GEO by engine: the criteria each AI weighs before citing your brand

Field findings: four engines, four citation logics

I look at 15 sites per week. All of them have the same problem. They optimize for Google. Not for answer engines.

ChatGPT, Claude, Gemini, and Perplexity choose sources instead of ranking pages. And the logic changes from one engine to another. Neil Patel published a detailed analysis. His teams dissected 68 000 AI-generated responses. Each engine weighs different criteria.

Of the 15 sites I review every week, 11 show the same flaw. An architecture built for Google. Missing semantic cocoons. Fuzzy entities. The product pages are often excellent. But they live in silos. No semantic link connects them.

The content holds up. The architecture needs work.

An e-commerce catalog can have 1 200 flawless product pages. Without semantic linking, ChatGPT cites none of them. Claude prefers a sourced guide. Gemini ignores the brand. Perplexity demands a fresh date. Every engine has its own logic. Most sites ignore these differences. They bet on a single signal, often authority.

Neil Patel’s analysis points in this direction. Teams that prioritize a single dimension of GEO miss 3 out of 4 engines. The worst mistake: producing more content. The right approach: building a structure that each engine reads differently.

Nothing theoretical about this. Since 2016, I’ve delivered 1 300 semantic cocoons to clients. The observation remains consistent: a machine-readable structure earns more citations than brilliant isolated content. Answer engines evaluate whether a site can deliver a reliable, sourced, contextualized answer.

The weighting changes by engine. ChatGPT reads the entity graph. Claude reads depth. Gemini reads brand entity. Perplexity reads freshness. A site that ticks only one box misses the other three.

Now the details, engine by engine. The weighted criteria. The mechanisms. The action levers for an e-commerce catalog.

Gardez cette matrice sous les yeux. Elle résume ce que ChatGPT, Claude, Gemini et Perplexity pèsent vraiment avant de citer un catalogue.

ChatGPT: semantic coherence and usage come first

ChatGPT rewards coherence between entities over raw authority. A single product page is not enough. You need a cluster of linked pages.

I saw this with a client: 945 pages. Each page mentions a reference, a brand, a category. No page links these entities together. ChatGPT ignored the site. After restructuring into three semantic cocoons of 47 queries each, ChatGPT cited 3 pages in its first response.

ChatGPT builds a semantic representation of the query. It looks for sources that cover the same lexical field and the same entities. An isolated page gets skipped. A semantic cocoon of 47 queries around the same need sends a strong signal.

What ChatGPT weighs before citing:

• Semantic coverage of the intent, not backlink volume.
• Coherence between entities mentioned across multiple pages of the site.
• The presence of an explicit internal linking structure.

Neil Patel observes this in his analysis: ChatGPT favors sources that answer the user’s intent, even the less popular ones. A small catalog can beat a large marketplace. With clear internal linking.

Marie Haynes has documented this for years: answer engines favor coherent entities. Aleyda Solis recommends thinking entities before keywords. That’s the core of ChatGPT’s citation mechanism.

Concrete application for e-commerce: group product pages by need, not by administrative category. Create hub pages that explain the choice. Link every product page to its hub page. That gives ChatGPT a reading path.

ChatGPT often cites lists, tables, and comparisons. A hub page with a comparison table of 12 references increases the likelihood of being cited. Internal links from every product page to that hub page strengthen coherence.

Result: on a comparison query, the site went from 0 to 3 citations in 14 days. The only change: a restructured internal link network.

Claude: editorial depth and primary sources

Claude is the engine that demands the most substance. It cites what is sourced. A 300-word page is not enough. A product page with 847 verified reviews and a compatibility table carries more weight than an optimized guide.

With an e-commerce client of 1 200 references, Claude cited only the buying guide. Never the product pages. Why? The pages had no primary data, no measurements, no comparative sources. The internal team enriched them with test reports, compatibility tables, and validation dates. Claude began citing two product pages.

What Claude weighs:

• Primary sources (studies, tests, measurements) before sales pages.
• Facts: tables, figures, direct quotes.
• Consistency between displayed data and external sources.

Neil Patel measured this too. Claude gives more weight to long documents and verifiable data. A dense, reliable page attracts citations. A thin page gets ignored by Claude.

For an e-commerce catalog, this changes everything. More content adds nothing. You need to add proof. A compatibility table, for example. Or a weight measurement, a tested delivery time. Every piece of primary data strengthens citation.

Claude often pulls excerpts from dense pages, precise figures, and sourced comparisons. A product page with a table of 12 measured characteristics has a higher chance of being cited than an empty marketing page.

Observed result: a site with 620 references saw Claude go from 1 citation to 4 citations after adding 12 sourced comparison tables. Still without ads.

Gemini: brand entity and local context

Gemini weighs the brand entity. Your Google Business profile. Your Wikipedia page. Your social profiles. Without a clear entity, Gemini skips you.

Neil Patel notes this: Gemini relies on Google’s Knowledge Graph to validate entities. A brand missing from the graph has little chance of appearing in an answer. A consolidated entity, however, opens up citations.

An e-commerce catalog with 3 physical stores saw its Gemini citations go from 0 to 4 after consolidating its brand entity. What they did: unify the name, address, and phone number across local listings, create an About page with structured data, and link the social profiles.

What Gemini weighs:

• The strength of the brand entity in Google’s graph.
• Local context: reviews, address, information consistency.
• The co-occurrence of the brand with entities in its sector.

Lily Ray has long documented the impact of structured data on visibility in rich results. Gemini applies the same logic to generated answers. A well-structured brand is cited more often.

For e-commerce, GEO optimization also goes through the brand ecosystem, not just the website. An incomplete business profile weakens citation. A local listing with 37 reviews and a verified address strengthens it.

What I observe: an increase from 0 to 4 citations on a category query after 21 days of entity consolidation. No new article. No artificial backlink.

Perplexity: freshness and data verifiability

Perplexity cites what is fresh and verifiable. A product page with a visible update date and cross-referenced sources wins the citation. A page frozen for 14 months disappears.

I saw a catalog of 620 references land 5 citations after adding update dates, availability badges, and links to independent external sources. Before, Perplexity cited only marketplaces. Not the site.

What Perplexity looks at:

• Freshness: publication date, update date, recent signals.
• Verifiability: cross-referenced sources, corroborated figures.
• Response speed: lightweight, structured pages that are easy to parse.

Neil Patel confirms this: Perplexity favors sources that display demonstrable freshness. A page updated 48 hours ago weighs more than a guide frozen for 2 years, even if that guide has more backlinks.

Search Engine Land regularly highlights the importance of freshness signals in generated answers. Perplexity, in particular, cross-checks dates and sources to validate an answer. A catalog without an update date remains invisible.

Application for an e-commerce catalog: display an update date on every product page. Add a real-time availability badge. Cross-reference data with a recognized external source. Three simple moves, and everything changes.

Observed result: a site with 1 200 references went from 1 citation to 6 citations in 30 days. No new page. No campaign. Just better data transparency.

Les chiffres sont têtus. Voici combien de citations chaque moteur a servi une fois le socle sémantique en place.

What these weightings change for your e-commerce catalog

The four engines don’t ask for four different stratégies. They ask for a single foundation that answers their respective weightings.

I apply the DOSE framework taught by Guillaume Attias at BMO Academy. It forces you to map entities before writing anything. A clear semantic architecture, consolidated entities, verifiable data, fresh dates. That’s the foundation.

For an e-commerce catalog, it looks like this:

• Semantic coherence: group product pages into semantic cocoons of 47 queries around the same need. Link each page to its hub page.
• Editorial depth: add comparison tables, tested measurements, and primary data on the pages.
• Brand entity: consolidate the Google Business profile, social profiles, and local structured data.
• Demonstrable freshness: display update dates, real-time availability, and cross-referenced sources.

Each engine pulls from this foundation what matches its weighting. ChatGPT reads the semantic cocoons. Claude reads the tables and primary sources. Gemini reads the brand entity. Perplexity reads dates and verifiability.

So you don’t need four content versions. You need a rich architecture. One single corpus, read differently by four engines.

That’s what I see with my clients: the more structured the foundation, the more citations climb from one engine to another. A catalog of 1 200 references, restructured around these four levers, went from 2 citations to 11 cumulative citations in 45 days.

Think of your catalog as a network of evidence, not a series of isolated pages. Every page counts. A hub page opens a door. A data point with a date gives a reason to cite.

Avoid the classic trap: optimizing only for ChatGPT. You’ll get citations on one engine. You’ll remain invisible on the other three. The shared foundation covers all four.

Un seul socle répond aux quatre logiques. Voici les étapes, dans l’ordre, telles que je les applique en audit.

Forge a single corpus without spreading thin

Visibility is shifting. Google is no longer the only playing field. Four AIs read your catalog. Everything depends on what they find there.

ChatGPT, Claude, Gemini, and Perplexity cite different sources. They weigh different criteria. But all of them look for the same thing: clear, coherent, verifiable, fresh content.

That’s why I show you the pages instead of selling a method. A 45-minute audit is enough to see if your catalog is ready. We map the semantic cocoons. We test a query on each engine. You leave with the exact actions, not an 80-page report.

The question to ask yourself: are your product pages ready to be cited by ChatGPT, Claude, Gemini, and Perplexity?

Stéphane Jambu

SEO & AI Engineer

I build growth systems / AI / Neuroscience | 650+ clients · 80 LinkedIn testimonials · 30 years of expertise · 15 years of systems running without me.

Follow on LinkedIn