Direct answer
For twenty years, "ranking well" meant one thing: ranking well on Google. That's no longer true. Each AI relies on different sources: ChatGPT uses Bing's index, Claude uses Brave Search's, Google (AI Overviews, AI Mode) its own index with Gemini, and Perplexity its own engine. The direct consequence: being first on Google guarantees nothing in ChatGPT or Claude, which don't read the same web. Search optimization is no longer a single playing field — it's split up by AI. And the only way to know where you show up is to measure each AI separately.
The problem
The old world was simple: one dominant engine, Google, and one discipline, SEO, to be visible on it. You optimized for Google, you measured on Google, and that was pretty much it.
That world is over, and many people haven't taken it in yet. Today your customers ask their questions to ChatGPT, to Gemini, to Claude, to Perplexity — and these AIs don't draw on the same web. Continuing to think "search optimization = Google" means optimizing for a single field when the game is now played on several, with different rules.
The idea to grasp
Here's what few people realize: each AI has its own source of information. These aren't different windows onto the same web — they're different indexes.
- ChatGPT (OpenAI) relies mainly on Bing's index, supplemented by its own crawler. One analysis showed that the vast majority of ChatGPT's citations match the top of Bing's results. In practice: if you're invisible on Bing, you risk being invisible in ChatGPT — even with an excellent Google ranking.
- Claude (Anthropic) relies on Brave Search, which has its own index, neither Google nor Bing. The overlap between Claude's citations and Brave's results is very high. Content that ranks well on Google may not on Brave — and therefore may not show up in Claude.
- Google (AI Overviews, AI Mode) relies on the Google index and its Gemini model. It's the field closest to classic SEO — but it's only one of the fields.
- Perplexity uses its own search engine, with its own judgment calls.
Put these facts end to end and you get a truth that changes everything: search optimization has fragmented. There's no longer one ranking, but several, on indexes that overlap only partially. You can be very present in ChatGPT (via Bing) and nearly absent in Claude (via Brave), without knowing it.
On top of that comes the memory layer: beyond web search, each AI also has its own training memory, which differs from one model to another. So there are two levels of fragmentation: the web sources (Bing / Brave / Google / own) and the models' memories.
The old reflex "I take care of my Google and I'm done" no longer holds. Taking care of Google helps for Gemini and a bit for the web in general, but tells you nothing about your presence in ChatGPT (Bing) or Claude (Brave).
What you hear everywhere
"I'm first on Google, so I'm visible everywhere." No longer true. First on Google tells you nothing about Bing (so about ChatGPT) or Brave (so about Claude). Three fields, three possible results.
"SEO is SEO, whatever the engine." The underlying principles are similar (authority, quality, structure), but the indexes differ: ranking well somewhere doesn't automatically carry over elsewhere.
"Just optimize for the biggest one, ChatGPT." And your customers who use Claude or Perplexity? Each AI has its own user base. Betting on a single one means ignoring the other fields where your visibility plays out.
My stance: only the facts. And the stubborn fact is that today there are several indexes, several memories, several results. Assuming they're aligned is a belief; measuring how they diverge is data.
My vision: measure each AI separately
From here on, the register changes: we describe the instrument.
If the playing field is fragmented, the measurement must be too. Measuring your AI visibility seriously means never blindly aggregating providers:
- Query each AI separately (ChatGPT, Claude, Gemini, Mistral, Perplexity…), because each has its own sources.
- Compare the results across providers: the gap is major information (present here, absent there).
- Separate memory from web for each one (model knowledge and live web aren't the same from one AI to the next).
- Repeat and date each measurement, by provider, to track the divergences over time.
A measurement that mixed all the providers into a single score would erase precisely the most useful information: which field you're winning on, and which you're losing on.