When AI agents buy instead of people: getting your store ready
- Increasingly the purchase is completed by an agent — ChatGPT, Gemini or Perplexity acting for the buyer.
- An agent does not look at your design. It reads price, availability, delivery and specifications in the data.
- Most of the work is not “new technology” — it is putting product data in order.
- A store without valid Product markup and without a feed never enters the shortlist.
A few years ago voice search promised to change retail and did not. Agents are a different story: the buyer does not ask “where can I buy this”, they say “find me this under 2000 with delivery by Friday” — and get a decision, sometimes an order.
What changed technically
In 2026 there are protocols that let an agent not merely find a product but pay for it: OpenAI’s Agentic Commerce Protocol built with Stripe, and Google’s Universal Commerce Protocol. The consequence is simple: there is now an intermediary between your product and your buyer, and it reads your site by machine.
That intermediary is fussy. It will not decipher a price written into an image, it does not understand “call to confirm availability”, and it will not guess that “delivery 1–3 days” applies only to the capital.
What an agent looks at
| What it reads | What a poor store shows | What it needs |
|---|---|---|
| Price | “from 1500” as text, or an image | A number in Offer markup, with a currency |
| Availability | “Please enquire” | InStock / OutOfStock, kept current |
| Delivery | A separate page with a table | Terms next to the product and in the data |
| Specifications | One paragraph of prose | Attribute and value pairs |
| Variants | A separate listing per colour | One product with variants |
| Returns | Not described | Stated in plain text |
What to actually do
Product and Offer markup. Not exotic — the same structured data Google has been asking for over a decade. Free to verify with the Rich Results Test.
A product feed. A Merchant Center feed powers Google Shopping and serves as a machine-readable source. For WooCommerce or OpenCart it is a one-time setup.
Specifications. The dullest and most valuable part. “Wooden chair, comfortable” tells an agent nothing. “Material: oak. Height: 92 cm. Load: 120 kg” tells it everything.
Stock levels. An agent that sent a customer to an out-of-stock product will send the next one to a competitor.
Clean product data pays for itself whether or not agents live up to the hype. The same data feeds Google Shopping, price comparison sites and your own reporting. A rare case where preparing for a trend is useful on its own terms.
What not to do
- Block AI crawlers. You disappear from the shortlist; the shopping carries on without you.
- Add a chatbot instead of fixing the data. A bot helps your visitors; external agents have no use for it.
- Rebuild the store from scratch. Nine tasks out of ten are solved on the platform you already have.
This is urgent for anyone selling comparable goods: electronics, parts, consumables. Less so where the product is chosen by eye — made-to-order furniture, designer clothing, services. Even there, clean data does no harm: search visibility starts from it in the first place.
We will audit your store — markup, feed, stock, specifications — and tell you what is genuinely worth fixing.