A person reads “feels like heaven” and pictures a cloud. A machine reads it and learns nothing.
That sentence is the whole shift, if you sit with it. For as long as there have been products to sell, copy has been written to make a human feel something — to take a list of facts and dress it in mood, story, and aspiration. The spec was always in there somewhere, but the job of the writing was to make you want the thing before you’d checked whether it fit.
The most important reader of product copy is now a model. And the model feels nothing, pictures nothing, infers nothing. Humans are forgiving readers. Machines are literal.
This is not a small adjustment to how we describe things. It’s an inversion of what the description is for.
The spec used to hide inside the story. Now it is the story.
A human can look at a photo and infer scale. An agent can’t. A human can read “all-day battery” and shrug, knowing it means roughly nothing. An agent has to decide whether your product clears a real constraint, and “all-day” is not a number it can compare.
So the craft flips. The literal title that used to feel lazy — “women’s waterproof hiking boot” — is now the strong one, because it says what the thing is. “Ocean Breeze” belongs in the creative, not the title. The dimensions, the materials, the weight, the compatibility, the care instructions, the certifications, the list of what’s actually in the box — the boring fields most catalogs leave half-empty — are the part the machine reads most carefully. If a shopper might ask it, you have to have stored it.
This feels like a step down in artfulness. It isn’t. It’s a different art: saying exactly what is true, with nothing load-bearing left to inference.
Uncertainty is the enemy of recommendation
Here’s the failure mode that catches people who think they’ve already done this work.
An agent will go hunting. It will read your returns page, your size guide, three reviews, and a spec sheet trapped in a PDF, because it has more patience than any human shopper. But it prefers clean inputs, and it is exquisitely sensitive to contradiction. When your product page says “water-resistant” and your feed says “waterproof,” the machine doesn’t pick one. It sees uncertainty. Uncertainty lowers its confidence, and low confidence lowers the odds it puts you forward at all.
So legibility isn’t only completeness. It’s consistency — one truth, everywhere. It’s structure — numbers stored as numbers, units made explicit, options enumerated, a parent product with real variants instead of five near-identical pages pretending to be different things. And it’s freshness, because if the agent recommends something that’s out of stock or mispriced at checkout, the shopper blames the agent, and the agent learns to stop trusting the source that embarrassed it.
Take one ordinary product, a USB-C cable. Across the web it’s a “USB-C to USB-A Cable,” a “Type-C charging cord,” a “USB 3.0 C-to-A cable,” and a dozen other things, each catalog naming it by its own house convention. A person knows these are the same object. A machine has to be told. Every place your product’s identity wobbles — a different name here, a missing unit there, a spec that contradicts the photo — is one more place the agent has to guess, and guessing is the one thing it’s trying not to do.
A product page can look perfect to a human and be unreadable to a machine. Those are now two different audiences, and you are writing for both.
The honest part
There’s something almost moral in this, and it’s the part I find genuinely interesting.
The most useful thing you can give an agent is often the thing old marketing worked hardest to hide: the constraints. Who the product isn’t for. What it doesn’t do. The fit that runs small, the feature it lacks, the case it doesn’t cover. A human salesperson buries that. An agent wants it, because its job is to protect the shopper as much as to serve them, and a product honest about its limits is a product the agent can place with confidence.
For a long time, persuasion rewarded a certain amount of strategic vagueness. Writing for machines rewards the opposite. The brands that win the agentic era won’t be the ones with the most evocative copy. They’ll be the ones that learned to write plainly for a reader that can’t be charmed, can’t read between the lines, and never stops checking.
Say what it is. That used to be the floor. Now it’s the whole game.