Signal
A new kind of buyer is reading the shelf, and it does not fall for the things that work on people. When two London business-school researchers put four leading AI models through thousands of simulated shopping rounds, only star ratings reliably raised a product's chances and only price reliably lowered them, while countdown timers, crossed-out prices, vouchers and bundles did nothing or backfired, and the reasoning models treated open persuasion as a warning sign. Accenture's survey of 25,590 people in sixteen countries put the split in one line: what agents can verify they will recommend, what people find meaningful they will remember. So the selling is being rebuilt around checkable facts. Shopify made every store readable to agents by default and from October passes them the crossed-out price as well, Alibaba connected its assistant to all four billion Taobao listings, and a trade has grown up around watching what the machines say, with Peec in Berlin signing 1,300 brands in its first nine months. The machine is not a perfect judge. Wharton found that an agent's pick depends on which review it saw first. But it reads the whole label and never buys the feeling. Persuasion was built for a buyer with feelings. The buyer arriving now has none, and it reads everything.
So what?
People buy stories. Agents buy proof. Marketing is splitting into persuading humans and convincing verifiers, and the budgets have started to move.
Evidence
Harvard Business Review, 12 May 2026 (study): two London business-school researchers ran four leading AI models through thousands of simulated shopping rounds in four product categories. Only star ratings consistently raised a product's chance of being chosen and price reliably lowered it, while scarcity, countdown timers, strike-through pricing, vouchers and bundles did not reliably work.
Shopify, 24 March 2026 (launch): products from millions of Shopify stores became discoverable in ChatGPT by default, with no app or integration, after Shopify's December pledge to make every store agent-ready by default.
Fortune, 24 February 2026 (startup): Profound raised 96 million dollars at a 1 billion dollar valuation to show more than 700 companies, including Walmart and Target, how AI models describe and recommend their products. Its own research found that up to 90 percent of the sources cited in AI answers change over time.
Worth asking
Think of the part of your work that people value most. If a machine that reads only raw data were auditing your job, how would you highlight that value?
