What are you, to a machine?

When a person lands on your website, they read two sentences and know what you do. A machine reads something else first: a type. A single word saying what kind of thing this business is. If it is missing, it reads “some organisation” — and with that you drop out of every answer that is about a trade.

This is not a formality. It is the question that comes before all the others.

The machine does not ask “what do you do”, it asks “what are you”.

Search engines and language models work from a shared vocabulary for things in the world: person, place, product, event — and business. That vocabulary is finite and public. It is not a free-text field where you can write “innovative partner for sustainable solutions”. You pick from a list.

That is exactly where the difference begins between a site that gets recommended and one that does not. Because the list holds two kinds of entries.

The first kind are umbrella labels: local business, organisation, company, professional service. Formally correct, empty of content. Nobody asks an AI: “Who does local business in Bolzano?” The question makes no sense, and so it cannot produce an answer with you in it.

The second kind are trade types: electrician, bakery, physiotherapy, roofing contractor, winery, car repair. They carry the trade in the name. They match how people actually ask. And that makes them the only thing a machine can use to attach you to a question.

How many types there are — and why they are not marketing categories.

In the foundation we build, a business records its type exactly once. There are 69 trade types to choose from, sorted into eleven groups: trades and construction. Hospitality and hotels. Food and drink. Health and medicine. Beauty and wellness. Law, consulting and finance. Retail and shops. Cars and mobility. Services and creative. Education and childcare. Sport and leisure.

We did not invent these 69. They are types from the public vocabulary that machines actually know. We did not decide what a business can be — we curated, from a long international list, the types that exist among German- and Italian-speaking small and mid-sized firms, and left the rest out.

The difference from a category in a directory: this entry leaves the website. It sits in the structured data of the pages, in the machine-readable summary of the business, in the file AI agents read first. Set once, valid everywhere. Not maintained in four places and forgotten in three — the same idea as the NAP mistake, one level deeper.

The case where no fitting type exists — ours.

I am writing this deliberately, because it is the more honest story: our own business sits on an umbrella label. Not out of neglect, but because for what we do, that list holds no trade type. Software foundations for businesses simply are not covered.

It is not a rare case, which is why we handled it in the system instead of looking away. There is a fixed list of these empty labels. If a business carries one, the trade-based match does not apply — instead the connection is made through the company name rather than the trade. Not as good as a real type, but honest: better to be found under your own name than under a question you are not an answer to.

For most businesses, though, that is not the situation. Anyone who installs electrics, bakes bread or rents rooms will find their type in the list — and far too often leaves it empty.

One type, several truths.

Businesses are rarely just one thing. A hotel with a restaurant is both. A bakery with a café is both. So the entry takes several values: you can set multiple types, and the first counts as primary. It decides what you appear as when there is room for only one word.

That order is a real decision, not busywork. A business that lives ninety per cent off its kitchen and lists “hotel” first is offered to the machine as a hotel — and is not considered in restaurant questions. Whatever comes first is what you want to be when someone asks.

Where that one word travels.

The reason the type carries so much leverage is its reach. It does not sit in a field; it appears in four places at once, without anyone copying it there.

• In the structured data of every page. That is the part search engines read before they look at the running text.
• In the machine-readable summary of the business. A compact file describing a firm in facts rather than sentences — name, type, place, contact, services.
• In the file AI agents read first when they visit a domain and want to know what it is about.
• In the context your own site assistant answers from. Someone asking “do you also do…” gets an answer that knows the type.

That is the difference between an entry and a source. An entry sits in one place. A source acts everywhere something about the business is produced — and when it changes, every output changes with it. Open a second line of work and you add a type and are done; nobody has to update four files afterwards and forget one of them.

Type, place and spelling belong together.

A trade type on its own is not yet a recommendation. A real question to an AI is almost never just “Who does electrical work?” but “Who does electrical work in Bolzano?” — and often in another language. Three things that mean little apart and much together:

The type says what you do. The place — address, service radius, municipalities served — says where the answer holds. And the spellings say which words find both; on that, A machine reads literally explains why a machine will not connect “electrician” and “electrical contractor” by itself.

If one of the three is missing, the chain breaks at a point you cannot see from outside. The business is reachable but cannot be matched to a question. It can be matched, but without a place. Or it fits, only under a word nobody types. In all three cases the website looks perfectly fine.

How these entries eventually form a picture of the business that machines can query is described in What is a knowledge graph?.

What you can check yourself, today.

Three questions, no tools needed:

1. Is there a type on your website at all? Not as text in the imprint — as structured data. If you cannot say, the answer is usually no.
2. Is it a trade type or an umbrella label? Does it contain the word a customer would use for your work, or something general like “company”?
3. If you have several lines of work — is the most important one first?

Answer those three honestly and you have found one of the most common reasons a business looks technically flawless and still never turns up in AI answers.

Why this comes before everything else.

Working on copy is tempting. Copy is visible, and you can show it to people. The type is invisible, it is a single word, and nobody thanks you for it.

But it comes before everything. A machine that does not know what you are has no place to file your texts. It reads them, matches them to no question, and forgets them. How that reading and splitting works in detail I described in How an AI reads your website; the overall technical architecture is in From SEO to GEO.

Setting the type takes two minutes. The difference is whether you are among the options at all.

How the vocabulary this type comes from actually works is in The vocabulary machines share.