Free AI visibility checker

How the AI visibility checker works

You give a brand, a website, and a short line about what you do. A location and up to three competitors are optional. The tool writes 10 questions. Three are discovery questions, such as who is a good fit for the job. Two are problem questions. Two compare alternatives. Two ask about your brand by name. One asks how to choose. The brand name stays out of the discovery questions, or the mention rate would be a trick.

Each question goes to three model families: a ChatGPT-class model, a Claude-class model, and a Gemini-class model. The exact model ids come from the server configuration, and the result prints them in small text. If one family does not answer, the others still count. If none answer, you do not lose a daily check.

Brand questions count for sentiment and for the accuracy check. They do not count toward the mention rate. Asking "is Acme any good?" and then celebrating that Acme was mentioned would be circular. The score weights are on the result: mention rate 35, share of voice 20, position 15, sentiment 15, accuracy 15.

Why AI visibility matters for your business

Some buyers now ask an assistant for a short list before they open Google. If the assistant names three firms and you are not one of them, you are absent from that conversation. That is not the same as losing a blue-link ranking, and it is not a reason to panic. It is a reason to see the actual paragraph.

The paragraph matters more than the score. A competitor named in seven answers and you in three is a gap you can read. A model that puts you in the wrong city is a different problem from a model that has never heard of you. I would rather you see the sentence than a dashboard with no text under it.

I write about this as GEO, generative engine optimization, on the AI search page. The work is still ordinary: a consistent name, pages that answer real questions, and mentions on other sites that say the same thing. The checker shows the gap. It does not close it.

Training data vs live search

These results come from asking models through their APIs. API answers mostly reflect what the models learned in training. The ChatGPT, Gemini, and Claude apps can also search the web, so an answer you see in the app can name a source this checker did not see. I label the result as a training snapshot for that reason. I do not pretend it is the live app.

Answers also move. Ask the same question twice and the names can change order. That is why the date is on the score and why I tell you to treat it as a snapshot. A weekly check is enough to see whether a fix changed the description. A check every hour will mostly show noise.

What the tool can tell you: whether the models know the brand, who they name instead, and whether the description fights the homepage. What it cannot tell you: your traffic, your Google rank, or what a logged-in app will say to one user tomorrow.

How to improve your AI visibility

Start with one name. The homepage title, the H1, the Organization or LocalBusiness schema, and the Google profile should use it. sameAs should list profiles that are really you. I have a longer note on entity authority on the blog. The short version is that machines merge records that share a name and a URL, and they split records that contradict each other.

Then publish the page the discovery question needs. If people ask for a service in a city, you want a page that says you do that service in that city, with the words in the HTML. Add a short FAQ in the customer's words. Earn a few mentions on sites that already get cited: a directory you control the facts on, a review profile, a piece of press that uses your name.

Do not buy a pile of random mentions, and do not stuff the brand into every heading. The accuracy check will flag a description that does not match the site. If a model says the wrong city, fix the site so there is only one city to repeat. Then run the checker again after the page has been live for a while. Training snapshots lag the site.

Example result

This is an illustration, not a client and not a measured case. Imagine a local firm, Northline Plumbing, with a competitor called Bolt Plumbing. The checker asks who to hire in Leeds, what to do about a leaking pipe, and what people know about Northline.

In the illustration, the brand questions mention Northline, and the discovery questions mention Bolt more often. The verdict says so by name. One answer places Northline in Manchester while the homepage says Leeds. That mismatch sits next to the score, because it is more urgent than a missing percentage point. The fix list says to make the city consistent, add LocalBusiness schema with the real profiles, and publish a page that answers the leak question in plain sentences.

Your result will name your brand and your competitors, and it will include the model text. Read that text. The score is only there so you can compare a later snapshot with this one.

How I read a snapshot

I read the verdict, then any mismatch, then one discovery answer in full. The score is for the next snapshot. The text is for today. If a competitor is named and you are not, that is the gap. If you are named and the city is wrong, fix the city before you chase more mentions.

A single word brand needs the domain in the match, and it still needs a human to check the sentences. "Apex" can be a dozen companies. The warning on the result is there for that reason. If the answers are clearly about someone else, the score is not your score.

I do not treat a good snapshot as finished. Models change, and a training answer lags the site. Change the page, wait, and run the check again. Three checks a day is enough. The weekly email option is there if you want the later snapshot sent to you. It is not a promise that the models will mention you.

Pair the snapshot with the SEO checker when you want to know whether the page itself is readable. A model can ignore a brand that is also blocked in robots.txt, or a page whose text only appears after JavaScript. Fix that on the URL, then ask the models again.