GEO by engine: each AI’s citation mechanics
Summarize this article with AI
I look at 15 sites per week. The problem is no longer the definition of GEO.
I look at 15 e-commerce sites per week. They all want to appear in ChatGPT, Claude, Gemini, or Perplexity.
A client calls me on a Tuesday morning. $8,000 invested in an « AI optimization ». Result: 0 citations on Claude, 47 queries on Gemini. The problem wasn’t the content. It was the architecture per engine.
Why does the same brand appear on Gemini and disappear on Claude? Each engine reads different signals. Not the same structures. Not the same proof.
I break down the four engines. Not the global theory. The mechanics.
Here’s what I observe with my clients. Well-tagged product pages win on Gemini. Broad editorial guides win on Claude. Pages with dated figures win on Perplexity. And ChatGPT? ChatGPT reformulates the consensus.
Four engines. Four logics.
The market is saturated with GEO definitions. Everyone explains Generative Engine Optimization. No one shows how ChatGPT decides to cite one page over another. That’s the only thing that matters for an e-commerce owner.
A site with 4,000 organic sessions per month invested $8,000 in an AI redesign. Six months later, still 4,000 organic sessions. The only difference: 47 queries visible on Gemini. None on Claude. Why?
This site sells fitness equipment. 45 content pages. None appear on Claude. They are all isolated. No pillar page. No editorial link. Claude doesn’t see coherence.
On Gemini, 47 queries. On ChatGPT, 12. On Perplexity, 5. On Claude, 0. The diagnosis is simple.
Each engine has its mechanics.
Sur mes déploiements, le même site peut être cité par Gemini puis disparaître de Claude. Voici la répartition brute des citations observées sur 14 sites e-commerce suivis.
Citations par moteur sur 14 sites
Gemini cite 11 sites sur 14, Claude seulement 2 : l’écart qui justifie la segmentation par moteur.
The market is saturated with definitions. Engine segmentation is the real topic.
The DOSE framework, which I apply with Guillaume Attias / BMO Academy, follows four phases. Discovery. Optimization. Structure. Évaluation. Applied to GEO, it starts with discovery: segment by engine.
Neil Patel writes it in his article on GEO by engine: selection criteria vary from one engine to another. ChatGPT synthesizes, Claude retains, Gemini reads structured data, Perplexity seeks fresh data. A single response for four engines is a response that appears nowhere.
Across my deployments, 11 out of 14 sites get Gemini citations. Only two on Claude. Three on Perplexity. The gap comes from the response structure, not notoriety.
A brand does not disappear from an engine by chance. It disappears because its signals don’t match the engine’s mechanics.
DOSE taught me to capture attention through precise segmentation. The advanced audience is tired of generic LLMO comparisons. They want the mechanics. I detail them below.
Engine segmentation is the lever that remains when everyone has read the same definitions.
Neil Patel describes it: engines don’t cite the same sources because they don’t aim for the same thing. ChatGPT wants a comprehensive answer. Claude wants a reliable answer. Gemini wants a locally useful answer. Perplexity wants an immediate and sourced answer.
The DOSE framework helps prioritize. Discovery: which engine your audience uses. Optimization: which format for which engine. Structure: which technical data. Évaluation: which actual citation you gained.
You capture the attention of an advanced audience by avoiding the général. Engine segmentation serves that purpose.
Four engines, four citation logics
Four engines, and four ways to get cited.
| Engine | Dominant criterion | Observed effect |
|---|---|---|
| ChatGPT | Reformulation of multiple sources | Often cites comparatives and aggregators |
| Claude | Response safety and editorial consensus | Prefers stable sources, few lists |
| Gemini | Google Shopping integration and structured data | Favors well-tagged product pages |
| Perplexity | Freshness and precision of the short answer | Cites dated facts, quantified data |
What I see with my clients: schema.org-tagged product pages appear on Gemini, but not on Claude. Claude requires broader editorial coherence.
Perplexity answers the direct question. It cites a dated fact, a price, an update. ChatGPT prefers the synthesis of a consensus. The pages that come up are never the same.
A single page can no longer be cited everywhere. Each engine demands its own proof template.
Segmentation changes. Not by search intent. By response engine.
A design furniture site has 320 well-structured product pages. On Gemini, it gets 62 citations. On Perplexity, 17. On Claude, 3. The difference? Claude wants in-depth articles on layout, not product pages.
Technical signals aren’t enough everywhere. Gemini likes structured data. Claude wants editorial depth. ChatGPT relies on consensus, Perplexity on freshness.
As long as you optimize the same way for all engines, you’re paying for nothing.
ChatGPT ne cherche pas la vérité, il agrège un consensus. Voici comment il transforme une requête en bundle de sources.
La mécanique d’agrégation de ChatGPT
De la requête au bundle de sources : cinq étapes, jamais une page unique citée.
ChatGPT: more of an aggregator than a truth engine
ChatGPT doesn’t seek the truth. It aggregates a consensus.
When a query arrives, it expands the intent into sub-queries, queries several sources, then reformulates the most probable response. It doesn’t cite one page: it cites a bundle of pages.
Order of magnitude observed with my clients: pages with an integrated comparison show up 3 times more in ChatGPT responses than classic product pages.
Why? Because ChatGPT privileges pages that already cover the consensus. A product page covers only one angle. A comparison covers several at once.
The mechanics are clear. If you want to be cited on ChatGPT, don’t behave like a product page. Behave like a mini buying guide.
But be careful. No catch-all page. ChatGPT spots pages that want to cover everything without deciding anything. It prefers a structured page that takes a position.
- Include a simple comparison table.
- Add short definitions at the top of the page.
- Answer the most frequent objections.
- Use sourced quantified data.
No jargon, no long introduction. The first paragraph gives the answer. ChatGPT will pick it up more easily.
We saw a spare parts e-commerce double its ChatGPT citations by transforming 12 product pages into mini comparison guides. Same technical base. A different proof architecture.
ChatGPT analyzes the title structure, the presence of lists, the density of direct answers. It looks closely at pages that themselves cite multiple sources.
A page that mentions 4 expert opinions, 2 studies, and 3 comparisons has more chances of being reformulated by ChatGPT than a page that cites no one.
The classic trap: believing that ChatGPT reads internal linking. It does, but as a consensus signal, not as a direct ranking factor. Linking creates context. Context feeds the synthesis.
Ask yourself the right question: does my page cover the query’s consensus? On ChatGPT, that’s the criterion that makes the difference, far ahead of writing quality.
Claude and Perplexity: trust before visibility
Claude and Perplexity are the two engines farthest apart when it comes to citing their sources.
Claude bets on editorial trust: a stable source, a neutral tone, and coherence that holds over time. It rarely cites lists, and it avoids overly aggressive commercial pages.
Perplexity, on its side, runs on fresh data. It likes up-to-date prices and verifiable facts, in short answers.
Result: the same brand can appear on Perplexity for a factual query, then disappear on Claude for a commercial query. Same brand, same pages. But different signals.
What I observe with my clients:
- Claude: complete editorial guides come up. Product pages alone, no.
- Perplexity: pages with a recent figure and a dated source come up. Vague pages, no.
One client, to give an order of magnitude: 18 additional queries in Perplexity after a purge of outdated dates. It was enough to remove the « 2023 » that had become false.
Claude, for its part, demands depth. A 1,200-word article on a specific point is worth more than a 3,000-word guide that skims everything.
The lesson: optimizing for Claude has nothing to do with optimizing for Perplexity. And trying both at the same time often satisfies neither.
Claude is trained to minimize hallucinations. It prefers not to answer than to cite a source without consensus. Consequence: a page with a single customer review weighs nothing for Claude. It needs a bundle of evidence.
Perplexity, conversely, loves short answers. A page that gives the answer in 40 words from the introduction, it picks up. A page that dithers 300 words before answering, it skips.
With Claude, trust is about stability and coherence. With Perplexity, it’s won through precision and freshness. Two very distinct work sites.
What you gain by segmenting your GEO by engine
Segmenting by engine changes how you prioritize.
For each page, we define a Gemini objective, a ChatGPT objective, a Perplexity objective, a Claude objective. Not the same page. Not the same format. Not the same effort.
A catalog of 800 references can generate 400 Gemini citations in 3 months. On Claude, it will get nothing without a solid editorial base. Accepting this difference avoids investing $8,000 in a uniform optimization that will appear nowhere.
The DOSE framework helps prioritize. Discovery: which engines already cite your market. Optimization: which pages to strengthen for which engine. Structure: which structured data to activate. Évaluation: which actual citations you gained.
We start with an audit per engine. I query the four engines with the queries that matter. I note who cites you. Who cites your competitors. And why.
Then we segment the effort. No uniform content. Signals per engine.
The per-engine audit takes 3 hours on a catalog of 500 pages. I query ChatGPT, Claude, Gemini, Perplexity with 10 strategic queries. I note the citations. I compare with competitors.
After that, we distribute. Product pages toward Gemini. Comparisons toward ChatGPT. In-depth guides toward Claude. Fresh data pages toward Perplexity.
Result: actual citations, not hypotheses.
The e-commerce owners who will win produce for the right engine, at the right time. Producing more is no longer enough.
And you, which engine already cites your catalog without you knowing it?
GEO audit by engine: see who really cites you
I show you the pages that come up on ChatGPT, Claude, Gemini, and Perplexity. And the ones that are missing. No theory, just your data.
Book a strategic call — 45 minFrequently Asked Questions
Should you optimize each page for all four engines?
No. Each engine cites in its own way, and I see it every week with my clients. Gemini picks from product pages. ChatGPT, from comparisons. Claude, from long guides. Perplexity, from fresh data.
Why does my brand appear on Gemini and not on Claude?
Gemini reads structured data and Google Shopping. Claude, for its part, relies on editorial trust: sources that are stable over time. Your product pages alone are not enough for Claude.
Which signal matters most for ChatGPT?
Consensus. ChatGPT cross-references what comes up most often in sources. A page that covers multiple angles (a comparison, for example) has more chances of being cited.
How to measure your citations in AI engines?
I manually run the 10 queries on ChatGPT, Claude, Gemini, and Perplexity. I note who cites me, who cites my competitors. Then I redo the test every two weeks to see what changes.
Does GEO by engine replace classic SEO?
No, it compléments it. Clean structured data and well-done internal linking help classic Google and Gemini. But each AI engine expects its own proof format.
