The short version
An AI visibility check asks several assistants the question a customer would ask and records whether your business is named. It measures what a model already knows from training rather than live search, so a new or thinly-covered business will usually not appear regardless of how good its website is.
What this measures, and what it does not
It measures recall. When you ask an assistant for a recommendation without letting it browse, the answer comes from what the model absorbed during training. So this is a test of whether your business exists in that absorbed knowledge, not of whether your website is good.
That distinction matters because the two are only loosely related. A business with an excellent site that nobody writes about will not be recalled. A business with a mediocre site that gets mentioned in local press, directories, forum threads and roundups will be.
It is also not a test of AI search. When an assistant is allowed to search the web before answering, the dynamics are completely different and much closer to ordinary SEO. This tool deliberately asks without browsing, because that is the harder question and the one nobody can fix with a quick content edit.
Why results vary between assistants and between runs
Do not read a single run as a verdict. Three things make these answers unstable.
Models have different training data and different cutoffs, so one may know a business another has never encountered. Answers are also generated with some randomness, which means the same question can produce different names on different runs. And the phrasing of the question shifts the answer substantially: asking for the best in a city produces a different list to asking who to call for a specific problem.
So treat a result as a signal rather than a measurement. Being named by none of them repeatedly is meaningful. Being named by two out of four once tells you very little.
What actually moves this
The uncomfortable answer is that it is largely the same thing that has always driven reputation: being written about by other people.
Models learn from text at scale. A business mentioned in local news, industry roundups, supplier directories, forum answers and review platforms accumulates the kind of coverage a model absorbs. A business whose entire web presence is its own website has given the model almost nothing to learn from, because one site is a vanishingly small share of any training corpus.
This is why the tactics being marketed as AI SEO deserve scepticism. Publishing a file that declares what your business does cannot substitute for other people writing about you. Google has said it does not use llms.txt for its AI features, and we do not generate such files for customers.
The durable work is unglamorous: get mentioned in places that discuss your category, keep your listings consistent so a model does not encounter three different versions of your name and address, and publish content specific enough that someone might cite it.
- Mentions in local and trade press
- Inclusion in category roundups and directories
- Consistent name, address and phone across listings
- Content specific enough that others quote or link it
- Reviews on the platforms your category actually uses
Reading the competitor names carefully
When an assistant does not name you, the tool lists the businesses it named instead. That list is useful and it deserves a caveat.
We extract those names from prose, which is imprecise. A capitalised phrase might be a competitor, or a city, or a product, or a trade body. Treat the list as a rough read on who occupies that mental space, not as a ranking.
What it is genuinely good for is noticing patterns across several assistants. A name that comes up in three answers out of four is a competitor with real presence in the training data, and that is worth knowing about regardless of how you feel about the methodology.
Questions about this tool
Why did no assistant mention my business?+
Most often because there is not much text about you in their training data. This measures what models already absorbed, not the quality of your website, so a newer business or one that is rarely written about elsewhere usually will not appear.
Does this use AI search or live browsing?+
Neither. It asks without browsing on purpose, because that tests what the model already knows. When assistants are allowed to search first, the dynamics are much closer to ordinary SEO.
Will I get the same result if I run it again?+
Not necessarily. Model responses carry some randomness and the phrasing of the question changes the answer. Being named by none of them repeatedly is meaningful; a single mixed result is not.
How do I actually improve this?+
By being written about elsewhere: local and trade press, category roundups, directories, forums, reviews. Models learn from text at scale, and your own website is a tiny share of any training corpus.
Do llms.txt files or AI SEO tactics help?+
Be sceptical. Google has said it does not use llms.txt for its AI features, and publishing a file about yourself does not substitute for other people writing about you. Venbit does not generate these files.
Which assistants does it ask?+
Whichever are configured, from ChatGPT, Claude, Gemini and Grok. The result names any that were skipped because no key is available, rather than quietly leaving them out.