You show up in AI answers, but in the wrong context: your restaurant is presented as a chef, your company as a private individual. It is not a random error: the AI has classified you in the wrong category, and that classification decides which questions make people find you — and which do not. You are losing visibility on every query where the right customers would be looking for you. Fixing the classification takes less than an hour, with no technical skills.
Your brand shows up in AI answers but in the wrong context. Ask Perplexity “best Michelin-starred restaurants in Bologna” and it names you. Ask “famous Bologna chefs” and it names you again, but confuses your restaurant with your chef name. Ask “gastronomic experiences in Emilia” and your venue disappears, while your personal name reappears as a chef, with no link to the restaurant.
Only one thing is happening: the AI has classified you as PERSON instead of ORG. Or it has split you into two disconnected entities. And as long as the entity type is wrong, you will show up in the wrong answers, to the wrong queries, with the wrong copy. Let me explain why this happens and how to tell whether it is happening to you.
What entity type classification is
When an AI engine reads your name, the first step is not deciding whether you are “good” or “authoritative”. It happens much earlier: deciding what you are. A person, a company, a product, a place, an event, a concept. This choice is called entity type classification, and it shapes everything the AI does afterwards with your brand.
If the AI marks you as ORG (organization), you fall into queries like “best restaurants”, “book”, “order”, “where to eat in Bologna”. If it marks you as PERSON (person), you fall into queries like “who is”, “biography”, “resume”, “books by”. If it marks you as PRODUCT (product), it is even worse: you show up in “where to buy” or “price of”, which for a Michelin-starred restaurant are meaningless queries.
In the world of research on entity recognition with LLMs, the classification mechanism is documented explicitly. Kim et al. (2024) describe how the first generations of NER systems treated each entity type in isolation:
“However, a critical limitation of the MLC ing techniques significantly improves LLMs’ approach is its requirement for constructing a sepa- understanding and processing of nested enti- rate model for each entity type, hindering its ability ties, marking a departure from conventional to recognize nested entities of the same type.”
Translated: previously you needed separate models for each entity type (one for people, one for companies, one for places). Now LLMs can recognize different types and even nested entities (e.g. “Osteria Francescana Restaurant in Modena”, where Modena is a place inside a company name) in a single pass.
For you this means something practical: the AI is doing the job of distinguishing “you the person” from “you the company” simultaneously, looking at the same text. If your site, your listings, Wikidata, your articles push in different directions, the AI gets confused. And when it gets confused, it picks the most likely classification given what it finds — which is often not the one you want.
Why it sits upstream of everything else
Entity type classification comes before all the other things I talk about in this series. Before the Named Entity Recognition that decides whether you are a recognized entity. Before the entity linking that connects your name to the right record. Before the entry in the Google Knowledge Graph.
Think it through: if the AI classifies you as PERSON, the linking system will never look for your company record on Wikidata. It will look for your personal record. If that one does not exist or is incomplete, the link breaks, and you become invisible for every query that concerns the restaurant as an organization.
Same thing for how the AI reads your content. In my previous articles I explained how brand representation works in the vector space: you are placed near other similar brands. But “similar” with respect to type. If you are a chef-person, you end up near Cracco or Bottura-the-person, not near Osteria Francescana or Al Cambio. The queries “where to eat in Bologna” fish in the group of restaurants, and you disappear.
“Massimo Rossi Restaurant” — the AI reads “Massimo Rossi”, classifies it as PERSON, ignores “Restaurant” as noise.
Why declaring it in one place is not enough
The paper by Kim et al. (2024) makes another useful observation:
“In contrast, is common to both GPT-NER and our study. our proposed instruction allows the identification and recognition of all entity types in a single exe- GPT-NER’s instruction (for DNA entity cution.”
Recent approaches evaluate all entity types together, not one at a time. It follows that when Perplexity reads “Osteria del Cappello, Bologna”, it evaluates all the hypotheses simultaneously and picks the type with the most coherent signals. If your homepage uses `Restaurant` schema, your Instagram bio says “Bologna chef”, your Google Business name is your surname, and you have a Wikipedia entry as a person — you have four contradictory signals, and the most frequent one wins, which is usually PERSON.
In the site’s editorial content, name the restaurant in full within the first 100 words of every page, and the chef as a role (“Leading the kitchen, [name]”), not as a synonym for the venue.
The 15-minute test to find out how the AI classifies you
Here are the steps. They are deliberately entry level: a serious diagnosis requires professional tools, but these three steps already tell you whether you have a problem.
First step: Google Knowledge Graph. Open Google and search for the exact name of your brand. If the “knowledge panel” appears on the right (the box with photo and data), look at what it says under the name: “Restaurant”, “Chef”, “Writer”, “Company”. If the panel does not appear, search for your brand on Wikidata and see whether an entry exists. If it does, open the record and check the “instance of” property: it should be “restaurant” or “business”, not “human”.
Second step: Rich Results Test. Go to Google’s Rich Results Test, paste your homepage URL, and click “Test URL”. In the result, look for the “Organization” or “Restaurant” section. If there is nothing, the AI has never received an explicit entity type signal from you. If there is “Person” but no “Organization”, you are telling Google that you are a person, not a company.
Third step: the conversational query. Open ChatGPT or Perplexity and ask three questions in a row: “who is [brand name]?”, “what does [brand name] do?”, “where is [brand name] located?”. Read how the AI presents you in the first words. If it answers “[name] is an Italian chef…”, you are PERSON. If it answers “[name] is a Michelin-starred restaurant in Bologna…”, you are ORG. If the three answers are inconsistent with each other (one says chef, one says restaurant, one says “I can’t find information”), you have an unstable classification, which is worse than a wrong one.
Binary threshold for the decision: if all three tests give you the type you want (ORG for a restaurant, ORG for a company, PERSON for an individual consultant), you are fine. If even one of them gives you the wrong type, the problem exists.
The errors I see most often in high-end dining
Working with Michelin-starred restaurants, historic trattorias, and Emilia-Romagna wineries, these are the patterns that recur:
- The chef’s name coincides with the venue’s name. “Massimo Rossi Restaurant” — the AI reads “Massimo Rossi”, classifies it as PERSON, ignores “Restaurant” as noise. Solution: on the site and the listings, the venue name must be treated as a separate entity, with `Restaurant` schema and a `founder` pointing to the person. They are two distinct entities connected by a relationship, not a single one.
- Wikidata has the chef’s entry but not the restaurant’s. Typical for starred chefs: journalists write about them, wiki-editors create the person record. The restaurant stays without an entry. Result: the AI has a solid entity for you-the-person and zero for you-the-company. Solution: request the creation of the Wikidata entry for the restaurant, with the property “instance of: restaurant” and the relationship “owned by: [you-the-person]”.
- Inconsistent schema markup across pages. The home has `LocalBusiness` schema, the “chef” page has `Person` schema, the “menu” page has no schema. The AI samples the pages and sees different types. Solution: a single dominant schema for the site (Restaurant for the venue), with Person as a sub-entity linked via `employee` or `founder`.
- Google Business Profile with the wrong category. Primary category “Chef” instead of “Italian restaurant” or “Fine dining restaurant”. The primary category on Google Business is one of the strongest signals for the entity type. Solution: check and fix it.
What to do, in order
These are the operational steps, from the intervention that moves the needle most to the one that fine-tunes:
- Define a single primary entity you want to get classified correctly (the restaurant, not the chef) and one or more secondary entities linked to it.
- Align the Google Business Profile with the correct category.
- Add `Restaurant` schema on the homepage with a stable `@id`, `name`, `address`, and `founder` pointing to the person.
- Create or fix the restaurant’s Wikidata entry with “instance of: restaurant” and the relationship to the person.
- In the site’s editorial content, name the restaurant in full within the first 100 words of every page, and the chef as a role (“Leading the kitchen, [name]”), not as a synonym for the venue.
- Repeat the three tests every three months: Knowledge Graph, Rich Results Test, conversational query.
Keep in mind that this is not a magic fix. Entity type classification stabilizes over months, not days. The AI has to see the new signals, compare them with the old ones it already had, and converge on the new type. If you have a 10-year history of articles presenting you as a chef-person, you will not change the classification in three weeks.
How it fits with the rest of your AI visibility
Getting the entity type right is the premise for everything you can do afterwards for visibility in AI answers. If you are misclassified, every E-E-A-T for AI effort applies to the wrong entity. Every author citation you earn flows into the wrong place.
In the following articles we go into detail: how entity linking works, how to obtain an entry in the Google Knowledge Graph, how to map the relationships between entities. But it all starts here: the AI must know what you are before it can know how much you are worth. Compare your knowledge panel with that of the 3-5 restaurants Perplexity cites for “best starred restaurants in Emilia Romagna”: if they have “Restaurant” and you have “Chef” or nothing, you have one of the explanations for why you do not show up.
