"Why does ChatGPT recommend my competitor and not me?" It's the question more and more merchants are asking. The answer owes less to luck than to a fairly logical mechanic — once you understand how a model goes from a shopper's question to a handful of cited products.
First: ChatGPT doesn't "see" your store the way you do
A model doesn't browse your site like a human scrolling. Depending on the engine and mode, it relies on what has been crawled and indexed, on structured data, sometimes on a live web search. In every case, it doesn't "feel" your nice photo or your storytelling: it manipulates facts and text. That gap explains most of it.
The three steps of a recommendation
1. Find the candidates
Faced with a question, the engine gathers a set of plausible products: those it knows, that match the topic, that are reachable in its data source. If your product is nowhere in a usable form — no structured data, no clear presence — it doesn't even enter the starting list.
2. Filter on the facts
The question almost always contains constraints: "under $100", "gluten-free", "iPhone-compatible", "for sensitive skin". The model eliminates anything that doesn't meet them — or anything it can't verify meets them. A product that ticks the constraint but says so nowhere readable is eliminated as if it didn't.
3. Choose what to cite
Among the survivors, the model favors clear, complete and consistent sources. A precise, structured page with no price/stock contradiction inspires more "trust" than vague text. There are only two or three slots: completeness and reliability break the tie.
An example: "earbuds for running, waterproof, under $100"
This question sets three constraints: use (running), waterproofing, price. Here's what the model tries to verify for each candidate:
- Use: is the product described for sport / running? (an attribute, not just a photo of an athlete)
- Waterproofing: is a clear rating (IPX4, "sweat-resistant") present and structured?
- Price: is it under €100, in a clean currency, consistent with the feed?
A product that exposes these three facts cleanly is citable. An identical product whose waterproofing is only mentioned on an image is dropped — not because it's worse, but because it's unreadable.
Why some products never appear
- No structured facts: everything is in marketing text or on visuals.
- Contradictions: price or stock differ between feed and page → loss of trust.
- No identifiers (GTIN): impossible to match the product against comparisons.
- Too thin a page: not enough information to break ties.
What you can influence (and what you can't)
Let's be clear: you control neither the model's weights nor the exact algorithm, and no one can guarantee you a citation. What you control is the raw material: the presence, structure, completeness and consistency of your facts. That's exactly what the model uses to decide. You don't drive the decision — you make the right decision possible.
You can't force an AI to cite you. You can remove every reason it has to ignore you.
In practice
- Expose your facts as structured data (
schema.org/Product). - Fill in the decisive attributes of your industry.
- Remove contradictions between feed, page and markup.
- Measure where you appear, and fix where you're missing.