An AI visibility check tells a business whose expertise is the product which of three separate things is failing: being cited as a source, being named, or being recommended to a buyer, because each one has a different fix.
Someone in an online forum counted it out. ChatGPT had cited about 700 pages off their site. It had named the site five times.
Seven hundred to five.
Their question was the right one. Is a citation worth anything without the name?
I spend more time in forums like that one than I would like to admit. It's where people say the true version of what is happening to their business. The polite version goes on LinkedIn.
Three things are getting counted as one number. AI citing your page as a source. AI naming your business. AI telling a buyer to call you. Those three are separate, and pulling them apart is the first job of an AI visibility check.
A citation means a model used your page to build an answer. Your sentence is in there. Your name isn't.
So you wrote the reference material for an answer that sent the reader somewhere else.
For a business whose expertise is the product, that one lands in a specific way. The thing you spent years learning is doing the work in the answer. It's doing it for a stranger, for free, under a heading with somebody else's contact form at the bottom.
Being named is a narrower question. It asks whether AI is confident you exist as one thing. Same business name everywhere it appears, a person attached to it, a stated service area. That's the most fixable of the three, and the one every technical audit already covers.
Recommended is a different problem.
Someone in the same forum put it plainly. Ask a model what it knows about a brand and it will tell you, accurately. Ask that same model who it recommends for the category and the brand is not on the list.
A model can know you fine and still leave you off the list.
One question is whether it can find you. The other is whether it has any reason to pick you over the three names it already has. Those aren't the same failure and they don't have the same fix. Being known and being recommended are two different events.
I keep seeing this in the sites I audit. Most of them fail three or more of the six things I score, and the naming problems are almost always the cheap ones.
Which one is broken changes what you should pay for.
Schema, site structure, meta descriptions, entity cleanup. All of that moves the naming problem. It does close to nothing for the recommending problem. If a model can already describe your business correctly and still hands the buyer to someone else, more technical work on the site is money spent on the part that's already working.
Being known and being recommended are two different events.
Worth saying plainly, because I sell one of these. An audit tells you which of the three you have. It does not fix the third one. Finding out and fixing are two different pieces of work, and they cost different amounts.
What moves a recommendation is having something specific to say about you that cannot be said about the other three names. Not better adjectives. Real numbers, a stated way of working, a page that answers the actual question a buyer asks with specifics in it. And a lot of it is not on your site at all, because the pages a model cites are usually somebody else's pages talking about you.
That's the split nobody separates before they buy. Someone else in the same thread said it from the other side: being cited does not matter much, because almost nobody clicks the links a model shares when it cites. Worry about being recommended.
You can tell which one you have without buying anything.
Open a model you don't use for work, in a window with no history, because your own chat history will feed you back the answer you want. Ask it what it knows about your business. Then, separately, ask it who it recommends for the thing you sell, for the kind of buyer you want.
I run this on my own business every month, and I have never once looked forward to it. There is a particular feeling that comes with reading a confident, well-organised paragraph about your own industry that does not have you anywhere in it.
Run it more than once, on more than one model, and write down what you got. A single answer on a single afternoon is a data point, not a verdict.
If you sell outside your own market, run it there too. Two people working in different parts of the world said the same thing within days of each other. AI visibility doesn't travel. Strong English-language results at home, then almost nothing on the same queries one country over, because the model leans on American sources even when the local work is better.
So whichever of the three is broken, it's broken per market. You check each one separately or you don't know.
If the first answer is wrong or empty, you have a naming problem, and a technical fix will move it.
If the first answer is right and the second answer is a competitor, stop fixing the website.