Your product feed is your campaign now

📋

This piece is a response to the Google AI Max for Shopping announcement. The system analyses product attributes including fabric softness, material durability, and fit to understand what a product actually is, then matches those attributes against conversational search queries.

For years, shopping teams measured catalogue health the same way. Schema coverage. GTIN fill rate. Stock availability as a site-level average. The dashboard turned green. Everyone moved on.

Those averages worked when Google was matching keywords. They stop working when the system reads your feed like a document and decides whether your product belongs in the answer to "soft organic cotton that does not shrink."

And they stop working entirely when the buyer is an AI agent, not a person.

An AI shopping agent does not buy your catalogue. It buys one product. If the one it picks happens to be in the 13 per cent without proper schema, or the variant showing InStock when size 10 sold out this morning, the agent moves on. The sale is gone.

Per-product, not per-site

At Aidō Lighthouse, we score every product individually against one question: can an AI agent actually buy this? Five axes. Not a site average. Every SKU.

🔍

Identifiable

💰

Priceable

📦

Stockable

🛒

Purchasable

↩️

Returnable

Example — SPF 50 tinted moisturiser

On the surface it looks simple. In practice it has to be identifiable across a dozen shade variants, each with its own stock status. The price has to be machine-readable with a declared currency. And a per-product return policy matters here because the consumer cannot try the shade before buying, so a mandate-bound agent will check whether returns are declared before it commits.

That is one product. A large beauty retailer might have three thousand like it.

The constraints nobody is looking for

What surprised us most while building Aidō Lighthouse was how much of a typical catalogue is structurally not purchasable, even when the schema looks fine.

B2B trade packs sitting next to retail SKUs. The feed does not distinguish them. An agent browsing for a single unit finds a case-of-24 and either buys the wrong thing or abandons.

Furniture listed at "call for price." The price field is unpopulated or contains a string. An agent cannot evaluate it. The product does not exist for agentic commerce purposes.

Age-restricted products with no declaration. An agent has no mandate to verify age. Without a machine-readable declaration, it cannot know the product requires one.

Subscription-only items that redirect one-time purchase attempts into a sign-up flow the agent has no authority to complete. The checkout starts. It does not finish.

These are not schema errors. They are business logic that was never meant for autonomous buyers. Google AI Max will surface them as reach and relevance problems. Agents will choke on them at checkout. Neither will tell you why your conversion dropped.

Google AI Max is a good reason to look at your feed properly

If Google is now reading your product attributes to decide reach and relevance, auditing what those attributes say is overdue. The same logic applies to agentic commerce, one level deeper: not at the site, but at each individual product, and the business logic hiding inside it.

The campaign starts with the feed. Agentic commerce starts at each individual product, and the business logic hiding inside it.

See which products are blocking you

Run a scan in Aidō Lighthouse. Look at which products score lowest. Usually at least one turns out to be a structural problem you did not know you had.

Scan your catalogue