What AI agents actually see when they visit your website
Everyone in your organisation has agreed that AI agents are coming. Nobody agreed on what to do next. Here is where to start.
Someone in your organisation has said "we need to be ready for AI agents." Everyone nodded. Then the meeting ended and nothing happened, because nobody could say what "ready" actually looks like.
One way to find out: when an AI agent visits your site, it does not see your homepage. It does not see your brand story or your hero image or your seasonal campaign. It sees what it can parse. For most e-commerce sites right now, that is not much.
This is not a future problem. Research published this year found that 38% of European consumers are already using AI to research and decide what to buy. By the time they arrive at your site, or their agent does, most of the evaluation has already happened. The consideration set has already been narrowed. If you were not legible to the agent upstream, you were not in that set.
"The question is no longer whether AI agents will mediate shopping. It is whether your site is built for a customer that does not read the way humans do."
What the numbers look like
We scanned 345+ retail and fashion sites across the US and Europe to see how ready they are for agent-driven commerce. The numbers held up across markets and brand sizes.
These are not obscure brands on slow platforms. They are established names with working online stores that humans shop on every day. Agents just cannot.
Three questions an agent is asking
Forget the jargon for a moment. When an AI agent is shopping on someone's behalf, it does three things in sequence: find your products, understand them, then buy them. Most sites fail at least one of these. Many fail all three.
Not Google. An agent. These are different things. An agent needs to enumerate your catalogue, pull a list of products it can reason about. If your site serves a JavaScript wall, blocks automated traffic, or has no structured product feed, the agent hits a dead end before it even reaches your products. Good SEO does not mean you are agent-discoverable. Most brands assume it does.
Agents do not read marketing copy. They read structured attributes. Material. Dimensions. GTIN. Variant SKU. Price and availability per size. If that data is not present in a format the agent can parse and compare, it has nothing to work with. It does not make a judgement call. It moves on. Most brands score highest on this dimension, and it is still not enough on its own.
This is where almost everyone falls over. Completing a purchase requires a programmatic cart API, a checkout flow that does not require solving a CAPTCHA or clicking through a modal, and payment infrastructure that supports delegated transactions. Most sites have none of this available. Publishing a UCP manifest says you want agent traffic. It does not mean an agent can actually get to checkout.
What actually breaks
In practice, a handful of issues come up again and again.
Many brands run CAPTCHA or WAF rules that treat any automated request as a threat. A legitimate AI shopping agent looks the same as a scraper to an unconfigured WAF. The agent gets blocked. It does not retry.
A human can read a product description and infer the material, the fit, the occasion. An agent cannot infer; it needs the attribute stated cleanly. Many sites have rich copy and weak schema. That is a problem when the customer does not read copy.
Multi-step forms, address validation modals, cookie consent interruptions, SMS verification at payment. Each one is a point where an agent stalls. Humans navigate these by instinct. Agents need a programmatic path.
Some sites have no exposed endpoints for cart, checkout, or product data. The only way in is a browser session. For an agent, that is a wall.
The gap is not between brands that have done the work and brands that have not. It is between brands that know where they stand and brands that do not. Most are in the second group, not because they have ignored this, but because nobody has looked at their site from an agent's point of view before.
What to actually do
You do not need to rebuild your site. You need to know where the gaps are, specifically, not in general. "We need better structured data" is not an action. "Our product schema is missing GTINs across 60% of our catalogue and our WAF is blocking all non-browser traffic" is.
The brands that get ahead of this are not necessarily the ones with the best tech stacks. They are the ones who look first. They find out their transactability score before an agent does. They fix the cart API before a competitor's cart API is the one that works.
"Agentic commerce is going to route demand the same way search did, toward whatever is most accessible and most legible. The brands that were easy to find in 2005 were not always the best brands. They were the ones that understood how the new customer read."
The new customer does not read. It parses. That is the whole thing.
Want to know where your store sits?
We scan e-commerce sites and tell you exactly what an AI agent can and cannot do on them. It is genuinely quite nerdy and we love it. Drop us a line if you want us to run the numbers on yours.
Get in touch