If an AI agent is going to recommend a product, it has to understand that product first. Most catalogues were built for human eyes, and a good photo can hide a lot of missing information. Agents cannot be charmed by a photo. Here is what we believe a catalogue needs.

Stable identifiers

Every product and variant needs an identity that does not change. Where they exist, manufacturer part numbers and global trade item numbers let an agent recognise that two listings describe the same thing.

Attributes, not adjectives

"Spacious" is an opinion. A capacity in litres is an answer. Dimensions, materials, compatibility, power and care instructions should be separate, consistent fields rather than sentences buried in a description.

Honest variants

Each size and colour should have its own price, stock level and image. An agent that picks a colour which turns out to be unavailable has wasted someone's time.

Prices that are final

People dislike surprise costs at checkout, and an agent comparing options needs the real total. Taxes, delivery charges and any conditions should be available before the final step.

Delivery and returns as data

"Fast delivery" means nothing to software. Estimated dates by destination, carrier, return window and who pays for the return can all be expressed clearly.

Freshness

Information about stock and price goes stale quickly. A catalogue that is accurate once a day will mislead agents for the rest of it.

A simple test

Take any product and ask whether someone who could only read your data, and could not see your page, would know what it is, whether they can have it, what it will cost and what happens if it is wrong. If the answer is no, the page is doing work the data should be doing.

This is the groundwork behind our commitment to the Universal Commerce Protocol, and it improves things for human shoppers as well.