Free AI Visibility Tool
This AI visibility tool asks ChatGPT, Perplexity, Gemini and Claude the kind of questions your buyers ask, then reports whether they named you, who they named instead, and which sources they cited. It runs on your own API keys, so there are no limits and no signup - you will need a key from at least one provider before you start.
Run the check on your own API keys - no signup, no limits.
Paid reach and AI-search reach are the two halves of the same visibility question - work out the paid side with the CPM calculator.
AI Visibility Tool at a Glance
How to run your first check
Five minutes, most of which is the key.
Get an API key from one provider. OpenAI is the quickest. Create a key, add a small balance, and paste it in. You only need one to get a result, and you can add the others later.
Enter your domain and brand name. Use the name people actually say, not the registered company name, unless they are the same thing.
Describe the topic you want to be recommended for. This is what the prompts are built from, so be specific. Project management software beats software.
Name two or three competitors. Optional, but it turns a score into a comparison, and the comparison is the part worth acting on.
Run it, then read the crawler table first. If the crawlers are blocked, nothing else in the report is telling you what you think it is.
Run it again in a month rather than tomorrow. Assistants vary between runs, and a day-to-day comparison is measuring that variance rather than your work.
What the report looks like
Label it unmistakably as an example. Acme Analytics is fictional and every figure below is illustrative; do not swap in a real brand, because a page publishing “ChatGPT does not mention [real company]” is making a public claim about a third party from a single sample run.
How to read this example. The two weak scores have completely different causes, and the crawler table is what separates them.
Claude, 3 of 10 - an access problem. Both Anthropic crawlers are blocked in robots.txt. The brand still surfaces occasionally because a model can name something it learned during training, but nothing on the site can be fetched or cited now. No amount of content work moves this one; it is a robots.txt edit.
Gemini, 0 of 10 - a coverage problem. Crawlers are allowed and the brand still never appears, so access is not the constraint. This is the harder of the two to fix and the more useful to know about.
The competitors column is where the answer usually sits. Asana appears in all four assistants, so the sources they trust on this topic contain Asana and do not contain Acme. That points the work at the citation list rather than at the site.
How this compares with the other free checkers
Most people arrive here having already run one of the free brand checks, or having read a list that ranks them. It is worth being straight about where this one sits.
The trade is a real one. A checker with no key gets you a number in thirty seconds and stops there. This one costs you a few minutes of setup and then does not stop - unlimited runs, your own prompts, four assistants including Claude, and the crawler check that explains most zero scores. If you want the thirty-second version first, take it, then come back when the number turns out to be a problem worth understanding.
What AI visibility means
AI visibility is whether an assistant names your brand when someone asks it a question you should be the answer to.
It is not a ranking. There is no position one, there is no results page, and two people asking the same question can get different answers. What exists instead is a probability - across many phrasings of the same question, how often does your name come up, and when it does, does it lead the answer or trail it behind three competitors.
That distinction matters because the old instincts do not transfer. A page can rank first in Google and never be mentioned by an assistant answering the same question. The assistant is not picking a link, it is composing an answer from what it has read, and what it has read is shaped by what it was allowed to crawl, what it found there, and how clearly that content stated a position.
AI visibility, GEO and AEO
Three names circulate for work that overlaps almost entirely, and the distinction matters less than the vocabulary suggests.
AI visibility is the measurement - whether assistants name you, and how often. Generative engine optimization, GEO, is the work done to change that - the same work described by people who include voice assistants and featured answers in the scope. Vendors pick whichever term their category uses; the underlying job is being present, quotable and crawlable in the sources an assistant reads.
None of it replaces search work. An assistant that answers from live results is reading pages that had to be findable in the first place, so the two disciplines share most of their foundations and diverge on what happens after the page is found.
How this tool checks it
The tool builds a set of prompts from your topics, market and audience - the questions a real buyer would type rather than the keywords a marketer would target - and sends them to each assistant you supplied a key for.
For every answer it records whether your brand appeared, how prominently, which competitors appeared alongside or instead of you, and which sources were cited. Those citations are the most actionable part of the result: they are the pages the assistant trusted on your topic, and they are where your absence is costing you.
It runs on your API keys rather than ours, which is why there is no signup, no daily cap and no queue. It is also why you need a key before you can run it. If you do not have one, the fastest route is an OpenAI key - it takes a few minutes and a small balance covers many runs.
The four assistants, and why the answers differ
Expect the results to disagree, and do not treat the disagreement as an error. A search-first assistant reflects what is findable and current on the open web right now. A model-knowledge answer reflects what was absorbed during training, which is slower to change and harder to influence.
Read the pattern rather than the individual scores. Strong on the search-led assistants and weak on the others usually means your content is fresh but your brand is not yet well established. Weak everywhere usually means an access or a coverage problem, and the next section is where to look first.
Can the assistants reach your site at all?
Before optimising anything, check whether the crawlers are allowed in. A site that blocks them cannot be cited, no matter how good the content is - and this is the single most common cause of a zero score.
Each company runs several crawlers with different jobs, and blocking the training crawler does not block the search crawler. They are separate directives in robots.txt and they have to be set separately.
This is where a lot of sites have quietly locked themselves out. Blocking GPTBot to keep your content out of model training is a legitimate choice, and on its own it costs you nothing in ChatGPT’s search results - those are governed by OAI-SearchBot, a separate directive. The damage happens when a site blocks both, usually by pasting a blanket AI-crawler block from a blog post. It has then removed itself from ChatGPT’s search results while believing it only opted out of training.
If you want the training opt-out without the visibility cost, block the training crawler and allow the search one. Google-Extended works the same way: it controls Gemini training and has no effect on ordinary Google Search.
One caveat worth knowing: Anthropic states that all three of its crawlers honour robots.txt, including the user-triggered one. Not every provider treats user-triggered fetches the same way, so a robots.txt rule is a reliable control over indexing and a less reliable one over a single user asking about your page directly.
Two rows in that table behave unlike the rest. Microsoft runs no separate AI training crawler - Copilot draws on the Bing index, so a Bingbot block removes you from both at once, and there is no way to keep one without the other. CCBot belongs to Common Crawl, which is not an assistant at all but an open archive that many models have been trained from; blocking it affects future training sets broadly rather than any single product.
Fix what you find here first. Once your crawlers are allowed in, make sure they have something worth reading - the llms.txt Generator builds the file that hands an assistant your site’s summary directly, rather than making it guess from your HTML.
How to improve your AI visibility
Let the right crawlers in. Everything else is wasted until this is true. Check it first, fix it in an afternoon.
Answer the question in the first paragraph. Assistants extract answers, they do not reward a build-up. A page that states its conclusion plainly at the top gets quoted; one that arrives at it in paragraph nine does not.
Be specific enough to be worth citing. Numbers, dates, named methods and stated positions survive extraction. Generic marketing language does not - there is nothing in it for an assistant to lift.
Get named in the sources the assistants already cite. Your own site is one input among many. The citation list in your results is a target list: those are the pages the assistant already trusts on your topic, and appearing on them moves the needle faster than another post on your own blog.
Keep entity details consistent. What your company does, where it operates, what it is called - stated the same way across your site, your profiles and third-party listings. Assistants assemble a picture from fragments, and contradictory fragments produce a vague answer or no mention at all. The Schema Markup Validator checks that those facts are stated machine-readably, not just in prose.
Re-run and watch the trend. A single run is a snapshot with real variance in it. The number that means something is the direction across several runs a few weeks apart - and since you cannot put parameters on a link an assistant composed, AI referrals arrive untagged and have to be separated by referrer rather than by campaign. The UTM Builder still covers the traffic you do control.
Common mistakes
Reading one run as a verdict. Answers vary between runs of the same prompt. Treat one run as a sample, not a score.
Blocking every AI crawler, then wondering about visibility. The most common self-inflicted wound on this list.
Optimising for the assistant instead of the reader. Content stuffed with question-shaped headings and no substance gets ignored by both.
Ignoring the competitors column. Who gets named instead of you is more useful than your own score. It tells you which sources the assistant trusts and what it thinks the category looks like.
Expecting fast movement. Search-led assistants can reflect a change within weeks. Model knowledge shifts on training cycles you do not control.
Assuming Google rankings carry over. They do not, reliably. Ranking first and being uncited on the same question is a normal result, not a bug.
If AI search is where your buyers are asking, that is a channel with its own work. See how the Zaprev team approaches generative engine optimization.
FAQ
Frequently Asked Questions
What is an AI visibility tool?
Software that asks AI assistants questions from your category and records whether they name your brand. It exists because assistants do not publish a ranking you can look up - the only way to know how often you are recommended is to ask repeatedly and count. This one checks four assistants and reports mention rate, rivals named and sources cited.
Why do I need my own API key?
Because the tool queries the assistants directly on your behalf, and those queries cost money per request. Running them on our account would mean caps, queues and a signup wall. Using your own key means no limits, no daily cap and nothing to register for, at the cost of a few minutes’ setup the first time.
Is this the same as checking my Google ranking?
No, and the two often disagree. A search engine returns an ordered list of pages; an assistant writes an answer and may name nobody at all. You can hold the top organic position for a query and go unmentioned when the same question is asked conversationally, because the assistant is choosing what to say rather than what to link.
Why do different assistants give different answers?
They draw on different material. Some answer mainly from live search results, others lean on what the model absorbed during training, and most blend the two depending on the question. That is why one assistant can recommend you consistently while another has never heard of you - and why the pattern across all four tells you more than any single figure.
Does blocking GPTBot hide me from ChatGPT?
Not from its search results. Blocking the training crawler keeps your content out of future model training, which is a separate thing from being surfaced when someone searches inside ChatGPT - that is governed by a different crawler and a different robots.txt line. Sites that want one without the other frequently block both by accident.
How often should I re-run the check?
Monthly is enough for most brands, and weekly during an active push. Any single run carries variance, so what you are watching is the trend rather than the number. Running it daily produces noise you will misread as movement.
How long before changes show up?
It depends which assistant. Ones that answer from live search can reflect new or updated content within a few weeks. Ones answering from model knowledge change only when the underlying model is retrained, which is outside your control and considerably slower.
Is there a free AI visibility checker that does not need an API key?
Yes, several, and they are worth running. The large SEO platforms and one or two marketing suites publish one-shot brand checks that need nothing from you but a brand name. What they give back is a fixed set of questions across two or three assistants, once. This tool asks for a key so that it can ask your questions instead of theirs, as many times as you want, without a cap or a queue.
What is the difference between AI visibility, GEO and AEO?
Mostly vocabulary. One names the measurement, the other two name the work, and different vendors prefer different labels depending on whether they count voice assistants and answer boxes in scope. Anyone promising you something fundamentally different under one of those acronyms is selling the label rather than the practice.
Can I check a competitor instead of my own brand?
Nothing stops you, and it is a useful exercise before a pitch. Put their name in the brand field and yours in the competitors field, and you will see the category the way an assistant describes it to their buyers. Treat the output as one sample of a probabilistic system rather than a finding about another company, and do not publish it as one.