What shopping agents actually need

Ask ten merchants what an AI shopping agent needs and you will hear about better descriptions and nicer photos. Those help people. Agents need something plainer: facts they can trust, in a form they can read, at a moment they can act on them.

Facts, not copy

A product description written to persuade is hard for an agent to use. What it needs is the underlying facts: dimensions, materials, compatibility, size availability, price, currency and delivery estimate. When those facts live inside paragraphs of marketing language, the agent has to guess which sentence is a claim and which is a specification.

Stability

Agents compare. If a price changes between the listing and the cart, or a product shows as in stock and then fails at checkout, the agent learns to distrust that store. Stable, consistent data is a form of trust a machine can measure.

Policies it can read

Return windows, shipping fees, restocking charges and warranty terms decide whether a purchase is a good one. If these are buried in a PDF or spread across several pages, an agent buying on behalf of a person cannot weigh them. A short, plain statement of each policy, linked from the product page, is enough.

A checkout it can finish

The last step is where most agent attempts fail: forced account creation, one-time codes sent to a phone, captchas and multi-step forms with unclear errors. Some of these protect real customers from fraud. The work is to protect them without making a legitimate agent acting for a legitimate person impossible to serve.

• Guest checkout or a delegated-authority path.
• Clear error messages that say what field is wrong.
• Totals that include tax and shipping before the final step.
• A confirmation the agent can read and hand back to the person.

What not to do

Do not build a separate site for agents and hope it stays in sync with the real one. Make the real site readable. Everything that helps a machine read it also helps search engines, screen readers and slow connections.