Almost every ecommerce founder arrives at the same question about AI search. It usually comes out as: “How do I show up in ChatGPT?”
The instinct is understandable. ChatGPT was the thing everyone saw first; it became shorthand for the whole shift. If your customers are asking an assistant what to buy, and that assistant is ChatGPT, getting recommended there feels like the whole game.
The problem is that the ground has already moved underneath the question.
Twelve months ago, ChatGPT accounted for roughly three-quarters of generative AI traffic. The most recent readings put it closer to half, with Gemini’s share more than doubling over the same window and Perplexity and Copilot both pulling real volume. None of this means ChatGPT is in trouble — it’s still enormous. It means the assumption underneath “optimise for ChatGPT” is quietly expiring. You’re no longer trying to be found in one place. You’re trying to be found across a handful of them, each reading the web in its own way.
Why fragmentation changes the job
If discovery were consolidating onto a single assistant, the right move would be tactical: learn that one engine’s quirks, work them, win the slot. That’s the SEO muscle most teams already have — find the algorithm, find the lever, pull it. A comfortable problem, because it’s a familiar one.
But discovery isn’t consolidating. It’s splintering. And you can’t run four separate optimisation tactics for four engines that each change every few weeks. You’d never finish, and you’d be chasing each one as it moved.
The brands that win the recommendation aren’t gaming a single engine. They’re the ones whose product information is coherent enough that any assistant — reading from any source, on any given week — arrives at the same answer about what they sell and who it’s for.
The real failure mode: illegible, not invisible
Here’s where it gets uncomfortable. Most catalogues aren’t coherent. The specs on the product page say one thing. A marketplace listing says something slightly different. An old blog post or a retailer’s feed says a third thing, two years out of date. To a human, that’s noise you’d never notice. To an assistant assembling a confident answer, it’s contradiction — and a contradictory catalogue is one the model quietly skips in favour of a competitor it can describe without hedging.
So the failure mode isn’t being invisible to AI. It’s being illegible to it. You’re in the data. You’re just not telling a consistent enough story for any assistant to stake a recommendation on you.
That reframing is the whole thing. The work stops being “rank in ChatGPT” — a tactic you re-run forever — and becomes “make my product information agree with itself everywhere it appears” — infrastructure you build once and maintain. And it pays off in every engine at once, including the ones that haven’t launched yet.
The side-by-side test
Pick a category you sell in. Ask three different assistants — ChatGPT, Gemini, Perplexity — the same buying question a real customer would ask. Read the three answers side by side.
If you get a consistent picture of your brand back, you’re in good shape. If you don’t appear, or you appear differently in each, that inconsistency is your work for the quarter. It’s a slightly deflating exercise the first time. That’s rather the point.
Discovery is moving from search engines to AI assistants. It’s also moving from one assistant to many. Plan for the second shift, not just the first — coherence infrastructure is how, and Geoffy builds it from the catalogue you already have.
See how consistently the engines read you: get your free GEO Score — it checks four engines, not one.