An afternoon with the stock list: checking four assistants' answers

What to have open before the first question

The check is a reading of your store's AI visibility on one day, so everything you read the answers against has to be from that day. Before you type anything, open:

• The day's stock list, exported from the register or the inventory system, with the brands and lines you carry and what is on the shelf.
• Your own pages: the brands page, each location page with its hours, the stock and pickup page, and the service pages.
• The Business Profile for each location, and any directory listings you know of.
• A sheet with one row per answer: the date, the assistant, the question, whether the store was named and for what, each fact the answer stated marked right or wrong, the page it cited, and the other stores it named.
• The four assistants, with web search on wherever an assistant offers the choice, run the same way for every question: signed in or not, and noted.

Run every question in one sitting. An answer read against Tuesday's shelf and another against Friday's cannot be compared, and a brand that came in on Thursday makes Tuesday's "wrong" right.

Twelve questions, written from the stock list

Write the questions from the stock list itself, not from memory. Take two categories you sell, two brands you carry, one brand you stopped carrying, one item that is on the shelf today, the conditions shoppers attach, and a few questions about the store by name. The brackets are yours to fill in:

Twelve questions on four assistants make forty-eight answers, which is an afternoon's reading if each question runs once. Our published methodology asks each of its twelve questions once, of one assistant, Claude, and dates the reading, because answers differ from one run to the next; it reads how often you are named across the twelve and, with monitoring, across months. One afternoon is enough to find wrong facts; it is not enough to say how often you are named.

Reading each answer against the shelf and the site

For each answer, note first whether the store is named, and for what: a category, a brand, a service, a condition, or the store asked about by name. Then go fact by fact. Every brand the answer says you carry is checked against the stock list and the brands page, every day or time against the location page and the Business Profile, every stock or pickup claim against the shelf and the stock and pickup page, and every address against the locations you have open. Mark each fact right or wrong, and paste the whole answer into its row.

Then record the source. Where an assistant shows the pages it drew on, write down which: your own page (and which one), your Business Profile, a directory, a marketplace listing, another store's page, a review site or a forum. That source citation is where the trace of a wrong fact begins.

An answer with no source to show may have come from what the model learned before its knowledge cutoff. Anthropic describes its web search tool as giving Claude web content beyond that cutoff, with citations for the sources it draws from the results, and an answer written without a search has neither. That is one way a brand you dropped a year ago can still turn up.

Four wrong answers, and what to do about each

Four kinds of wrong answer have a fix a store can make within the week:

The fourth case needs care. Which page an answer cites is the operator's choice, and no page of yours can settle it. What your page can do is state the fact the answer used, so that a page of yours is there to be cited. The routine for the first three is in the same-day change routine.

A wrong fact that appears nowhere, on no page you can find, is a different case, and may be a hallucination. Note it, look for it again next month, and do not rewrite your pages around it. For wrong facts on pages you cannot edit, and what the operators themselves document, see the article on wrong facts about a business.

The week after: fixing at the source

The sheet is only useful if something changes because of it. Work through it in this order, the same week, while the answers are fresh:

1. Fix each wrong fact where it is written, starting with your own pages: the brands page, the hours on every copy, a closed location's page and its listings.
2. Add the facts the answers had nothing for: a line on the brands page, a sentence about holds on the stock and pickup page, a service page that did not exist yet. Stating availability and pickup shows the shapes for the first two.
3. Check each changed page the way a fetch for an answer sees it: view it with scripts off and confirm the new words are in the HTML.
4. Ask the owners of pages you do not control, such as a directory or a maker's list of stockists, to correct what they show, and note the date you asked.
5. Keep the sheet. Next month the same twelve questions are read against it, and it becomes the first row of the store's panel.

What one afternoon can tell you, and what it cannot

It tells you where the store stood on one day: which questions named it, for what, and which facts were wrong. It cannot tell you whether that will hold next week, or whether a change you make next month worked. For that, the same questions have to run again, unchanged, every month.

Our own probe was a snapshot of the same kind, run at scale. Beside our audit of 9,483 small business websites, we put one buyer question per business to Claude, with web retrieval on, on one day. Claude named 14.9% of the 9,240 businesses asked about, and 17.7% of the 5,928 answers that came back as a clearly differentiated list. Those figures cover all 40 trades, with no retail-only cut, and they are directional for the same reason an afternoon is.

To read change, turn the afternoon's questions into a fixed monthly panel; the monthly reading for one store shows how, and today's sheet becomes its first row. For Google's own AI features, Search Console's generative AI performance report adds impressions of your links in AI Overviews and AI Mode by page, country, date and device, though not the queries behind them. The guide for retail stores shows where the check fits among the other work.