“Get me milk.”
Three words. No brand name, no preference, no loyalty signal. When a consumer says this to an AI agent instead of “get me Oatly,” the brand loses. Not to a competitor — to an intermediary that has never seen a Super Bowl ad, never felt an emotional connection, and never walked down a grocery aisle.
Some of that shift is already visible, alongside forecasts of further growth. McKinsey estimates AI agentic commerce could redirect $3-5 trillion in global retail spend by 2030, with nearly $1 trillion from the U.S. alone. Forty-five percent of consumers already use AI for at least part of their buying journey, and 60% expect to use agentic AI for purchases within the next twelve months. Among power users, 82% would delegate even high-stakes purchasing decisions to AI agents.
Eighty-one percent of retail executives believe generative AI will weaken brand loyalty by 2027. They are not worried about a new competitor. They are watching the entire architecture of brand power shift from the consumer’s mind to the algorithm’s logic.
For a brand team, the problem goes beyond where to advertise. A business built on recognition and loyalty now has to reach a decision-maker that does not experience either in the human sense.
The Quartz Crisis, in Reverse
In the 1970s, Swiss watchmaking controlled 85% of the global market. Then Seiko introduced the quartz movement, and a $10 watch kept better time than a $10,000 mechanical one. The performance differentiator that justified premium pricing vanished overnight. Swiss market share collapsed to 15% in units. Hundreds of manufacturers went bankrupt.
Paul Graham captured the aftermath: “Brand is what’s left when the substantive differences between products disappear.” The quartz crisis proved him right. Survivors did not compete on accuracy — they pivoted to identity. Nicolas Hayek merged ASUAG and SSIH into the Swatch Group. Patek Philippe deliberately oversized the Nautilus at 42mm versus the golden-age standard of 32-33mm, not for ergonomics, but so it could be recognized across a room. By the mid-1980s, mechanical watches had been repositioned as wealth displays, not timekeeping instruments.
Today the four largest Swiss watch brands control 49.1% market share while capturing 76% of total industry profits. Watches above CHF 50,000 represent 1.4% of units shipped but account for 89% of industry growth.
When technology commoditizes the current value proposition, survivors redefine what they sell. The difference this time is scope. The quartz crisis hit one industry over a decade. AI intermediation hits all industries simultaneously, compressing the adaptation timeline from years to quarters.
The quartz crisis created the Brand Age. AI intermediation may be creating the Post-Brand Age.
Watches went from selling performance to selling identity. AI is now commoditizing identity itself.
The Algorithm Is Already the Storefront
Amazon Rufus has reached 250 million users, handling 274.3 million daily queries. During Black Friday 2025, Rufus was active in 38% of shopping sessions — up 127% from July. Shoppers using Rufus are 60% more likely to complete purchases, putting the assistant on pace for $10 billion or more in incremental annualized sales.
Only 22% of products appearing on Amazon’s traditional first results page coincide with those recommended by Rufus. Brands spent years optimizing for one gatekeeping system. A second, incompatible one has been layered on top.
Rufus recommendations are 83% self-serving and only 32% accurate.
Brands are now optimizing to please a biased judge. The bias itself may erode consumer trust in AI intermediaries over time — but the intermediation trend persists because consumers optimize for convenience, not accuracy. The gatekeeper does not need to be fair to be powerful. It just needs to be there when people are buying.
We have seen this dynamic play out already with food delivery. Which pizza do you order? Often whichever one appears first in your app of choice, preferably under a promotion.
AI traffic to U.S. retail sites increased 805% year-over-year on Black Friday. ChatGPT now drives more than 20% of referral traffic to Walmart and nearly 15% to Target. This is not an Amazon-only story. It is an ecosystem-wide shift — from e-commerce and CPG to travel, where airline CEOs warn that AI agents will determine which carriers secure bookings, to professional services, where legal tech alone is projected to grow from $20.8 billion to $65.5 billion by 2034.
The discovery funnel is collapsing alongside it. When an AI overview appears on Google, 80% of the time there is no click. Gartner projects a 50% decline in organic search traffic by 2028. CTR drops 34-46% when AI summaries appear. HubSpot’s blog traffic fell from 13.5 million to roughly 6 million monthly visits after AI Overviews rolled out. The brand consideration phase — where consumers used to browse, compare, and form preferences — is being compressed into an ask-and-receive model where the AI’s recommendation is the consideration set.
Your Brand Equity Does Not Transfer
AI tools produce different brand recommendations 99% or more of the time across identical queries. The semantic similarity of prompt pairs is only 0.081, meaning even slight rewordings generate substantially different outputs. Carnegie Mellon found that minor prompt changes shift AI brand recommendations by up to 78.3%. Google’s Gemma model went from recommending Instant Pot 0% of the time to 100% based on a single word change.
A CMO who spent $50 million building brand awareness now has another question to answer: will that work influence an assistant’s recommendation? These results suggest the words in the request can outweigh the awareness the team worked to build.
Across platforms, the instability compounds. There is a 62% brand recommendation disagreement between ChatGPT and Google AI. Tesla scores 94 out of 100 on AI visibility in its category. Apple hits 96. In CRM, Salesforce dominates at 92 — while Oracle and SAP barely register in AI recommendations.
A brand can be beloved by millions of consumers and invisible to the algorithm that serves them.
When product attributes are missing from AI-readable data, selection probability drops 20-40%. This is the machine legibility gap, quantified by researchers at Yale, Columbia, and the University of Chicago. The mechanism is structural: AI agents construct their own preference logic from available structured data. Incomplete digital footprints create systematic disadvantage regardless of how much consumers love the product.
Kantar states it plainly: “If the model doesn’t know a brand, it won’t choose it.”
Share of Mind is being replaced by Share of Model. They are not the same metric, and they do not correlate the way brand teams assume.
Where consumers still choose the brand
One limit to this argument is that brand loyalty creates override behavior.
When an AI agent recommends generic oat milk, the committed Oatly buyer corrects the recommendation. Consumers already override algorithmic suggestions much of the time. Fifty percent remain uncomfortable with end-to-end AI transactions, and shoppers trust retailer-built agents three times more than third-party agents. When brand attachment is high, the AI intermediary becomes a convenience layer that the consumer corrects rather than a decision-maker.
This objection is right. However, override behavior works for identity purchases — perhaps five to ten brands per consumer. The average grocery basket contains thirty to fifty items. For the remaining 70-80% of purchases, the consumer has no strong preference. That is where the AI agent decides.
Then there is the luxury question. HBR captures it well: “The purpose of a car like a Lamborghini is to drive it, not be driven in it.” If AI handles the selection, does the status signal survive? A Rolex on someone’s wrist signals taste regardless of who chose it — display value persists. But experiential luxury categories where the selection ritual is the product — fine dining, bespoke tailoring, art collecting — face genuine risk when the discovery process becomes algorithmic. Prada seems to understand this: their “Artisanal Intelligence” framing treats the tension between craft and computation as brand strategy itself, not something to resolve but to inhabit.
For the time being, AI may actually be concentrating brand power rather than eroding it. Tesla’s 94/100 and Apple’s 96/100 AI visibility scores suggest dominant brands are winning the recommendation game. But this reflects AI immaturity, not permanent equilibrium. As agents mature beyond brand-as-heuristic toward attribute-based selection, the early advantage of being a known name will erode. The risk is complacency — assuming that because AI currently favors you, it always will.
Who Survives and Why
When you map brands against two dimensions — how often consumers specify them by name and how much value survives algorithmic comparison — five zones emerge.
Verb Brands (Google it, Uber there, Venmo me): Category-defining names that consumers use as verbs. AI cannot route around them because the brand name is the request.
Identity Brands (Oatly, Patek Philippe, Prada): Consumers specify by name because the brand is part of their identity. The override defense works here — these are the five to ten brands per consumer that survive delegation.
Training-Data Defaults (Zillow, Salesforce): Brands so dominant in AI training data that algorithms default to them. A precarious position — it depends on maintaining data dominance as models retrain.
Habitual Repurchase Brands (the vast middle): Moderate loyalty, repurchased out of habit and shelf visibility. These brands are particularly exposed. AI agents eliminate both habit triggers — shelf placement disappears, and workflow inertia breaks when an algorithm re-evaluates every purchase.
Commodity Brands: No meaningful differentiation. These become white-label suppliers to AI platforms. BCG and Kantar predict the bifurcation: trusted brands negotiate outcomes; distrusted brands become invisible commodities.
To assess that exposure, ask whether a customer would still request your brand after an agent compared price, specifications, reviews, and availability. If the answer is no, you have yet to establish a reason to choose you that survives that comparison.
What to Do About It
Start by checking whether an assistant can find and interpret the information a buyer needs. Brands adopting llms.txt machine-readable formats saw 12-25% increases in AI-generated traffic, and missing attributes cause a 20-40% selection probability drop. If the algorithm cannot read you, it cannot recommend you. This is the table stakes.
Second, build the dual-interface brand: one expression for the human who responds to emotion and story, another for the machine that responds to structured data and attribute completeness. The brands that treat AI as a second audience — not an afterthought — will maintain visibility across both channels.
Third, control the relationship layer so AI intermediation becomes optional rather than mandatory. Own the customer data and the direct channel. When the consumer can reach you without going through an algorithmic gatekeeper, your pricing power survives.
Finally, stress-test through the quartz crisis lens. Swiss market share collapsed from 85% to 15%, yet the survivors built a fortress by finding a dimension of value the disrupting technology could not commoditize. They did not just survive — they captured 76% of industry profits. The brands that find their uncommoditizable dimension will do the same.
Run the AI purchasing simulation this quarter. Ask ChatGPT, Google AI, and Rufus to buy your product category. Measure your Share of Model. The number that comes back may look nothing like your Share of Mind — a gap worth measuring alongside conventional brand awareness.
The quartz crisis killed brands that believed their reputation would carry them through a technological shift. It did not kill the industry. It killed the value proposition.
Start with the request a customer would actually make. See whether the assistant finds your product, what it says about it, and what the customer would need to add for your brand to be chosen.
References
- McKinsey Global Institute — AI agentic commerce market sizing, 2025
- Consumer AI adoption survey — purchasing journey and agentic AI expectations, power user cohort
- Amazon Rufus AI shopping assistant performance data, Black Friday 2025 analysis
- Carnegie Mellon University — AI brand recommendation instability study
- Yale, Columbia, University of Chicago — machine legibility and selection probability research
- Swiss watch industry historical data — quartz crisis market share analysis
- Swiss watch industry concentration metrics — luxury segment growth analysis
- BrightEdge / Gartner — AI Overviews impact on organic search traffic and CTR
- EverTune — cross-platform AI brand recommendation disagreement study
- llms.txt adoption impact — machine-readable format traffic analysis
- Consumer trust in AI commerce — retailer vs. third-party agent preference study
- AI referral traffic to major retailers — Black Friday 2025 analysis
- Kantar — “The Non-Human Consumer” report on AI brand selection
- Retail executive survey — generative AI impact on brand loyalty projections
- HBR — luxury brand and AI selection compatibility analysis