Named at the right hour: measuring a pharmacy in AI answers

Named, recommended, first: reading one answer for one need

Every answer on the panel can get three verdicts; our methodology records two of them, named and first. Named: the pharmacy appears anywhere in the answer, under any form of its name, even in a closing line such as "[Pharmacy name] on [street] is another option". Recommended: the answer puts it forward for the need in the question, as the place open past ten, say, or the one that delivers to that neighborhood. First: it is the first pharmacy the reader meets.

Each verdict belongs to a line, and a line is a need. The same pharmacy can be first for "pharmacy that delivers to [neighborhood]" and absent from "pharmacy open on [the holiday]" in the same month, and both readings are true. Keep them line by line. The visibility score set out in the method does not combine or weight them: it is the share of the twelve lines in which the pharmacy was named.

The base record every answer gets, from the date and the run to the answer pasted in full, is in the guide to building a panel and its record. This page adds the pharmacy's part: lines that carry an hour, and three columns that check the time each answer gave.

Four needs, and the time each one carries

The needs in a pharmacy question each carry a time, and it is a different time for each. Open late means today's closing time. A holiday means that date's hours. A vaccination means the days it runs and whether it is walk-in. A transfer or a delivery means how long it takes and when the cut-off falls. An answer can be right about your pharmacy and wrong about the time, and the time is what the reader acts on.

Each of these can change on a given date: a holiday, summer hours, a pharmacist's day off, a new delivery round. So the check in this article is made on the day a line runs, against the hours your pages state for that day, and not against your usual week.

A panel written by service, hour and neighborhood

Write twelve lines, which is how large our panels for paying clients are. Each line names a need, an hour and a place, typed the way someone nearby would type it, with the neighborhood as locals say it. Everything in brackets is yours to fill. The lines are templates, not questions we recorded, and the first full run of your question panel is your baseline.

Each line names its hour instead of saying "now", so a line run at noon still tests the evening. Line 10 names you on purpose: it tests what an answer tells someone who already has your name and wants to know one thing. And every line is about finding a pharmacy. What to take, or whether a medicine suits someone, is a question for a pharmacist, and the panel does not measure it; a line that named a medicine would also be measuring what answers say about the medicine, which is not yours to change.

Neighborhoods get a line each, and only where you want customers from, named the way the people there name them: the station, the hospital, the street everyone knows. A neighborhood you add later becomes a new line with its own start date, so its months never mix with the others'.

Three columns a pharmacy adds to each record

The base record holds the verdicts, every other pharmacy named, and the answer in full. A pharmacy adds three columns, each filled from the answer as it was written and then checked against your own pages.

"Right that day" is the column that makes this a pharmacy's reading. The check is against the date of the run, not a typical week. On [the Monday of a holiday weekend], the right hours are the holiday hours your page posted, and an answer that gives your usual Monday is wrong on that day, though right on every other Monday. Keep "not stated" apart from "wrong": an answer with no hours leaves the reader to call, and one with the wrong hours sends them to a shut door.

The cited page, the answer's citation for you, says where to start: your own hours page, a directory entry, a map listing. When the column is blank it tells you nothing about your pages, because an answer that shows no source may rest on what a model learned in training.

Reading the first month, times before totals

Read the hours column across every line before you look at the score. A score that rises while the hours in it are wrong means more people sent to your door at an hour you cannot serve them. The fixes are on your own pages and listings: the late-night and holiday article covers the after-hours page, the overview for pharmacies lists what else the site states, and tracing a wrong fact covers finding the page that still says it.

What a month of readings cannot tell a pharmacy

It cannot tell you how many people asked. Nobody outside the assistant companies can see that, and the panel does not pretend to. It tells you what an answer says when one of your lines is asked. Answers vary between runs, so one answer is a snapshot: the weight sits on the pattern across the lines and across months, and a panel you run yourself can ask each line more than once.

Nor can it tell you why an answer changed. A better month after you rewrote the hours page may be the rewrite, a change an operator made, or ordinary variation. What the record gives you is the order of events: the date the page changed, and the answers before and after it.

The only published number that touches any of this is ours, and it is not a pharmacy number. In our audit of 9,483 small business websites, one buyer question per business went to Claude with web retrieval on, for 9,240 of them, and 14.9% of the businesses were named. It is directional, one question to one assistant on one day, across all 40 trades; we publish no pharmacy-only cut. For your pharmacy, the reading that counts is your own panel, run the same way each month, by hand or through monitoring at $290 a month, which you can cancel in any month.