AI visibility for cafés: what being named means, and how to count it

What does it mean for a café to be named?

Four things can be recorded about any answer: whether the café appears, whether it is recommended or only mentioned, whether it comes first, and which other cafés share the answer. Our methodology records three of them and leaves out the line between recommended and mentioned; its visibility score is the share of the twelve questions in which the café was named, with first place and the other cafés set beside it. A café's reading needs two more, because a café answer carries reasons and facts.

The first is what the café was named for. An answer to "quiet coffee shop to work from near the university" that names you for your pastries has named you and has still not answered the person's question. The second is whether the facts attached to your name were right: the hours, the milk, the wifi, whether you are still open. A recommendation that sends someone to a locked door on a Sunday does you more harm than silence.

So a café's reading has a reason column beside the name column. After a few months it shows what the answers say you are good for, whether coffee, working, meetings, the bakery or the terrace. Whether that matches what you want to be chosen for is a decision about your own pages.

The three kinds of question a café panel holds

A panel is a fixed list of questions, each one a prompt written the way people type it and asked every month without changes. For a café the questions fall into three kinds, and each tells you something different when its answers move.

Weight the panel toward the first two kinds, because they are asked by people who do not know you yet. The name questions show what gets repeated about you; when one of them comes back wrong, it goes to the fact check for your café to be traced and fixed, not into the score as a bad month.

Writing café questions that give the same reading next month

Two words in café questions need handling. The first is "near me". It means wherever the person asking happens to be, so the same words typed at your counter and across town are two different questions. Write the place into the question instead: the neighborhood, the station, the landmark people give when they give directions.

The second is "now". "Open now" asked at three on a Tuesday and at eight on a Sunday are different questions too, and only one of them tests your Sunday hours. Put the time into the words ("open sunday at 8am"), so the question means the same whenever it is asked, and keep one line for each time you care about: early on a weekday, a weekend morning, a holiday.

Seasonal lines need a decision. An iced-coffee question belongs in the panel only if you will ask it every month of the year; one added in June and dropped in September leaves a hole in the series. How to keep the file and what to record for every answer is in setting up prompt monitoring. How many runs make a reading, and how to tell a change from noise, is in measuring AI visibility.

What to write down for each café answer

The standard record covers whether you were named, recommended and first, and who else was named. A café's record needs three more fields, filled in by the person reading the answer:

• The reason given. The words the answer attaches to your name, copied as written. Over a few months this column shows what you are named for, and whether it is what you want.
• The facts stated. Hours, milks, wifi, prices, the address. Mark each one right, wrong or out of date; a wrong fact goes on your fix list rather than into the score.
• Who was named for your attribute. When the question is about working or oat milk and you are absent, the café named in your place is the one whose pages to read next. Who is named, and how often, is your share of voice for that question.

Keep the answers themselves too, pasted in full. A reason column filled in from memory drifts toward what you hoped the answer said.

Reading a first baseline without overreading it

A first reading that rarely names you would be ordinary. In our audit of 9,483 small business websites (September 2026), one buyer question for each of 9,240 of the businesses went to Claude with web retrieval on, and 14.9% of the businesses were named. Among the 5,928 answers that gave a clearly differentiated list rather than a hedge, 17.7% were. That covers all 40 trades at once, with one assistant, one question and one day; it is not a café rate, and it describes no assistant but Claude.

The first baseline is for comparison, and it reads best by kind of question. A café the answers describe correctly when asked by name, but never mention for an attribute, has pages that say who it is and not what it offers. A café named for its early hours but never for oat milk has hours a machine can read and, perhaps, a menu it cannot. Each pattern points at a different page.

Then set the reason column beside the list of facts on the café industry page. If the answers never mention working and you want laptop guests on weekday afternoons, the working section is the page to write first, as the laptop article sets out. If they name you for the terrace in January, a seasonal fact has outlived its season.

What the reading cannot tell a café

It does not count cups. Someone who hears your name from an assistant and walks in leaves no trace in the reading, and the register does not record why they came. The reading says whether the answers name you, for what and with which facts; sales say the rest, and the two only meet when you ask customers how they found you.

It covers only the questions in the panel and the four assistants in it. That is the reason to build the panel from what customers ask your staff, about milks, seats and Sunday hours. A question nobody asks is a line that measures nothing.