When the Buyer Is a Bot

Agents are starting to do the shopping, and they were never going to fall for your billboard.

For thirty years, the entire discipline of marketing has been built on a single assumption: that a human being is on the other end. A human who can be charmed, nudged, flattered, made to feel a little insecure and then a little better. We wrote headlines for the way attention drifts. We picked colors for the way moods shift. We A/B tested buttons because people are impulsive and a slightly bigger “Buy” can tip them over the edge.

That assumption is quietly expiring.

The reason is a class of software that didn’t really exist in a usable form two years ago and is now being wired directly into the tools people already shop with. Ask ChatGPT to find you a moisturizer under thirty dollars that won’t wreck sensitive skin, and increasingly it won’t just talk about products. It will compare them, weigh them, and offer to buy one. Microsoft’s Copilot can carry a checkout from start to finish. Stripe and OpenAI have published a protocol for it. Google and Shopify have published another. The plumbing for machine-driven purchasing is being laid down right now, in public, by the largest companies in the industry.

Which raises a strange and slightly vertiginous question for anyone who sells anything. What happens to your marketing when the buyer can’t be charmed?

Here is the uncomfortable part. An agent doesn’t have a mood. It doesn’t get tired at 11pm and buy the thing it’s been eyeing. It doesn’t recognize your logo and feel the warm flicker of brand familiarity you spent a fortune cultivating. It does not see your billboard. It will never watch your ad. When an agent goes shopping, it does something no human customer has ever reliably done: it reads everything, forgets nothing, and applies the same criteria to you that it applies to your competitor, in the same instant, without prejudice or fatigue.

This sounds like the death of branding. I think it’s closer to the opposite.

Consider what the agent actually needs to do its job. To recommend a product, it has to understand what the product is, who it’s for, what it costs, how it compares, and whether the source can be trusted. It assembles that understanding from whatever it can find and parse: your specifications, your reviews, the way third parties describe you, the structured data on your pages, the consistency of your claims across the web. The agent is, in effect, conducting due diligence at machine speed on every option in the consideration set. And then it makes a shortlist.

That shortlist is the whole game now. Not the click. Not the impression. The shortlist.

Getting onto it has very little to do with persuasion and almost everything to do with legibility. Can a model actually tell what you sell? Are your facts the same in five places or do they contradict each other? When someone asks the agent a question your product answers, does your name come up cleanly, attached to a claim the model can verify? The brands that win in this environment are not the loudest. They’re the most parseable, the most consistent, and the most independently corroborated. The agent rewards the dull, honest virtues that human marketing spent decades learning to paper over with vibes.

I keep coming back to a small thought experiment. Imagine the most skeptical, well-informed customer you’ve ever had. Someone who read every review, cross-checked every spec, ignored your packaging entirely, and could not be flattered. Now imagine that customer doing all of that research in four hundred milliseconds, for every single purchase, forever. That’s not a horror story. That’s just the new median buyer. The work, then, is to be the kind of company that does well in front of a customer who cannot be fooled.

There’s a temptation to treat this as a far-off thing, a 2030 problem. It isn’t, quite. Today the share of retail flowing through these channels is small, a rounding error against the whole. But it is growing fast, and the more telling signal is who is building. When the payment companies and the search companies and the commerce platforms all start shipping the same capability in the same eighteen months, they are not guessing. They’re paving a road they expect traffic on.

The honest strategic move is not to panic and not to wait. It’s to start auditing yourself the way an agent would. Pull up the questions your best customers ask before they buy, and check whether a model answering those questions today would mention you, describe you accurately, and have any reason to trust the description. Where it wouldn’t, you have found your actual marketing backlog, and it probably looks less like a campaign and more like a cleanup: contradictory specs reconciled, vague claims made checkable, the boring facts of your business stated plainly enough that a machine can repeat them without flinching.

For a generation of marketers trained to make people feel something, this is a genuinely disorienting shift. The craft we were good at, the emotional sleight of hand, the manufactured urgency, the carefully lit aspiration, has a shrinking audience. What’s growing is an audience that wants exactly one thing from you: to be true, clearly, in a form it can read.

The buyer is becoming a bot. The good news is that bots can’t be manipulated. The bad news, for a lot of brands, is exactly the same sentence.