They ask an AI instead. And the AI answers from what it already knows - not from a list of links. If it doesn't know you, you were never in the running.
For twenty years the question was "where do I rank on Google?" That question still matters. But a growing share of buying decisions never reaches a search results page at all - someone types a question into a chat window, gets one answer, and acts on it. There is no page two. There is no scrolling. There is one recommendation, and either you are in it or you are not.
Ten blue links. Your competitors are there, you are there, an affiliate blog is there. The user compares, clicks around, forms an opinion. You had a chance to compete on the page - with your title, your copy, your reviews.
Three names. Maybe two. The model answers from what it absorbed during training - and it never saw your site, or it saw it once and forgot. You did not lose the comparison. You were never in it.
A single number would be easy to sell and hard to trust. So we ask each model eight separate questions about your domain, and the score is what comes back.
Has the model actually seen this domain, or does the name just look familiar? Zero means it has never encountered you at all.
How well the model knows you when it sees the name. Nothing at all, the topic only, or the full picture.
Would it point someone to you as a source? This separates "has heard of you" from "trusts you enough to send people your way."
How many concrete, checkable statements it can say about you. Who runs it, where it is, when it started, what it sells.
The one that pays the bills. If someone asks for a solution in your field without naming you - do you come up?
Where you sit among everyone else in your category, as the model understands that category.
Does it know what business you are in? A high score here with a low Recommend means it knows you but does not rate you.
How current its picture of you is. A model can know the company you were in 2019 and nothing about the one you are now.
No crawling, no scraping, no clever tricks. We ask the models directly and write down what they say.
Every scan runs against the model's own training knowledge. We explicitly block web lookups, because a model that can search will find your site in a second and tell you it knows you. That would measure Google, not the model. We want the part that is actually baked in.
Language models are not calculators - ask twice, get two slightly different answers. So we ask three times and keep the median, the middle value. One odd run cannot drag your score around, and you are not looking at noise dressed up as a trend.
Claude and GPT were trained on different data at different times. One knowing you and the other not knowing you is a real, useful signal - and a single model would have hidden it. You always see both scores separately, not just the blend.
The eight metrics average into a score per model, and the two models combine into your AI Visibility Score. If a model fails to answer, we throw the whole scan away rather than show you half a picture with a number that looks complete.
Most tools in this space are vague about their limits. Here are ours, up front, before you pay anything.
The ones we would ask if someone showed us this page.
Because two very different things get called "the AI knows me." A model that searches the web will find any live site and describe it back to you - that tells you your site exists, which you already knew. What we measure is the other thing: whether you are part of what the model carries around by default, before it looks anything up. That is what decides whether you get named when someone asks a question without mentioning you.
Not today's score - that one is fixed, because it reflects a model that already finished training. But the next version is a different story. Structured, well-sourced content that gets picked up and cited is exactly what shapes what future models learn, and that is a real practice worth investing in. What a low score does tell you now is real: the fastest-growing way people find things has never heard of you. That is worth knowing while you decide where your attention goes, and worth watching for the moment it changes.
Because they disagree, and the disagreement is the interesting part. Claude and GPT were trained on different material at different times, So one can know you well while the other has never heard of you. A single number from a single model would have quietly averaged that away. You see both, side by side, always.
Yes, and plenty of people do exactly that. Scanning is free for any domain, and a slot lets you keep any domain in your dashboard - it does not have to be yours. Your score alone is a number. Your score next to the three companies you lose deals to is an argument.
One scan, thirty seconds, no account and no card. You will know more about your AI visibility than most of your competitors know about theirs.
Scan My Domain