AI knows Duolingo. It doesn't know Duolingo can teach you chess.

Everyone asks me how to rank in AI answers. It's the wrong question — nobody asks how AI decided what they are in the first place.

Market perception used to be a soft science. It lived in surveys and focus groups, spread across a million heads, and no single version of it could be wrong. Now there's a version that's written down. One machine, one stored answer, stated as fact to every buyer who asks. Your brand perception has a source of truth now — and most of the time, it's wrong.

We checked. Across 24,488 brands, 70.4% were misperceived by AI in at least one way. Not stylistic drift — actual factual error. 17.8% were seriously wrong. 5.5% were fundamentally wrong, as in the AI described a company that no longer resembles the one you'd find if you looked it up yourself.

It's not AI being generally sloppy — it's AI being confidently, specifically wrong in predictable ways. Established brands that rebranded or launched new products: AI still repeats their old messaging, so researching buyers never know. New brands trying to break in: AI has no record of them and fills the gap with something else — a different company, a generic definition, or an "I don't know."

There's a reason this happens. AI learns from a snapshot of the internet at training time, then stops. The brand keeps living. The model doesn't update.

Everyone's answer to this is the same: just make AI search the web. Get crawled, get cited, get into the sources. We tested that too. On 12,782 scored answers with verified ground-truth data: from memory, AI misrepresented brands ~75% of the time on a per-answer basis. With live web search enabled, that dropped to about 25%. One in four answers still wrong even when the model searched first — because search gets you found, and the model still describes the wrong you. Being in the sources is not the same as being perceived correctly.

Misperception comes in two flavors. Your brand is either miscategorized — AI knows you, but has you in the wrong aisle — or misidentified — AI doesn't know you, and describes someone else wearing your name.

One hurts brands you've heard of. The other hurts brands you haven't.

Duolingo does not have a visibility problem. It might be the most visible brand in AI answers about learning anything. Ask about language apps and it ranks first, described glowingly, every model, every time.

Duolingo launched a full chess course in June 2025. Free, on iOS, Android, and web, PvP live. Seven million daily active users. Not a side experiment — a real product in a competitive market.

Ask any AI "what's the best app to learn chess for beginners" and you get Chess.com, Lichess, Dr. Wolf, Chessable. Duolingo doesn't appear — on Gemini or Perplexity, with live search on.

The demand is real. The product is real. The buyer is actively looking. AI perceives Duolingo confidently, warmly, accurately — as a language app. And that's the trap: every additional article, review, and mention of Duolingo-the-language-app makes the model more certain of the category it filed the brand under. Duolingo's massive visibility is actively working against its newest product.

This is why you can't publish your way out. More content, more press, more earned coverage — a Muck Rack study on 25 million AI citations found that 84% come from earned media, so being in the sources matters. But coverage compounds whatever perception already exists. If you're miscategorized, more visibility just makes AI more confidently wrong.

The outcome: Chess.com acquires a customer Duolingo built a product for — every day, at scale.

This one hurts new companies and new product lines trying to break through.

AI can only find you if it can pin your name to you and only you. Without an unambiguous signal, it attaches your name to whatever it knows best — the dictionary definition of the word, or the incumbent with the most weight in its training data. It doesn't hedge. It answers confidently about something else entirely.

I tested this with a small AI software company whose name is also a common English word. "What is Shipper" — across three models, two turns each of detailed freight logistics answers. Not "which Shipper do you mean?" Consignors, linehaul rates, dimensional weight. For a buyer trying to research a software product. In some sessions, when I pushed further, a competitor's name surfaced instead — a real e-commerce shipping platform with enough corpus weight to steal the answer.

Then I used the URL. "What is shipper.now" — the actual domain. Every model came back correct on the first turn. No freight definitions, no wrong company.

Same brand. Two queries. Completely different outcomes across every model.

Notice what the URL did. A domain isn't a ranking signal — it's an identity signal. It gave the model nothing to resolve, so there was nothing to get wrong. No AEO tactic, no content calendar, no authority campaign does that, because those are rank interventions aimed at an identity problem.

Your domain is now part of your brand name — practically, not legally.

The outcome: You are known by your top-level domain (TLD), or you are known by your competitors.

AI has become the first stop in the purchase journey — the one that shapes what buyers expect before they reach your site, store, or sales team. When the answer is wrong, the confusion doesn't land on the AI. It lands on the brand. The buyer doesn't think "AI misled me." They think: that company isn't in this space, that product doesn't exist, I've never heard of them. And they move on.

Because AI queries don't show up in your referral analytics, you may never know it's happening.

Last week I ran an AI-generated brand audit pointed at myself. Its verdict: I appear to have no owned publication…which is ironic because you're reading it. It found my name attached to different handles on different surfaces, and noted it couldn't confirm they were all the same person. My own personal website—which frankly, I had forgotten about—doesn’t mention my company by name (oops).

My problem wasn't presence. I have earned press, a strong LinkedIn, coverage. AI's perception of me just didn't add up to one person. I've spent months documenting misperception across 24,000 brands, and I was one of them.

Here's the honest caveat: if your buyers never ask AI anything — and they buy your products based on shelf recognition or a golf-course handshake — you can close this email. AI search rank anxiety really doesn’t apply to you.

But if you care about what your customers and the market think of you, your brand and your products—then misperception in AI answers does matter. Because whether or not anyone is querying you today, AI has already formed a perception of you — an identity, a category, a story — and it's sitting there, wrong or right, waiting for the first buyer who asks. When the answer is wrong, the confusion doesn't land on the AI. The buyer doesn't think "AI misled me." They think: that company isn't in this space, that product doesn't exist, I've never heard of them. And they move on. AI queries don't show up in your referral analytics, so the loss is silent.

Every tool in the AI-visibility gold rush is measuring the scoreboard — do you show up, do you sound good. Presence and sentiment. Not one error in this piece would trip either metric. Duolingo shows up constantly and sounds great. You can rank first, glow, and be miscategorized under a product you discontinued two years ago.

So don't start by asking where you rank. Ask what AI perceives you to be.

Open ChatGPT, Claude, and Gemini. Ask each one: "What is [your brand]?" Then ask: "What is [yourbrand.com]?"

Screenshot all six answers.

If the two sets match and they're accurate — enjoy it, you're in the minority. If the answers change when you add the domain, you've just watched yourself get misidentified. If they're consistent but describe last year's you, you're miscategorized. Either way, you now know which flavor of misperceived you are — and everything you spend on AI visibility from here will compound it or correct it.

Methodology: The 70.4% divergence figure covers 24,488 brands profiled across AI assistants in Optimly’s AI Brand Index. The 75% / 25% figures are from our grounded audits — 12,782 scored answers with verified ground-truth data.