26 March 2026 · LinkedIn
Sixteen percent more sales. No additional staff, no price changes, no new products. Just generative AI quietly rewriting product descriptions and restructuring search results across millions of transactions on a single retail platform.
That's the headline finding from a Columbia and Wharton-led field experiment, run over six months on a major cross-border e-commerce operation. Return rates didn't budge. Customer satisfaction held steady.
The more revealing number is buried deeper. The consumers who gained most were the least experienced. Newer shoppers, less familiar with the platform, less practised at navigating choice overload. The AI compressed the gap between the confident buyer and the uncertain one. Frictions that previously penalised inexperience were quietly removed.
In effect, the AI was negotiating on behalf of the customer. Surfacing relevance, filtering noise, matching need to product with a precision no human sales assistant could sustain across millions of simultaneous interactions. For the less experienced buyer, it functioned as an advocate they never asked for and never knew they had.
That sounds like progress. And in isolation, it is.
But zoom out. What this study captures is the opening move of a much longer game. Hyper-personalisation doesn't stop at tidying up search results. It learns what holds your attention, what triggers your trust, what sequence of impressions moves you from browsing to buying. Every interaction refines the model. Every transaction trains the next one.
When the organisation behind that model is working in the customer's interest, the result is genuine value. When it isn't, the result is something else entirely.
We have history here. Tupperware parties monetised trust between friends. Ann Summers leveraged social pressure in living rooms. Multi-level marketing turned personal relationships into sales infrastructure. None of it was illegal. Influence, exercised face to face, has never been legislated against with any seriousness.
Now scale that capability by several orders of magnitude. An AI that knows your cognitive patterns, your emotional triggers, your purchase history, your hesitation points. Deployed not by a friend in your living room but by every platform you touch, every hour you're online. Persuasion that doesn't feel like persuasion, because it's been personalised to the precise shape of your attention.
Sam Altman observed in 2023 that superhuman persuasion would likely arrive well before superhuman general intelligence. The Columbia study isn't that. It's the scaffolding. Population-level optimisation that already works at scale. Individual-level persuasion, tailored to the precise shape of your attention, is what comes next. And there's no legislation waiting for it when it arrives.
Join the conversation on LinkedIn →
← All writing
