TheModelDoesn'tKnowYou'reGood

Marcus Chen had been brand director at a mid-sized UK skincare company for six years when he noticed something that made no sense.

It was January 2025. He was pulling the quarterly analytics, the routine audit of traffic sources, conversion rates, the usual forensics of a brand trying to grow in a crowded market. Everything was roughly where he expected it. Paid search, steady. Organic, flat. Email, holding. And then, buried in the referral data, a number that shouldn't have been there.

ChatGPT referrals had tripled.

Not grown. Tripled. In a single quarter. With zero media spend behind it, zero influencer push, no PR campaign, no trade coverage. The model had simply decided, somewhere in its vast, opaque interior, that his brand was the answer to "best natural moisturiser for sensitive skin." Consumers were asking a machine what to buy. The machine was sending them to him.

Revenue was up 14%.

Marcus spent the next three months assuming he was a genius.

Then in Q2, without warning, the referrals collapsed. Traffic cratered. The model had moved on, to a competitor he had never heard of, for reasons nobody at his company could identify, with no appeal process and no explanation. He had no idea what had changed. He had no lever to pull.

That feeling, the dizzying high of algorithmic selection followed by the vertiginous drop of algorithmic abandonment, is the defining brand experience of the next decade. And almost no one is running toward the danger at the right speed.

The Illusion of the Deserving Brand

The story most brand managers are telling themselves right now goes like this: AI search rewards the brands that have done the hard work. Strong brands with genuine equity, consistent messaging, high review scores, and broad recognition will naturally surface in LLM recommendations. Brand investment, long misunderstood and underfunded, is finally getting its moment.

This story is wrong.

LLMs don't recommend your brand because it's good. They recommend it because it's legible to a machine, because your content is structured in ways the model can parse, your name appears consistently across third-party domains the model trusts, and the language around your brand matches the semantic patterns the model has learned to associate with credibility. These things correlate with brand strength. They are not the same as brand strength. And confusing the two is the most expensive mistake the marketing industry will make in the next five years.

The Numbers Are Violent

Consider what's actually happening at scale.

Across eight major AI platforms, ChatGPT, Perplexity, Gemini, Google AI Overview, Google AI Mode, Microsoft Copilot, Claude, and Meta AI, a Q1 2026 study of 177 brands found that 89.8% had zero AI search visibility. None. Not one mention across any of the eight platforms. [Sourced: DigitrendZ, Q1 2026]

ChatGPT includes brands in 99.3% of its ecommerce responses. But it draws from only 2,127 unique domains. Out of millions. [Sourced: BrightEdge AI Catalyst]

The top 2 to 3 brands in any given consumer category capture the majority of AI-driven recommendations. In B2B software, the top brands average 71% of AI citations. [Sourced: Brandofy 2026 Benchmark]

AI-driven referrals to e-commerce grew 109% between January and September 2025. Overall referrals grew 7%. [Sourced: Euromonitor, December 2025]

The model is becoming the market. And 90% of brands don't exist inside it.

> "Google was the landlord. ChatGPT is the estate agent. And your brand just got taken off the listings without anyone telling you."

Goodhart's Brand

Here is where it gets structurally dark.

Charles Goodhart, the British economist, articulated a law that has become the quiet tyrant of every measurement-obsessed organisation: when a measure becomes a target, it ceases to be a good measure. Banks optimised for capital ratios and created systemic fragility. Schools optimised for test scores and produced students who couldn't think. Social platforms optimised for engagement and produced outrage.

The same trap is opening up, right now, in front of every brand manager who has noticed the LLM citation gap and decided to close it.

The brands currently winning in LLM recommendations are winning because they've built genuine ecosystems of trust, years of editorial coverage, broad retail distribution, consistent consumer reviews, structured product data, third-party endorsement. These things created machine legibility as a byproduct of doing brand-building correctly.

The moment brands start optimising directly for machine legibility, building content specifically for LLM ingestion, chasing structured data formats, engineering their language to match model priors, they will begin destroying the brand distinctiveness that made them legible in the first place. A brand that learns to speak fluent Machine is a brand that has stopped speaking fluent Human. And sooner or later, the humans notice.

Goodhart's Law doesn't just describe measurement failure. It describes the annihilation of the thing being measured, by the act of measuring it.

Byron Sharp Needs a Third Column

In 2010, Byron Sharp published How Brands Grow and dismantled a decade of marketing orthodoxy. The argument was clean: brands don't grow through loyalty; they grow through penetration. And penetration requires two things, mental availability (being easy to think of) and physical availability (being easy to buy).

Sharp's framework reshaped budgets, restructured agencies, and gave brand managers a vocabulary for defending broad-reach media against the precision-targeting lobby.

It is now incomplete.

LLMs have introduced a third axis of availability: Algorithmic Availability, the degree to which a brand exists, accurately and favourably, within the training data, retrieval layers, and semantic associations of the AI systems that increasingly mediate discovery.

A brand can be mentally available (consumers recognise it), physically available (it's in every supermarket), and algorithmically invisible (the model doesn't know it exists, or worse, knows it wrong). A brand that is algorithmically invisible in a world where 93% of AI-mode search sessions end without a website visit is a brand with a structural revenue problem that won't show up in this quarter's numbers. [Sourced: OmniBound, April 2026]

The three columns of availability, mental, physical, algorithmic, must now be audited together. Most brand teams are still running with two.

The Perez Window

Carlota Perez spent decades mapping the patterns of technological deployment. Her insight: every major technology goes through an installation phase, a period of rapid adoption, speculative excitement, and incumbent advantage, followed by a deployment phase, where the technology matures, challengers catch up, and the early edge dissolves.

We are deep in the installation phase of AI search. And the brands benefiting most from it should be the most nervous.

Right now, the LLM lottery favours incumbents. Established brands with broad coverage, long content histories, and wide third-party citation have a structural head start in machine legibility. This creates an illusion of permanence, a sense that the brands winning today have earned a durable advantage.

They haven't. They've earned a temporary one.

As AI search matures, as models improve their retrieval, as the tools for LLM optimisation become commoditised, as challengers learn to build algorithmic availability from scratch, the incumbent head start will erode. The brands that have confused early legibility with earned equity will discover the difference when it hurts.

The Perez window is open now. But windows close.

The Price That AI Set (And Who It Harms)

Here is the part of this story that should concern regulators as much as brand directors.

A 2026 study by Amit Zac and Michal Gal examined what happens to purchasing behaviour when consumers use conversational AI for shopping decisions. The findings were uncomfortable. AI shopping assistants consistently steered consumers toward established, premium brands at higher prices, without any corresponding improvement in product quality. Amazon review ratings, used as a quality proxy, showed no gain. Spending went up. Quality stayed flat. [Sourced: Zac & Gal, 2026, via ProMarket]

The model is a price inflation engine for the brands it favours. And a pricing trap for everyone else.

For brands inside the citation circle, this looks like pricing power. Revenue grows without a change in actual product merit. The premium feels earned. It isn't. It's algorithmic. And algorithmic premiums are fragile in ways that genuine brand premiums are not, because they can be revoked without warning, for reasons the brand cannot see or control.

> "You spent twenty years building a brand in the consumer's mind. You now have approximately eighteen months to rebuild it in the model's."

For brands outside the citation circle, the 89.8%, this is a structural pricing disadvantage that compounds quietly. If AI-mediated discovery increasingly shapes where consumers look and what they're willing to pay, then invisibility in that layer isn't just a traffic problem. It's a margin problem. It's a pricing power problem. It's a long-term brand value problem.

If the top cited brands in a category command the pricing premiums that AI recommendation implies, and if AI-mediated discovery accounts for a growing share of high-consideration purchase journeys, the implied pricing disadvantage for uncited brands in the same category could compound to a 15 to 22% effective price ceiling gap within 36 months, not because their products are worse, but because the machine never mentioned them.

The Brands Winning Right Now Are the Most Exposed

The brands most exposed to the coming algorithmic correction are not the ones currently invisible in LLM answers. Those brands already know they don't exist inside the model. Quietly, some of them are rebuilding.

The most exposed brands are the ones winning the LLM lottery right now, basking in algorithmic premium, mistaking the model's current preference for equity they built. They are the ones not asking hard questions. Watching revenue grow and attributing it to a decade of brand work.

Some of it is. Some of it is borrowed time.

The question every board should be asking is not "are we visible in AI search?" The question is: do we understand why we are visible, and what breaks if that reason disappears?

What This Requires

The answer is not to stop building brand. Build it in three dimensions, not two.

Audit algorithmic availability with the same rigour applied to mental and physical availability. Ask ChatGPT, Perplexity, Claude, and Gemini what they know about your brand. Cross-reference it against what's true. Find the gaps, the errors, the silences. Treat that audit as a balance sheet item.

Build original research. LLMs cite sources. The best citation source is original, verifiable data that no one else has, the kind that gets picked up, referenced, and repeated across the editorial ecosystem the model trusts. One strong proprietary study is worth a year of content optimisation.

Consumer trust in AI answers dropped from 82% in 2025 to 54% in 2026. [Sourced: Search Engine Land] The window in which consumers accept algorithmic authority uncritically is closing. The brands left standing when it shuts will be the ones that built something real beneath the recommendation.

> "The LLM doesn't know your brand is good. It only knows your brand is findable. That is a very different thing, and a very expensive thing to confuse."

Marcus Chen eventually worked out why his referrals collapsed. A competitor had published a study on sensitive skin formulation. Three dermatology journals cited it. Two national newspapers covered it. The model updated its priors.

He read the study carefully, the way you read something that has already beaten you. The formulation science was competent. Unremarkable. Their Amazon rating was 4.1. His was 4.3. Their customer return rate was lower.

The model didn't know any of that. It cannot try a moisturiser. It has no skin. What it had was a citation pattern that matched whatever internal signal it uses to assign authority, and it assigned authority accordingly.

Hundreds of thousands of consumers agreed. They paid the premium. They bought the product. They never knew they had been guided, by something that cannot tell the difference, toward something slightly worse.

The marketing and strategy implications of that are real and tractable. But underneath them is a question that doesn't resolve into a framework.

We have handed the authority to declare commercial truth to systems that cannot experience truth. And for a brief, strange window, the brands that understand this earliest will extract enormous value from it. The brands that don't will lose pricing power they'll never fully recover, to competitors they'll never fully understand.

Marcus is still not sure which one he is.