Being cited by ChatGPT is not the same as being found

More and more people no longer put their buying question to Google, but to ChatGPT or Perplexity. Not "AI agency Zeeland" in a search bar, but "which company can build an AI agent that reads my invoices?" in a conversation. The answer they get is a short list of names. If you are not on it, you do not exist for that buyer.

I was curious how that mechanism actually works, so for a stretch of time I tracked which sources AI search engines cite when they answer this kind of question in the Dutch AI market. What came out was not what I expected, and it is useful for any business that wants to be found online, not just AI agencies.

What AI actually cites, and what it does not

The first surprise: AI search engines almost never cite a homepage or a service page. They pull their answers from a handful of fixed formats. In the market I followed, the vast majority of citations went to four kinds of pages: overview guides ("how to choose a partner for X"), rankings ("the best providers of Y"), how-to articles and comparisons. Together they accounted for the lion's share of everything cited.

So your most polished service page barely shows up in those answers. Not because it is bad, but because it is not the format an AI reaches for when someone asks "who can do this for me?". For that question the model picks a list, and that list is almost always on a site other than yours.

Being cited does not bring you visitors

The second surprise was less comfortable. You can be cited plenty and end up with zero visitors from it. During my measurement window there was steady citation activity, but the number of people who actually clicked through from such an AI answer to the site was practically zero.

That makes more sense than it seems. An AI answer is the end point in itself. The user reads the summary, sees three names, and maybe remembers one. They do not click through the way they would on a Google result. Your visibility does not sit in clicks, then, but in whether your name appears in that answer and in what context. That makes the old reflex ("chase more traffic") less important than a new question: am I being mentioned, and is what is being said about me correct?

Why a perfect site of your own is not enough

This is where the two findings meet. If AI answers "which company for X" questions from other people's lists, then whether you appear in the answer is not decided by your site, but by whether other, trusted sources mention you. I held my own findability up to the light and saw exactly that: technically everything was in order, the content was complete, and yet I was missing from the answers to the concrete buying questions. The reason was not on the site, it was next to it. I was not on the lists the model was drawing from.

That is not a pleasant message, because your own site is in your hands and someone else's is not. But it is the honest one: at a certain point, more polishing of your own pages is a waste of your time, and the gains shift to the outside.

What to do about it in practice

A few things that do work, in order of how much they return.

Make sure AI is actually allowed to read you. Sounds obvious, but plenty of sites unknowingly block the crawlers from OpenAI, Anthropic or Perplexity. Check your robots settings and explicitly allow those bots. This is free and takes five minutes.

Tell the model who you are, in structured form. With schema markup (organisation, services, frequently asked questions) you give an AI hard facts instead of leaving it to guess. That lowers the chance of nonsense being told about you, and raises the chance that you are classified correctly.

Publish in the formats AI cites. An honest buyer's guide or comparison in your field has more chance of being cited than the tenth service page. One caveat: if you crown yourself the winner of such a list, a model weighs that lightly. It has to be genuinely useful to the reader, otherwise it backfires.

Build presence in places AI trusts. This is the real work and it takes time. Being mentioned in trade media, in neutral overviews, in a guest article: those are the sources models cite from. One mention in an authoritative place does more for your AI visibility than a month of tweaking your own copy.

Measure whether you are being mentioned. You do not need an expensive tool for this. Each month, put a few real buying questions from your market to ChatGPT and Perplexity, and note whether your name comes up and what is being said about you. That is your new position check.

Staying level headed

AI search engines are changing how people find companies, but the underlying logic is not new. Whoever appears in the answers is whoever gets mentioned by others in the places the model looks. Having your own house in order is the baseline, not the winning point. The gain sits in other, credible sources mentioning you, and you earn that over the longer term, not with the flip of a switch.

For most businesses that is more reassuring than discouraging. It means you do not have to write a new page every week. It means your energy is better spent outward: on the kind of presence a model reads as trustworthy. Less polishing, more being seen where it counts.