Why your competitors appear in ChatGPT answers and you do not

Not because anyone compared your machines with theirs and preferred theirs. An answer is assembled from sources that can be retrieved and weighed, so the company that is easiest to describe accurately tends to be the company that gets described.

A supplier written up consistently across its own site, a trade register, an association listing, a technical article and a distributor directory gives an answer five places to draw from, all saying the same thing. A supplier whose only description of itself sits on its own website gives it one, unsupported. The second is harder to name — and from where you are sitting, harder to name looks exactly like judged and rejected.

That difference is testable. Ask the questions your buyers ask, record who is named, and look at which sources each answer rests on.

Is this a judgement about our equipment?

No. Nothing in the process assesses your machines. Systems synthesise from the material they can retrieve, and weigh it by how well it is corroborated. Quality of engineering is not an input unless somebody has written about it somewhere a system can reach.

Which is why the pattern is often the opposite of what you would expect. A specialist with thirty years in aseptic fillers can be absent from an answer that names a generalist with a large published library. The generalist is not better. It is better described.

What is being assessed is the material, not the machine. How an answer is assembled out of retrieved documents is set out in how AI builds supplier shortlists.

What makes a company easy to name

Three things, and only one of them lives on your website.

The first is whether an answer cites any supporting source at all, rather than simply asserting. The second is how strong and how relevant those sources are — a trade register and a supplier's own brochure are not equal evidence. Those two are Citation Presence and Citation Authority, and both are defined on Prove, the step of our method that exists to improve them.

The third is consistency. If your company is described one way on your site, another way in a directory entry written in 2019, and a third way on a distributor's page, there is no settled version of you to work from. Contradiction is not neutral. It is worse than silence, because it gives a system a reason to reach for a company it can describe without hedging.

Where the evidence actually lives

Outside your website, mostly. Trade registers, association membership listings, standards and certification bodies, machinery directories, technical press, conference programmes, distributor and integrator pages.

Your own site establishes what you say about yourself, which matters and is not sufficient. Independent sources are what turn a claim into something corroborated.

Your website is your statement. Everywhere else is your evidence.

The practical version: a case packer manufacturer with an excellent site, listed in one machinery association directory under a five-year-old description that does not mention glass handling, will struggle to be named for glass jar questions no matter how good the site is.

On the underlying mechanism: Google's generative features are built on its core Search ranking systems and use retrieval-augmented generation and query fan-out to surface material from the Search index — Optimizing your website for generative AI features on Google Search, Google Search Central.

But our website is better than theirs

It may well be. That is largely beside the point, and it is the hardest part of this to accept.

Design quality, copywriting and photography are not what is being weighed. What is being weighed is whether the information can be found, understood and confirmed elsewhere. A beautiful site making claims that appear nowhere else is still one source with nothing behind it.

Trade evidence for polish and the answer changes. Keep the polish and add the evidence, and it changes further.

Are they just bigger than us?

Often, yes. Size correlates with coverage: bigger firms have more listings, more press, more people writing about them without being asked.

But size is not the mechanism. Coverage is, and coverage can be built deliberately at any size. A mid-sized specialist who publishes real technical answers about a narrow application can be named for that application while a multinational is named only for the general category. The specific question is where a smaller company can win, because the general question is where scale already has.

This is also what AI Visibility measures across a whole question set rather than one answer — how often you appear at all, defined on Baseline.

What you can check this week

Take five questions your buyers ask before they know who to call. Put each to ChatGPT, logged out, and read the answer twice.

The first read tells you who was named. The second is the useful one: look at what the answer cited. If the sources are your competitor's own pages, they have won on first-party clarity. If the sources are trade bodies, directories and technical press, they have won on corroboration, and that is the harder position to take from them.

Then ask the same system to describe your company, and see whether what comes back is what you would have written. If it is out of date or wrong, that is a different problem with a different fix, and it usually moves faster than anything else on this list.

What this article cannot tell you

It cannot tell you which specific source put your competitor into a given answer. Systems do not publish their weighting, and anyone who claims to know it is guessing with confidence.

It cannot rank the sources that matter in your category from here. That varies by industry and by country, and finding out is work rather than knowledge.

And answers move. AI systems are probabilistic and change continuously, so one reading is a snapshot rather than a position. What tells you anything is the same set of questions, asked the same way, on a schedule.

What to do next

The discipline that deals with all of this has three competing names, and the differences between them are narrower than the people selling them suggest — we have set that out in GEO vs AEO vs SEO.

If you would rather have it measured properly — 30 to 50 agreed questions, all six systems, your named competitors measured alongside you, and the sources recorded for every answer — that is the AI Visibility Audit. Either way, the useful move is the same: stop guessing why they are named and go and look.