For twenty years, being visible online meant one thing: climbing into Google's ten blue links. That world is closing. When a shopper asks ChatGPT "which serum for sensitive skin under $40?", they don't get a list of links — they get an answer, and two or three cited sources. If your product isn't in that answer, it doesn't exist for that shopper.
From ranking to citation
SEO is a ranking game: ten results, and you fight to sit higher than the competitor. GEO (Generative Engine Optimization) is a citation game: the engine reads the web, synthesizes an answer, and decides which sources deserve to appear in it. There are no longer ten slots — there are two or three.
The behavior shift is the real earthquake. The shopper no longer scans ten links: they read one answer and act. This is zero-click search. In this world, the citation is the new click — and if you're not cited, you're nowhere.
This shift isn't a distant hypothesis. More and more shoppers start their research by asking an AI rather than typing a Google query — especially for purchases where advice matters: cosmetics, electronics, sports gear, baby products. The question is no longer whether your customers will use these engines, but when you'll become visible inside them.
Why your SEO doesn't automatically carry over
Good SEO helps — a fast, clean, well-linked site gets crawled better. But it is not enough. Answer engines don't look for the best-ranked page: they look for usable facts. A page that pleases Google's algorithm can stay mute to a language model if it doesn't contain, in a machine-readable way, what the shopper is asking for.
- Google ranks pages; AI extracts facts. The former wants a title and keywords; the latter wants a precise attribute ("fragrance-free", "SPF 50", "GTIN").
- The meta description vs JSON-LD. One seduces a human in a SERP; the other gives the model structured data it can cite without getting it wrong.
- Backlinks vs factual completeness. Popularity is still a signal, but a model favors a source that is complete and consistent about the requested product.
This doesn't mean your SEO work is wasted. A fast site, clean architecture, tidy URLs and well-linked pages: all of it helps AI crawlers explore you. But that's the foundation, not the building. GEO happens one layer up — the layer of facts a machine can extract and cite.
A concrete example: one page, two readings
Take an ordinary product page: a sunscreen. To a human it looks perfect — nice photo, persuasive copy, five-star reviews. To a language model, it can be almost empty. Here's why.
What the shopper sees
A marketing promise ("protects and enhances your skin"), a price, an "Add to cart" button, and copy written to seduce. The essential information — SPF, water resistance, sensitive-skin compatibility — is often buried in a paragraph, or worse, only present on the packaging image.
What the model sees
A model doesn't read the packaging image and doesn't "understand" marketing. It looks for facts: SPF 50? yes/no. Fragrance-free? unknown. Water-resistant 40 min? missing. Result: to the question "a fragrance-free SPF 50 water-resistant sunscreen?", your product — which may tick all three boxes — isn't cited, because nothing states it in a readable way.
The tragedy is that the product is the right answer. It's just missing the layer that tells the machine so. That's exactly where GEO plays out: not inventing qualities, but exposing the ones that already exist.
What answer engines actually read
Beneath the visible text of your page sits a layer the shopper never sees but machines read first: structured data (schema.org), your product feed, and clean factual text. A product described this way becomes comparable — the model can place it in an answer.
{
"@context": "https://schema.org",
"@type": "Product",
"name": "Sérum Vitamine C — Peau sensible",
"brand": { "@type": "Brand", "name": "Veridian" },
"gtin13": "3401590012345",
"description": "Sérum sans parfum, formulé pour peaux sensibles…",
"additionalProperty": [
{ "@type": "PropertyValue", "name": "Type de peau", "value": "Sensible" },
{ "@type": "PropertyValue", "name": "Sans parfum", "value": "Oui" }
],
"offers": {
"@type": "Offer",
"price": "34.90",
"priceCurrency": "EUR",
"availability": "https://schema.org/InStock"
}
}With this block, the question "a fragrance-free serum for sensitive skin under $40?" has a clean answer: skin type, fragrance-free, price and availability are explicit. Without it, the model has to guess — and it doesn't guess in your favor.
The three pillars of citability
Making a page citable comes down to three requirements. None is magical; together, they make the difference between a product the AI can recommend and one it ignores.
1. Structure
Data must be machine-readable: schema.org/Product markup, named attributes, normalized identifiers. It's the grammar engines expect. Without it, your facts live in free text, where they're harder to extract with confidence.
2. Factual completeness
An empty structure is useless. The decisive attributes of your industry must be filled in: INCI and skin type in cosmetics, size and material in fashion, compatibility and specs in electronics. A product whose pages mostly expose the key attribute will be cited far more often than one where it's missing.
3. Consistency
The engine cross-checks several sources: your page, your feed, your markup. If the price differs between feed and page, or the stock is wrong, the model loses confidence — and an inconsistent source is one that gets cited less. Feed ↔ page consistency isn't a technical detail: it's a reliability signal.
Five mistakes that make you invisible
- Putting specs only on the image. The model doesn't read text baked into a packaging visual.
- 100% marketing descriptions. "Enhance your skin" answers no question; "fragrance-free, dermatologically tested" does.
- No identifiers. Without GTIN/EAN, your product is hard to match against comparisons and databases.
- Missing or broken markup. A malformed
schema.org/Productis sometimes worse than none at all. - Price/stock inconsistencies between feed and page, which sabotage your reliability in the eyes of engines.
GEO in practice: where to start
- Structure. Expose a clean
schema.org/Productand the decisive attributes of your industry. - Complete the facts. Identifiers (GTIN), specs, price, availability — without inventing anything, from your real data.
- Measure. Track your citations and share of answers, engine by engine, to know where you actually appear.
- Iterate. Close the gaps on the queries that matter, then re-measure.
How long before you see an effect?
Let's be honest: GEO isn't a switch. Engines re-crawl the web at their own pace, and citation also depends on the competition for each query. Think in weeks rather than days before your presence in answers shifts. The right approach: structure, measure a baseline, fix the gaps, then re-measure. It's ongoing work, not a one-shot — and no honest party can guarantee you a citation.
SEO puts you in the list. GEO puts you in the sentence.
SEO isn't dying — it's changing target
Answer engines still crawl the web: a healthy site remains essential. What changes is the target of optimization — from rank to citability. The two coexist, but the share of traffic that flows through an AI answer will only grow. Better to be machine-readable today than invisible tomorrow.
Who should start now? Any merchant whose products are chosen on criteria — in other words, almost everyone. The richer your catalog is in attributes (beauty, electronics, sport), the higher the stakes: that's exactly where AI needs facts, and exactly where most pages stay silent.