Our own numbers, since they make the point better
An AI visibility audit of inite.ai returns 90 out of 100 for readiness and 25 out of 100 for visibility.
That means the crawlers can parse everything we publish and the engines still rarely mention us. Two of six engines recall the brand when handed the name. Zero of six recommend us in category, which is the question a buyer actually asks.
Search Console tells the same story from the other side: 30 clicks and 3,262 impressions across a month, and zero impressions against any commercial query. Not low positions. Absence.
We are publishing this because the alternative is writing about AI visibility from behind a number we have not earned, and because the gap itself is the useful thing to understand.
Readiness is cheap. Visibility is not.
| AI readiness | AI visibility | |
|---|---|---|
| What it measures | Whether an engine can read you | Whether it chooses to mention you |
| Who fixes it | A developer, in a fortnight | Evidence accumulated over months |
| Improves | Immediately | Slowly, if at all |
| Cost | Low and one-off | Ongoing |
| What it is worth alone | Nothing | The whole thing |
The two get priced as if they were the same product. A vendor who sells visibility and delivers readiness has delivered something real, much cheaper than what you thought you bought, and you will not notice for a quarter because the readiness score moves immediately.
The number to watch
Category recommendation, not brand recall.
Brand recall asks whether an engine knows you exist once your name is supplied. It is easy to improve and worth very little, because someone who already knows your name has other ways to reach you.
Category recommendation asks whether you appear when somebody describes a problem and asks who solves it. That is the question a buyer types. Our own split of 2 of 6 against 0 of 6 is the difference between being findable and being recommended, and only the second has revenue attached.
Ask any vendor which of those two their headline number describes. The answer is informative.
What actually seems to move it
Nothing exotic, and nothing that can be finished in a fortnight.
Answer a specific question completely, on a page that is about that question, in the form the question is asked. Mark it up so the answer is extractable rather than buried in a narrative. Then get corroborated somewhere that is not your own domain, because an engine weighing whether to recommend a vendor is looking for something that is not the vendor's own claim about itself.
What does not appear to move it is the thing most commonly sold. llms.txt appeared on 10.13% of domains and SE Ranking's November 2025 study across 300,000 domains could not detect a citation lift attributable to it. We publish one anyway because it is inexpensive to maintain. That is a much weaker claim than the one usually made for it, and the full argument is here.
Why bother, when 93% of AI Mode is zero-click
Because the traffic is not the asset.
In a zero-click answer the assistant states a conclusion and names sources. Being one of those names puts you in the shortlist a buyer carries into their next conversation, including the one where they eventually type your name directly.
Treating it as a traffic channel produces the wrong measurement and then the wrong conclusion, which is almost always that it did not work. Count whether you are named in answers to the questions your buyers ask, and watch branded search over the following months. If your reporting counts only sessions, a successful programme and a failed one look identical.
What an operator should do about it
Three things, in order, and none of them is buying a tool.
Find out what an assistant currently says when asked who does what you do in your city or sector. That takes an afternoon and it is the only baseline that matters.
Fix readiness once, because it is cheap and because being unreadable makes everything after it pointless. Whether you let the crawlers in at all is a business decision rather than a technical one, and the allowlist post sets out the trade.
Then spend the ongoing effort on evidence rather than markup: cases with numbers, answers to real questions, and corroboration on domains you do not own. That is slow, and it is the part that separates the two scores. The mechanics, with the schema that matters and the part that does not, are in the AEO guide.
The honest summary
We are good at the cheap half and bad at the expensive half, and we can prove both with numbers.
Anyone selling you the cheap half at the price of the expensive one will show you a score that improves in week two. Ask what it measured.
