makefield

Week 3 · the first dimension

Recognition: does AI know you exist?

Before an assistant can describe you well, it has to recognise you as a specific thing. Most visibility problems start here — and so does the fix.

Makefield · Berlin

The first question

Before anything else, recognition

Every other part of AI visibility depends on one thing being true first: the model has to know that you are a specific, distinct entity — a company, a product — and not a vague phrase or someone else's name. This is recognition, the first dimension, and it's the floor everything else stands on. There's no point worrying whether an assistant describes you accurately if it doesn't recognise you as a thing to describe at all.

Recognition has a precise shape. A model recognises you when it can place you as a named entity, sort you into the right category, and tell you apart from the things you could be confused with. Miss any of those and you're not yet visible — you're noise.

The namesake problem

When the AI knows other things by your name

The clearest way to see recognition failing is a new or uncommon name. I ran this on my own: I asked six AI assistants about a brand-new name with almost no web presence yet. Between them they confidently returned five completely different, unrelated things — townships, a golf brand, an insurance agency — and none of them were the company I was asking about.

That's not the model being wrong, exactly. It's the model doing its best with a name it doesn't yet recognise as a distinct entity: it reaches for whatever it does associate with those letters. For a new brand, this is the starting position — you don't begin invisible so much as collided, sharing your name with whatever the web already knows. Recognition is the work of becoming the thing the model means when it sees your name.

What recognition looks like

Named, categorised, distinct

When recognition is working, three things hold at once. The model names you without prompting — you come up when someone asks about your category, not only when they type your exact name. It categorises you correctly — a B2B analytics tool, not "some software," and not your competitor's category. And it distinguishes you — from your namesakes, and from the other companies in your space, so a description of you is actually about you.

If any of those slips, the symptom shows up downstream: you're left out of shortlists, lumped in with the wrong group, or described with facts that belong to someone else.

How recognition is built

Consistency, identity, corroboration

Recognition isn't won with a trick. It's built the way a careful researcher would build an understanding of you:

Why it's the floor

You can't correct an answer that was never about you

This is why recognition comes first in the method. Accuracy, fair competitor framing, recommendation — every later dimension assumes the model already knows who you are. Fix accuracy before recognition and you're correcting a description the model isn't even attaching to you. Get recognition right and the rest of the work finally has something to hold onto.

Next — Accuracy: once the model knows you exist, is what it says about you actually true? New here? Start with the overview.

Common questions

Questions this lesson answers

How do I tell if AI knows my company exists?

Don't test it by typing your exact name — a model can echo your name back without recognising you as a distinct thing. Test it the way a buyer would find you: ask several assistants about your category ("who makes B2B analytics tools for logistics?") and see whether you come up unprompted, sorted into the right category, and told apart from companies you could be confused with. If you only appear when someone types your name letter-for-letter, and never when they describe what you do, the model is matching a string, not recognising an entity. Run it across a few assistants, not one — recognition varies by model.

Why does AI confuse my company with something else that shares our name?

Because a name the model doesn't yet recognise as a distinct entity gets filled in with whatever it already associates with those letters. I ran this on a brand-new name and six assistants returned five different unrelated things — townships, a golf brand, an insurance agency — none of them the company I asked about. A new or uncommon brand doesn't start invisible so much as collided, sharing its name with whatever the web already knows. The fix isn't to correct each wrong answer; it's to become the thing the model means when it sees your name, by building a consistent account of who you are across sources.

Isn't the AI returning my name proof that it recognises me?

No. Returning the letters of your name is not the same as recognising you as a specific entity. Recognition has three parts that have to hold at once: the model names you without prompting when someone asks about your category, categorises you correctly (a B2B analytics tool, not "some software"), and distinguishes you from your namesakes and competitors. A model can produce your name while getting all three wrong — describing someone else's company, or filing you under the wrong category. Name-matching is the surface; recognition is whether the description is actually about you.

How do I get an AI assistant to recognise my brand?

Not with a trick or a single tag. Recognition is built the way a careful researcher would build an understanding of you: a consistent account of who you are, repeated across many sources so it becomes the model's default; an unambiguous, machine-readable statement of your name, category, and what you make, so the model isn't guessing from fragments; and corroboration beyond your own site, because the account firms up when it appears in places other than your homepage. When the web's account of you is thin or contradictory, the model has nothing stable to recognise — so the work is making that account clear and consistent everywhere you appear.