AI Is the New Chooser. Brands It Can't Tell Apart Don't Get Named.

Ask ChatGPT to recommend a clinic in your city. This is how AI recommends brands now: you get two names, sometimes three. Ask again tomorrow and one of them changes. What never changes: the answer is short, it is confident, and almost every brand in the category is not in it.

The answer is the new front page, and there is no page two.

The short version: Search matched keywords and let a human choose from ten links. AI reads meaning and chooses for the human: it names two or three brands, and it can only name what it can tell apart. Because nearly every brand now produces content with the same AI models, most brands read as interchangeable, so the machine averages them out. The fix is not more content. It is engineered distinctiveness: measured against your competitors, built off your own domain, watched daily.

The machine used to match. Now it chooses.

For twenty years, discovery was a matching game. Google read keywords, returned ten links, and a human did the choosing. The mechanics were public and the game was mechanical: more pages, better keywords, more links. Volume won, and ten seats meant nearly everyone kept some traffic.

The machine in front of buyers now does something different. ChatGPT, Perplexity, Gemini and Google's AI Mode read meaning, compare everything they retrieved, and answer. They do not rank. They choose.

The old game (search)	The new game (answers)
Matched keywords	Reads meaning
Ten links, the human chooses	Two or three names, the machine chooses
Volume and backlinks won	Distinctiveness wins
Rank #8 still got some clicks	No page two

The numbers say the handover already happened. Around 68% of Google searches end without a single click, and 83% when an AI Overview appears (Pew measured this across 68,879 real searches). Buyers moved with the machines: 58% of shoppers research with AI tools, and by March 2026 traffic referred by AI assistants converted 42% better than Google, after converting 38% worse just twelve months earlier.

The channel that rewarded volume is emptying. The one replacing it is brutally concentrated: 3% of brands capture 71% of AI recommendations, and in most categories the citations collapse into three to five names.

The machine can only choose what it can tell apart

Why so few names? A reasoning machine has one requirement the old machine never had: it must distinguish you before it can name you.

That is the requirement most brands now fail, and the cause is the tool everyone adopted at once. Nearly every brand produces content with the same handful of AI models, and a language model predicts the most probable next words. It writes from the centre of everything ever written on the subject. I wrote about the mechanism in AI Is a Machine for the Average: the model is not pulling your content toward wrong, it is pulling it toward the middle, and the middle is where every competitor already sits.

Run an entire category on the same models and the brands converge into one voice. This is measured, not asserted. Shaw and Nave at Wharton showed, across 1,372 participants, that people adopt AI output without engaging their own judgment. In Science Advances, Anil Doshi and Oliver Hauser put the collective cost plainly: "generative AI–enabled stories are more similar to each other than stories by humans alone." Each writer improved; the group converged. The sea of sameness is a measurement, not a metaphor. AI can average your voice; it cannot evolve it.

So the chain closes: everyone writes with the same machine, every brand drifts to the same centre, the choosing machine cannot tell them apart, and it names the few it can. Sounding the same as your category used to cost you style points. Now it costs you existence.

The obvious fix makes it worse

The instinctive response is to produce more. It backfires twice.

First, more AI-default content means more centre, so you converge faster.

Second, your own site is the wrong battlefield. Roughly 85% of what AI assistants say about brands comes from third-party sources: communities, reviews, press. We audited a real-estate brokerage with 31,725 indexed pages, an enormous production operation, and found near-zero presence on the community surfaces AI trusts most. All that publishing, and the machine formed its opinion elsewhere.

The tool market doesn't fix this either. Content generators manufacture more sameness by design. Visibility trackers tell you whether you were mentioned: the symptom, never the cause. Being retrieved is not being chosen.

What works: engineer the difference, then hold it

If the machine chooses on distinctiveness, distinctiveness stops being a branding preference and becomes an engineering requirement. That is the discipline we built at Ivanooo. We call it Distinctiveness Engineering, and it has three moves:

1. Measure. Strip your logo from your content and test whether a machine can still tell it's you. Most brands fail. The Brand Distinctiveness Index turns that test into a 0-100 score against your real competitors, with the evidence attached: the phrases you share with them, the proof you're missing, the places you've been averaged into your category.
2. Build. Construct the three layers the machine selects on. Entity: the identity it resolves by name. Voice: the language no competitor shares. Topic Authority: the claims and proof only you can make. Then place them where the machine reads, which is mostly not your own site.
3. Hold. The pull toward the middle never stops, and teams slide back to the machine default without noticing. Each piece looks fine; the decay only shows across the body of work. Daily observability catches convergence the day it starts. That is the difference between a one-time rebrand and a durable position.

The question to ask this week

Not "how much content are we shipping," and not "are we mentioned." Ask the chooser's question: if our logo came off everything we published, could a machine still tell it was us?

If the answer is no, the machine standing in front of your buyers cannot name you, and in an answer with three seats, unnamed is invisible. Find out where you stand: paste your URL, get your weaknesses ranked, with the evidence. No call required.

FAQ

Why does AI recommend so few brands? An answer engine reads everything it retrieves, then names only what it can distinguish and trust. In most categories citations concentrate into three to five names, and 3% of brands take 71% of recommendations.

Why doesn't producing more content improve AI visibility? AI-produced content converges toward the category average, so more of it makes you more interchangeable. And about 85% of what AI says about brands comes from third-party sources, so your own domain was never the main battlefield.

What is Distinctiveness Engineering? The discipline of making a brand measurably distinguishable and attributable to the machines that decide recommendations: measure your distance from the category average, build the Entity, Voice and Topic Authority layers, then monitor daily so the distinctiveness holds.

What is the Brand Distinctiveness Index? A 0-100 score of how distinguishable your brand is to AI, computed against your real competitors, with cited evidence behind every component. The blind test at its heart: with the logo stripped, can a machine still attribute your content to you?

How is this different from AEO or AI-visibility tools? Trackers measure the symptom: whether you were mentioned. Distinctiveness Engineering measures and fixes the cause: whether the machine can tell you apart at all. Retrieved is not chosen.

Can't I just prompt the AI to be more original? No. Naming the edge relocates the machine to the average of the edge: ask for "original" and you get the statistical centre of everything labelled original. Distinctiveness has to be engineered from human judgment, then held.