The next customer at your checkout isn’t human

Somewhere this week, a customer asked their assistant to “find me a three-month supply and order it.” The assistant read a handful of product pages, compared prices and delivery, and bought from one of them. No page views you would recognise, no scrolling, no ad exposure. If your store was in that comparison, you either won it or lost it based on what your pages say to software.
What an agent needs from a product page
Not your photography. An agent needs the facts, machine-readable: what the product is, what it costs, in what currency, whether it is in stock, what shipping costs and takes, and what happens if it comes back. Humans infer these from layout and prose. Agents read structured data — and when the structured data is missing, many simply move to the next store rather than guess.
This is the quiet unfairness of the agentic web: the store with worse photos and a machine-readable returns policy beats the beautiful store whose shipping costs only exist inside a checkout flow no agent will complete speculatively.
The three layers, in order of effort
Structured data first. Product, price, currency, availability on every product page. This is decade-old technology, it also feeds Google Shopping, and it is the difference between being comparable and being invisible.
Policies as data second. Shipping and returns exposed as fields, not paragraphs. An agent asked to “order the cheapest with free returns” cannot act on a paragraph.
Direct transaction rails third. The newer standards — a discovery document that tells agents how to transact with you, tools exposed for the browser-native assistants that ship with Chrome — are where early movers are now. This layer is optional in 2026. The first two stopped being optional about a year ago.
How to know where you stand
You cannot see this failure in your analytics, because the visits that fail are nearly invisible. You have to test the pages the way an agent reads them. That is what our agentic commerce module does on every audit: reads your product pages as software, and tells you which facts an assistant could not find. Every missing field is a comparison you are silently not entered in.