For most of the twentieth century, a book in an American library lived in two places at once. It sat on a shelf, and it sat in a wooden drawer as a small stack of typed cards. The shelf held one copy. The drawer held several descriptions of it: a card filed under the author, a card filed under the title, and one or more cards filed under its subjects, each written to a standard the Library of Congress had made common when it began selling printed cards in 1901. A reader who knew the author walked to one drawer. A reader who only knew what she wanted to learn walked to another, and found the book only if a cataloger had thought to file a card there. A careful cataloger gave a book on keeping bees a card under Bees, another under Honey, and perhaps a third under Pollination, and nothing about the book changed. What changed was the number of questions that led to it.
That is the most useful way I know to think about the advertising that now sits under most product searches, and about the machines that are learning to shop.
The Drawer Moved Online
Last week Jackson Blackledge, who works on Shopping accounts for ecommerce brands, published a careful piece on how to "convince" Google to show your ads in more searches. His frame is the right one: Shopping is a coverage game. An advertiser does not buy keywords there. The product data decides which searches a product is even eligible to appear in, and then how well it competes once it is in the room. He reports that brands he has worked with reached 2-3x their Shopping traffic in as little as 30 days by fixing that data, and in his telling almost none of the work involves bids. It is cataloging.
His particulars read like a cataloger's manual. The title is the strongest relevance signal the system has, and he argues that its opening stretch, roughly the first 30 to 70 characters, carries most of the weight, so a title that opens with the brand name spends its best words on people who already know the brand. For prospecting he puts the product first and the brand last, and keeps a separate set of listings for branded searches. He wants every attribute filled, the real barcode and never an invented one, a store that passes Google's checks on shipping, returns, browsing and checkout, and enough reviews to earn seller ratings. Then comes the move most brands find strange at first: list your best products more than once, each listing written for a different reason to buy. His example is a magnesium supplement described one way for sleep, another for recovery and another for children, which is to say three cards in the drawer pointing at the same bottle.
The bottle does not change. The number of shoppers who can find it does.
A Card for Every Question
The drawer is now being read by something other than a person. Google has added what it calls conversational attributes to Merchant Center: fields for product questions and answers, links to manuals and specification documents, related products, variant options and a popularity rank. They do not affect whether a product is approved. They exist because shoppers have started asking longer questions, and in its third quarter remarks Google said queries in AI Mode run on average three times longer than traditional searches. A shopper who types "water bottle" is looking up a title. A shopper who asks for a bottle that keeps ice through a twelve hour shift and still fits a car's cup holder is looking up a subject, and only a product described well enough to match the subject can answer her.
This is where the series that began with The Toll Comes After the Road arrives. That piece argued that new roads are cheap while they are being paved and priced once the traffic becomes a habit, and that advertising inside AI answers will follow the sequence mobile followed more than a decade ago. The toll is about price. This is about entry. Before anyone can pay to travel the new road, the road has to know the car exists, and on the roads machines are building, the road knows only what the catalog says. The feed has become the brief. It is the document a machine reads to decide whether you belong in the answer, and unlike the person who receives a creative brief, the machine will never call to ask what you meant.
The Economics of a Thin Catalog
A thin catalog is cheap to maintain and expensive to own. Every empty attribute is a question the product cannot answer, and every question it cannot answer is a search it never enters, which means the loss never appears anywhere. A keyword you chose not to bid on at least shows up in a report as a decision. A query you were never eligible for leaves no impression and no click to count. The cost of a weak title is invisible in exactly the way the cost of a missing catalog card was invisible: the reader simply concluded that the library did not own the book, and went somewhere else.
That invisibility is why catalog work gets handed to whoever has time, usually a coordinator with a spreadsheet, while the serious budget conversation happens down the hall about bids and creative. It is the wrong way round. The bid decides how much you pay to compete for a question; the catalog decides how many questions you are allowed to compete for. A brand that doubles the questions it is eligible to answer can grow at the same bid, and Jackson's 2-3x is best read that way, as coverage gained rather than efficiency squeezed out of the same narrow set of searches.
There is a limit, and he names it plainly. Listings duplicated by hand do not update themselves, so a change in price or stock has to be carried to every copy, and a catalog that multiplies its cards carelessly multiplies its mistakes. A cataloger who filed a card under every conceivable subject would have made the drawer useless, and Google is no more patient with clutter than a reference librarian was. He suggests a few angles for the products that matter most, not the whole range. The discipline is in choosing the questions, not in answering all of them.
Who Writes the Cards
So the real question for a marketing leader is ownership. Somebody in the company decides what each product is called, which questions it answers and which buyer each listing is for, and today that decision is usually made by default inside a merchandising system built to count inventory rather than to explain anything. The fields that put a product into more Shopping searches are the same fields an AI shopping agent reads when it compares three bottles on a customer's behalf. A catalog written for people who already know the brand will increasingly be read by software that does not.
The repair is not glamorous, which is why it is still available. Give the catalog an owner who answers to marketing as well as operations. Write titles for the buyer who has never heard of you, and keep the brand's name for the searches that already contain it. Take the answers your sales and service teams give on the phone every day and put them in the fields Google now offers for questions and documents. Choose the handful of products where a second or third reason to buy is real, and give each reason its own listing. Then read the search terms the way a librarian read the slips left at the reference desk, looking for the questions people asked that led nowhere.
A library never held more books than its drawer admitted to. Neither does a store.
