Ask an assistant to recommend a supplier in your category and watch what happens. Two or three names appear, described confidently, with a source or two. If yours is not among them, the reason is usually mundane rather than mysterious.
The first and commonest is that your site says nothing a model can repeat. Pages full of adjectives — premium, bespoke, innovative — give a model nothing to stand behind. A model repeats facts, because a fact can be attributed and an adjective cannot. Say what you build, for whom, where, on what, at what scale, and you become quotable.
The second is disagreement. Your site, your listings, your profiles and an old press release describe the company four slightly different ways. Models resolve entities across sources, and an entity that contradicts itself is the one they leave out in favour of a competitor whose story is boring and consistent.
The third is structure. Without schema.org markup a model has to infer what is a service, what is a price and what is a location from prose — and inference is where it either gives up or gets it wrong. The fourth is access: robots rules written years ago, never re-read, quietly excluding the crawlers that feed these systems.
The fifth is the one nobody can fix quickly. Models weight sources that other credible sources mention. If nothing outside your own domain has ever said your name, you are asking to be trusted on your own word. That is a slow problem, and it is the same slow problem it always was.
The service behind it
Generative Engine Optimisation (GEO)
Optimisation for AI search: making a brand the source ChatGPT, Gemini, Perplexity and Google's AI answers are assembled from.