AI Shopping Assistants Are Recommending Your Products From a Feed You Never Checked
A shopper asks an AI assistant for the best waterproof hiking boot under 150 dollars, and the assistant recommends your boot, with the wrong price, the wrong waterproof rating, and a feature your boot does not have. You never wrote that copy. Your feed did, three months ago, and nobody has looked at it since.
The feed is the storefront now
AI shopping assistants do not browse your site the way a person does. They read structured data: your product feed, your schema markup, your titles and descriptions, and whatever review and listing data they can reach. When the feed says a jacket is waterproof and the product page says water resistant, the assistant picks one, and it usually picks the feed, because the feed is structured and the page is prose. When the feed lists a price from three months ago, the assistant quotes it, and the shopper arrives at checkout expecting a number that no longer exists.
The errors are mundane and that is why they survive. A title that says 12 pack when the listing is a single unit. A color field that says assorted. A size field that is empty on half the catalog. A description generated by an AI tool that invented a feature, waterproofing, a pocket, a warranty, that the product does not have. Each one is a small lie, and the assistant repeats it with confidence, because the assistant has no way to know it is a lie. The shopper finds out at the returns desk.
What to audit, in order
Start with the fields the assistants actually use: title, price, availability, description, and the attribute fields for the category, size, color, material, and the one feature that defines the product. Pull the feed and sort by the fields that are empty, inconsistent, or auto-generated. The auto-generated descriptions go first, because they are where invented features live: a product description tool that hallucinated a waterproof rating is a returns problem wearing a feed problem's clothes.
Then check the schema markup on the product pages against the feed. The two should agree on price, availability, and the core attributes, because an assistant that finds a disagreement has to choose, and it will choose the structured one. Then check the prices, because a feed that updates weekly while prices change daily is a feed that lies on the days that matter, like the first day of a sale. For the description version of this problem, see AI Product Descriptions Invent Features, and Returns Pay for It, and for the image version, AI-Generated Alt Text on Product Photos Is a Liability.
The check that keeps it clean
Feed hygiene is a weekly job, not a launch job, because prices, stock, and copy change every week and the feed does not update itself. Put the feed audit on the same schedule as the price changes: every time a price changes, the feed changes with it, and every time a description is generated, a human reads it before it ships. The assistant will keep recommending your products either way. The only question is whether the recommendation is true.
Frequently asked questions
How do AI shopping assistants find my products?
They read your product feed, your schema markup, and your site copy. The feed is structured data, so it usually wins over the product page when the two disagree, which is why feed errors become assistant errors.
What feed fields matter most?
Title, price, availability, description, and the attribute fields for the category: size, color, material, and the defining feature. Empty or inconsistent fields in those five are where most errors live.
Why do AI-generated descriptions cause returns?
Because the generation tool can invent a feature the product does not have, a waterproof rating, a pocket, a warranty. The assistant repeats it confidently, the shopper buys on it, and the return comes back with the product that never had it.
How often should we audit the feed?
Weekly, on the same schedule as price changes. Every time a price changes, the feed should change with it, and every generated description should be read by a human before it ships.