How this check works
This is the only check that asks an artificial intelligence. Scanaris gives a language model your domain (without the www) and your page title, and asks it to answer only whether it has real, specific prior knowledge of that site or brand, with the explicit instruction to say no if it would be guessing from the name. The model returns a yes or a no. If it is no, it opens an informational finding that costs no points. If the model fails or does not answer in the expected format, nothing is opened: when in doubt, no finding is invented. A small, fast model is used so the test is cheap and consistent.
What the result means
A "no" means that, in the knowledge that model was trained on, your brand does not appear enough for it to recognise you. It is an approximation, not an exact measure: each assistant is trained on different data and can also search live. But it is a useful signal, because models train on similar sources, and if one does not know you, the others probably do not either. And the consequence is concrete: when someone asks ChatGPT, Claude or Perplexity "which tool should I use for X", assistants recommend what they know. If they do not know you, they do not recommend you, however good your product is. For a new brand it is the expected result; the interesting part is scanning again in a few months and seeing whether it changes.
How to improve it
There is no technical shortcut: models learn from what is written about you on sites they crawl. What works, in order of impact:
- Be where models read. GitHub, Product Hunt, directories in your category, G2 and Capterra if you are software, useful answers on Reddit, Stack Overflow and Hacker News, articles in your industry's media. Consistent mentions of your name next to what you do.
- Content that answers real questions. Assistants cite pages that directly answer what people ask: "how to secure an app built with Lovable", not "innovative security solutions". Each question, one page or one section with its heading.
- Structured data with
OrganizationandsameAspointing to your profiles, so machines join your site to what they already know about you from other sources; structured data measures it. - A distinctive brand name, always used the same way. A brand that is a common word is almost impossible to recognise.
- Let them read you: do not block AI crawlers and serve the content in HTML, not only in JavaScript.
Count on months, not weeks: models are retrained periodically and live searches need sources to cite.
Check it is fixed
There is no command for this. The closest thing is asking three assistants yourself, in a window without your history: "What is yourdomain.com?" and then "Which tools do you recommend for [your category]?". Note the answers and repeat each quarter. The next Scanaris scan runs the automatic test again.
Frequently asked questions
The model says it does not know me but ChatGPT does find me.
ChatGPT with search enabled visits the web live, which is a different thing: it finds you as Google would. This check measures prior knowledge, the kind that gets you recommended without anyone having searched for you. How to check the other part is in how to know if ChatGPT can see your site.
Why is it informational and not penalised?
Because it is not a flaw of yours, it is the state of your reputation. A freshly launched site with everything perfect will get a "no" all the same. It is a baseline to measure progress against.
Can I "trick" the model?
Not from your site: the model does not visit it in this test, it answers with what it already knows. The only thing that changes the result is more reliable information about you existing out in the world.