When an AI Agent Does the Shopping, Your Data Is the Shop

The buyer stopped visiting

For twenty years the job was to get a person onto a page and persuade them. Every discipline downstream of that — design, copywriting, conversion optimisation, retargeting — assumes a human arrives and looks.

Agentic commerce removes the arrival. Google's Universal Commerce Protocol is an open standard that lets AI agents read a merchant's catalogue, pull live pricing and inventory, assemble a multi item cart, and complete a purchase inside AI Mode in Search and inside Gemini. Google co-developed it with Shopify, Etsy, Wayfair, Target and Walmart, with payment and retail names endorsing it across the ecosystem.

Read that mechanically. The agent queries structured data over an API. It does not load your homepage, does not see your photography, and does not experience your brand. It compares fields.

What this changes, and what it does not

It does not make brand worthless. A person still tells the agent what to look for, and what they ask for is shaped by what they already trust. Brand moves upstream, into the prompt.

What changes is the middle of the funnel. The comparison step — the part where a shopper opened four tabs and weighed things up — is being done by something that reads specifications, price, availability, and returns terms, and is completely immune to a hero image.

So the asset that decides whether you are in the consideration set stops being your storefront and becomes your product data.

The uncomfortable audit

Most merchants have never looked at their feed the way an agent does. Open yours and ask:

Is every attribute populated? Dimensions, materials, weight, compatibility, colour, size. An agent filtering on a field you left empty does not guess. It excludes you.

Is pricing live and correct? Including delivery. A feed that disagrees with the site is worse than an incomplete one, because it produces failed carts and the agent learns to distrust the source.

Is stock accurate in real time? Availability is a filter. Being listed as out of stock when you are not is invisible lost revenue.

Are returns and delivery terms machine readable, not buried in a footer PDF written for humans?

Do your product titles describe the product rather than perform for a keyword? Agents parse meaning; stuffed titles read as noise.

None of this is glamorous and all of it is now competitive. As the merchant guidance makes plain, the brands that get selected are the ones that are data complete and accessible, not the ones with the biggest ad budget.

Where this connects to search

This is the same shift covered in two thirds of searches end without a click and what structured data actually does, arriving in commerce.

Worth being precise, because agencies are already overselling this: Google has said there is no special schema that makes you eligible for its AI features. Structured data does not buy you a citation. What it does is remove ambiguity, and in an agentic transaction ambiguity is disqualifying rather than merely unhelpful. The requirement has changed from optional clarity to a functional prerequisite.

What to actually do this quarter

1. Audit the feed against the fields agents filter on. Completeness first, accuracy second, prose last.
2. Fix the price and stock sync. If the feed and the site can disagree, they will.
3. Write specifications like a spreadsheet, and descriptions like a human. You need both, for two different readers.
4. Publish the boring facts. Delivery windows, returns policy, warranty. These are decision fields now, which is the same argument as the product page is the whole shop.
5. Keep building the brand anyway. The agent executes a request. The request comes from a person who already had an opinion, and that opinion is built by everything an agent cannot read.

The honest uncertainty

Nobody knows what share of purchases will run this way, and the forecasts being circulated are mostly from companies selling readiness services. We are not going to add one.

What is checkable is that the protocol exists, that major retailers and payment networks are behind it, and that it is live inside Google's own surfaces. That is enough to justify fixing your product data, which was worth doing regardless.

Frequently asked

Does this apply to service businesses? Not directly, and be sceptical of anyone telling you otherwise. The equivalent for services is being clear, specific, and quotable, which is getting cited by AI search.

Should we block AI agents from our store? Only if you also want to be excluded from the carts they build.

Can Fixora do this? Yes: a feed audit against the fields that actually filter, the sync fixes, and the product content underneath. Send us your store and get findings within 48 hours.