AI knows you exist, but it doesn't know what you sell: if a buyer asks who makes jacquard fabrics in merino wool for prêt-à-porter, your name doesn't come up — yet if they ask about you directly, the answer is correct. The model knows your brand as a name, but doesn't connect it to your products, and that connection won't build itself. You're losing the most commercial queries — the ones where the customer is already trying to buy. Explicitly linking products and brand in your content is the fix that turns your visibility around for the searches that convert.
AI knows your brand but doesn’t know you make product X. Or it knows you make X but doesn’t link the brand to the product. Both are worlds that barely talk to each other, and the result is the same: when a fashion entrepreneur asks ChatGPT “who makes jacquard fabrics in merino wool for prêt-à-porter”, your name doesn’t come up. But if they ask “who is [your brand]”, the answer is correct. The engine knows you exist. It doesn’t know what you do.
In this article I explain why this happens, what the research says about the way AI engines link brands and products, and how to make every line you produce become an entity that AI can recommend on its own — not as an appendix to the company name.
Brand entity and product entity are two different things
For an AI engine, your company name and each product you make are distinct nodes in the knowledge graph. They are not the same thing, they don’t live in the same record, and they don’t automatically inherit each other’s attributes.
In the field of e-commerce search research, Dong Liu et al. (2025) describe how brand entity linking works this way:
“Each brand entity is unique across languages, stores and surface forms.”
Translated: every brand entity has a unique identity that the engine recognizes across linguistic, graphical, and spelling variants. When your brand is recognized, the system treats it as a single object in its graph.
The operational consequence is simple: if the graph contains only the “company” node, the engine has a single object to recommend — you, in your entirety. If, instead, there are also connected product nodes (jacquard line, bouclé line, carded wool line), the engine has several objects it can cite independently. With equal authority, the second configuration generates more opportunities to appear in AI answers.
What the AI reader asks when it searches for a product, not a name
The same authors explain why the problem is hard:
“The task presents unique challenges because queries are extremely short (averaging 2.4 words), lack natural language structure, and must handle a massive space of unique brands.”
On average a query is made up of 2.4 words. Short, without structure, often without context. That’s why the engine has to decide quickly which entity to attach it to.
It follows that the queries “boiled wool fabrics made in Italy” and “[your brand] fabrics” trigger different linking logic. The first looks for products. The second looks for you. If you want to come up in the first — the commercially richest one — your products need to be entities in the graph, not just mentions inside the “about us” page.
The link to what I told you about embedding and vector space is direct: the engine works on vector representations of concepts, and every entity has its own vector. If “jacquard fabric in pure wool for women’s jackets” is not a node but just an expression lost in a paragraph, its vector doesn’t exist as an independent object and can’t be retrieved as a result.
Without brand, offers, material, category, the node is too thin to be attached to specific queries.
The test you can run in ten minutes
Open the Google Rich Results Test and paste the URL of your homepage. Look at what it finds:
- If you see only Organization → the engine recognizes only the company.
- If you see Organization + Product (or multiple Products) → you have products as entities.
- If you see nothing structured → you are invisible as an entity, full stop.
Then repeat the test on the page of a single product or line. If only Organization comes up there too, the problem is systemic: your site talks about products but doesn’t declare them as objects to the engine.
Third step: open ChatGPT or Perplexity and ask “who makes [specific fabric type] in the province of Prato”. If your brand doesn’t come up but a smaller competitor’s does, it’s very likely the difference is here — and not in your volume of content.
Keep the limit in mind: the Rich Results Test is an entry-level check. It tells you whether the markup is there, not how rich it is or how it’s actually interpreted. The real analysis of how AI is citing your industry requires professional tools and observation work on the actual answers.
List your main lines or products — not every code, the 5-10 that truly represent your business.
The A/B comparison I observed in the Prato textile district
Over the past few months I’ve taken a close look at two very similar Prato textile companies: both produce regenerated wool fabrics and blends for women’s prêt-à-porter, both with revenue between 4 and 7 million, both with a four-generation family history, both suppliers to Italian and French luxury fashion houses.
The first — I’ll call it Company A — has a site with Organization schema, three descriptive “collections” pages, and no Product markup. The second — Company B — has Organization schema plus a dedicated Product schema for each of its five lines: boiled wool, bouclé, jacquard, flannel, technical fabrics. Each Product has name, description, brand, material, and the reference to the seasonal collection.
I ran ten queries on ChatGPT and Perplexity, varying between “who makes regenerated wool fabrics for women’s fashion”, “Prato textile companies boiled wool”, “jacquard producers in merino wool Italy”, “bouclé fabrics made in Prato” and variations.
Out of ten queries, Company B is cited seven times, four of them as the first result. Company A comes up twice, both times when its name literally appears in the query. It’s an indicative test, not a study: small sample, a single industry, a single district. The pattern, though, is clear, and it recurs in other similar comparisons I’ve seen in niche manufacturing.
The difference is not in the nicer site, the longer blog, or the backlinks: it’s in the fact that in the engine’s knowledge graph, Company B has six citable entities. Company A has one.
The mistakes I see most often
- A “products” page that lists everything in a grid without Product schema on each item. The user sees them, the engine sees a wall.
- Generic Product schema with only name and description. Without brand, offers, material, category, the node is too thin to be attached to specific queries.
- Internal company product names (item code, internal abbreviation) instead of semantic names. “PRM-47B” is not a recognizable entity; “Jacquard fabric in pure virgin wool 420g” is.
- Copy-pasted descriptions across different products. The engine doesn’t distinguish them, it flattens everything onto the company node.
What to do, in order
- List your main lines or products — not every code, the 5-10 that truly represent your business.
- For each one, create a dedicated page with Product schema: a descriptive and semantic name, a specific description, brand (yours), material or category, an aggregateRating if you have reviews, offers if applicable.
- Verify each one with the Rich Results Test. It must show “Product” as the detected type, with no errors.
- Link each Product to the Organization page with `brand` pointing to your brand entity. That way the engine understands the relationship “this line is produced by this company” and not two disconnected things.
- If you have a Wikidata profile for the company, consider adding items for the best-known lines too. It’s not mandatory and only makes sense if they have independent press visibility.
- Compare yourself with the 3-5 competitors the AI cites in your industry when you run the queries above: almost always you’ll find they already have this structure.
This work fits together with the one on author entity recognition and on E-E-A-T for AI: three levels of entity — person, company, product — that talk to each other in the engine’s graph.
Why this changes your visibility in AI answers
Having well-made Product schema doesn’t make you appear everywhere overnight, and on its own it isn’t enough: the company already needs to have authority, the content needs to be consistent, and the site structure needs to hold up. But it shifts the threshold for entry.
As long as you exist as a single undifferentiated block, AI has only one handle to cite you by — your name. When each of your lines becomes an entity with its own attributes, the handles become as many as your products, and each one can answer different queries. It’s the difference between being a closed book on a shelf and having every chapter indexed.
