past one billion — blog.google

Keep going

Turn one answer into something you can act on

A single transcript tells you the symptom. These are the tools that identify the cause and let you measure whether the fix worked, roughly in the order they are worth running.

AI Citation Gap Finder AI Share of Voice AI Visibility Tracker AI Crawler Access Checker Entity & Knowledge Panel Checker Google Business Profile Audit All 26 free tools →

Read the guide: how to run a full local SEO audit, Is Local SEO Dead? No, but It Has Changed and Benefits of Local SEO for Businesses. Want it handled for you? See our AI SEO (GEO) services.