Online shops
Your shop is already being read by agents
Not next year. Now. The question is not whether they arrive, but whether your product survives the comparison they make in the fraction of a second after they do.
Somebody asks an assistant which magnesium supplement to buy. It does not open ten tabs and read them. It fetches what it can, compares what is comparable, and answers.
If your product data is incomplete, your product is not in that comparison. Not rejected — simply absent, silently, with nothing appearing in any report you look at.
This is not a forecast. Across five shops we look after, more than one order in ten in recent quarters has come from an AI source. We know because WooCommerce stores the origin of every order, and we read it rather than modelled it.
Where that sits
- 1 in 10orders in the shops we run, from AI sources
- 5 of 6requirements we would have met anyway
- 0plugins, connectors or third-party services involved
Industry reporting currently puts typical traffic from these sources at well under one percent, with fifteen to twenty-five percent treated as a projection for the coming years. That gap is the point of this article: nothing unusual was done to get there.
What an agent does differently
An agent does not browse. It compares.
A person landing on a product page skims, scrolls, forms an impression, and tolerates a great deal of vagueness on the way. If the pack size is unclear they guess. If the delivery date is fuzzy they assume it will be fine.
An agent cannot do any of that, because it has to act on what it reads. Ambiguity is not something it can shrug off — it is a reason to move to the next candidate, which is easier and always available.
That single difference explains almost everything below.
The six things that decide it
Product data that is actually complete. A missing weight, a missing pack quantity, an unclear dosage. Each one removes you from a comparison somebody made about your category. There is no error message for this.
Structured data that agrees with the page. Price, availability and shipping in machine-readable form — and matching what the page says. Where the markup claims one thing and the visible text another, neither is trusted.
Real HTML rather than content assembled on arrival. A good deal of what a page builder produces only exists once a browser has executed it. Something reading the page as delivered finds an empty frame where the price should be. This is the one that surprises people, and it is the hardest to fix afterwards.
Delivery cost and date, stated plainly. “Ships in two to three days, depending on region” is not comparable. A date is. This is the same requirement a customer has, only enforced strictly.
A page that answers before it is asked. How long it lasts, what it costs per use, what the difference to the neighbouring product is. Exactly the four questions a customer has — an agent needs the same answers, as text it can read rather than as an impression it can form.
Structure a machine can follow. Headings in order, real labels, lists that are lists. Identical to what a screen reader needs, which is why shops built properly for people turn out to be readable by agents without anyone planning it.
Five of those six are things any competent build should have done in 2020. Only the second one — structured data that stays in step with the page — is genuinely new work, and even that has been a search requirement for years.
Three checks you can run yourself
None of these need a tool or an agency. Twenty minutes, on your own shop.
View the source
Open a product page, press Ctrl+U, and search for the price. If it is not in there as plain text, it is being assembled by the browser — and something reading the page directly will not find it.
Pick five products at random
Check whether weight, pack size, quantity and delivery time are all filled in. Not in the description — in actual fields. If three of five have gaps, that is your first job and it is not a technical one.
Read your own orders
WooCommerce records where each order came from. Open the last few months and look for AI sources. Whatever the number is, it is the only one about your shop that matters.
What this does not fix
Being readable gets you into the comparison. It does not win it, and the reporting on this is candid: orders arriving through these channels currently convert considerably worse than through established ones, largely because the checkout that follows was never built with them in mind.
So this is not a reason to rebuild a shop. It is a reason to finish the product data you already started, and to stop assuming that what a person can infer, a machine can too.
Nobody set out to optimise for this. The shops were built to be read, and something started reading them.
That sentence is the honest version of our own figure, and it is also the practical advice. There is no separate discipline here to buy — there is a standard of completeness that either exists in your shop or does not.
What that looks like when it is done deliberately is on our AI search page, including where we think the limits currently are.
Send us one product page.
We will read it the way an agent would and tell you what is missing — in writing, and yours to keep whether or not you hire us. If nothing is missing, that is a short email and a good day for both of us.