Why AI does not mention you
Four levers, in the order you should look at them. The first suspicion —“it must be my site”— is almost always the wrong one.
If AI does not mention you, in most cases it is not because your site is badly built: it is because outside your own domain there is nothing to read about you. And those two problems are fixed in completely different ways.
The clearest proof we have is our own tool. The vilnux site scores 93 out of 100 on AI readability —robots.txt, llms.txt, sitemap, schema, zero critical failures— and appears in 0% of the 90 questions in its sector. If the site were the problem, that number could not be zero.
The four levers, in this order
Checked from cheapest to slowest to move: site → entity → authority → coverage. Starting at the last one is what costs people months.
1. Your site: can an AI read it?
It is the only one entirely under your control, which is why you rule it out first. It takes minutes: that the server does not block crawlers, that the page returns 200, that content lives in the HTML rather than only in JavaScript, and that structured data says who you are.
The failure we find most here is not the site, it is the firewall: a WAF returning 403 to automated requests. GPTBot and PerplexityBot are exactly that, so the 403 shuts them out and nobody notices. If this lever comes out fine —and it usually does— the problem is in the next three.
Perplexity is the quickest place to check, because it always shows its sources: how to appear in Perplexity.
2. Entity: does AI know WHO you are?
A model can read your site and still not be sure you exist as an organisation, or mix you up with something with a similar name. What fixes that are points that confirm one another: profiles declared in your schema and linked back, listings in your sector's directories, presence in knowledge bases.
It is the fastest lever to move, which is why it comes before authority. One example measured on our own domain: declaring a verified profile in the schema and linking it from the footer moved the entity score from 17 to 31 on the same day. That is minutes of work.
3. External authority: is there anything to read about you that you did not write?
This is where the problem is 75% of the time, and it is the slowest lever.
The way to check is not to count mentions: it is to open the pages AI cites when talking about your sector and see whether you appear in them. Co-occurring in an answer is not the same as being written about — that shortcut produces false 100% scores.
When this lever reads zero, the diagnosis is always the same: the only pages citing you are your own. And a brand that is only its own source gives a model no reason to recommend it over one it has read about elsewhere.
4. Intent coverage: which questions do you show up for?
Appearing in your sector's generic question is worth little. What decides a purchase are others: the use case (“what tool would I use in a 50-person company”), the problem (“we do this by hand and it no longer scales”) and the alternative (“what are the options against X”).
A global percentage hides exactly that. You can appear in 20% of a sector's questions and in none of the ones with buying intent.
What each combination means
The diagnosis is not a number: it is which of the four reads zero.
- High site, zero authority. By far the most common. Nothing to fix on your site: you need to be mentioned elsewhere. It is slow and there is no shortcut.
- High site, low entity. AI reads you but cannot place you. The fastest to fix, and the one most people skip.
- Everything high, low coverage. You appear, but in the questions that do not sell. Here content does help — content of one specific kind: whatever answers the missing intent.
- Low site. Fix it first, because until it is fixed nothing else can be measured reliably.
How long it takes
It depends on the engine, and the difference is an order of magnitude.
Perplexity searches in real time: a new mention on a page it already reads can show up in days. ChatGPT leans heavily on its base knowledge, so it takes far longer. Google AI Mode sits in between, and it also changes behaviour: in August 2026 it cut the sources it cites by 67%, and eleven brands across ten sectors lost a third of their visibility in the same week having done nothing.
That last figure matters for the diagnosis: before hunting for a mistake of your own, rule out that the engine moved. If your tool does not tell you, you will chase a failure that is not there.
What this does not fix
Two things that get sold a lot and deserve to be put in their place.
The llms.txt file. Worth having —half an hour and it does no harm— but real adoption by the engines is close to nil. We stopped scoring it in our audits for that reason: recommending work that does not move the needle is worse than recommending nothing.
Effectiveness multipliers. The sector repeats that this or that markup triples your citations. No study supports it. The only peer-reviewed evidence we know of —Aggarwal et al., KDD 2024— measures up to 30-40% relative improvement from adding statistics, quotes and citations to a text, on their own benchmark. The mechanism is real; the multiplier is invented, and it cannot be extrapolated to schema or to llms.txt.
Where to start
By measuring the four levers instead of guessing which one is broken.
The free audit does exactly that on any URL: it asks real questions from your sector, checks how many you appear in, opens the most-cited sources to see whether they mention you, resolves your entity and scores your site. No signup, no card.
And it gives the result with its confidence interval, which on a small sample is wide. We would rather say so: a percentage without its margin looks firmer than it is.
Frequently asked questions
My site ranks well on Google but I do not appear in AI. Why?
Because they are different games. SEO rewards your page being the best answer; a generative model builds the answer from several sources and decides who to name based on what it has read about you, whether or not that sits on your site. You can be first on Google and not exist in an AI answer.
Is it my site’s fault that AI does not mention me?
Almost never. Our own site scores 93/100 on AI readability and appears in 0% of its sector’s questions: if the site were the cause, that number could not be zero. The site is a necessary and insufficient condition — you rule it out first because it is quickest to check, not because it is the likeliest cause.
Is the llms.txt file worth anything?
Little, today. Worth having because it costs half an hour and does no harm, but real adoption by the engines is close to nil. We stopped scoring it in our audits for that reason: recommending work that does not move the needle costs you trust in everything else you recommend.
How long before a brand starts appearing in AI answers?
It depends on the engine. Perplexity searches in real time and a new mention can appear within days. ChatGPT takes far longer because it leans on its base knowledge. And expect the engine itself to change: in August 2026 Google AI Mode cut the sources it cites by 67%.
What do I do if AI says something wrong about my company?
You do not fix it inside the model: you fix it in the sources it reads. First locate which pages the wrong figure comes from. In one of our own cases, 8 of 9 errors a model asserted came from an outdated figure on the client’s own site, cited by the model as a source seven times.
Why does AI recommend my competitor and not me?
Usually because it has read about them on sites that are not theirs, and about you it has not. The useful check is not counting your mentions but opening the pages AI cites when discussing your sector and seeing whether you appear. When that reads zero, that is your answer.
See which of the four reads zero
The free audit measures all four levers on your domain and tells you who shows up instead of you. No signup, no card.
Diagnose my site free →Home · Plans & pricing · Methodology · Research · Perplexity · Metrics · How to choose · Use cases · Changelog · Agencies · Alternatives · Check if AI cites you · Noise and margin of error · Four-step method · Optimizing for agents · X · Product Hunt