How does an AI assistant decide which products to recommend?

That step is called Bought, and it is the one that differs most from one industry to the next. A dental practice is measured on whether a patient can reach a booking. A software company is measured on whether a prospect can reach a trial. A retail brand is measured on the shelf, because the shelf is where a purchase happens.

So the thing being checked is more than who got recommended. It is the price quoted beside your name, the stock claim attached to it, and whether the buyer left with a named destination. An answer that praises you warmly and then points nowhere in particular is an answer somebody else finishes.

What gets read is the answer and the page together: the figure an assistant quotes, and the figure your own product page publishes in a form a machine can lift. Where the two agree, the shelf is intact. Where they diverge, there is a specific page and a specific claim to go and settle.

A retail brand is placed by distribution rather than by a map. Your products sit on shelves belonging to shops, marketplaces and stockists, and an assistant answering a buying question reads those listings much the way a shopper would. That is what separates this from a practice, which an assistant places by town, and from a firm, which it places by jurisdiction.

That is a different competitive position from most industries and it cuts both ways. A retailer listing carries weight no page of your own can, because it arrives with a price, a stock claim and a body of reviews attached. It is also written by somebody else, so putting it right is a conversation with an account team rather than an edit to a page.

Your own pages still matter, and they matter for a precise reason. They are what a crawler reads for your prices and specifications, and they are the version an assistant is compared against when a listing has drifted. The page you control is the evidence.

A retail brand holds every one of the things the site checks look for, which is the most concrete advantage this business type has. It publishes product markup and offer data, it publishes a price and an availability, and it can expose a machine route into its catalogue.

So all twelve read what they were built to read. Whether AI crawlers can reach the site. Whether an llms.txt file is published. Whether the organisation is identifiable as itself. Whether the pages can be read without running JavaScript. Whether Organization structured data is present. Whether pages declare when they last changed. Whether Product structured data is present. Whether product pages carry complete offer data. Whether prices are published where a machine can read them. Whether availability is published and current. Whether the links a buyer is sent to resolve. And whether a machine route into the catalogue exists.

The comparison is what makes that concrete. A practice has no product feed. A professional firm publishes no fee, and is right not to. A hotel keeps its rates in a booking engine on a host it does not own. Each of those is measured on a smaller set, and a retail brand is measured on all of it, so every gap here is a gap with a specific page behind it.

The broad category question is the one that decides who gets named at all. Best waterproof jacket, best gentle shampoo, best kettle for hard water. It is typed by somebody with no shortlist yet, and it is the widest door into a category.

The comparison question is the head to head against the brands a shopper is already weighing up, by name. The constraint question is budget, size, material and the rest of what somebody types once the category is settled and the choice is not. Those three carry the most, and they are where a purchase is won.

Three more sit underneath and each does something different. The job the product is bought to do. Where to get it, at what price, in stock where, which is the closest a shopper comes to typing a transaction. And the research question asked before the decision rather than at it, which shapes what somebody believes about a whole category and is kept out of recommendation scoring, because there is no recommendation inside it to win.

Retailer pages come first, because that is where an assistant checks price and availability and where most buying answers end up. Consumer review and testing sites come next, because they carry the verdicts assistants repeat back.

Behind those sit the places where a category is argued out rather than rated. Threads and video are where the real objections live: the sizing that runs small, the part that wears first, the thing no specification mentions. Buying guides and roundups are cited heavily on category questions, which is the question that decides who gets named at all.

Then there is the long tail of smaller sites, which is where a new category conversation usually starts. A measurement names those alongside the large ones rather than reporting one figure for everything else.

Coverage is the outward half. It is presence in the retailer listings, review verdicts, roundups and threads a buying answer is built from, described the way you would describe yourself.

The shelf is the half you own outright. Complete offer data on a product page, a price a machine can read, an availability that is current, and links that resolve. It is unglamorous and it is the quickest thing on the list, because nobody outside the building has to agree to it.

The two halves fail differently and that is what sets the order. A gap on your own shelf is a page you can edit today. A gap in coverage is a conversation with a retailer, a reviewer or a community, so it starts sooner and lands later. The measurement says which of the two you are short on, which is the part a preference gets wrong.