Stop 5 of 9 · Weekly
Quality checks with Claude in Chrome
The same eleven questions on twenty sites every week, a score per site, and the only output Dany reads: what changed since last week.
The same eleven questions, twenty times
Dany used to check client sites when something felt wrong, which meant she checked the loud clients and not the quiet ones. Claude in Chrome walks a live site the way a visitor does: it follows the links, reads the pages, submits the test form and reports what it found. Twenty sites, one checklist, once a week.
The point is not the report. It is that the same questions get asked of Aria Dance Studio as of the client who calls every Tuesday.
The checklist
| The question | The failure it catches |
|---|---|
| Every page loads, over https, no error pages | An expired certificate nobody noticed |
| Links, menus, footers | A menu item pointing at a deleted page |
| Phone, email and hours against the client row | Old hours on the contact page since the refit |
| The contact form, submitted end to end | A form that says thank you and delivers nothing |
| 390 pixels wide, on every template | Text off screen after a content edit |
| Titles, descriptions, one h1 per page | A page listed in search as an address |
| The heaviest file on the page | A four megabyte photo the client uploaded |
Eleven questions in practice, once you split the templates. Write them once, keep them in the studio Project, and change them there rather than in each week's prompt.
Walk every page of the site in this tab as a visitor. For each page report: pages that do not load or return an error, dead links, any phone number, email or opening hours that differs from [paste the client row from clients.csv], text running off screen at 390 pixels wide, images with no alt text, missing or duplicate page titles, and the heaviest file with its size. One table, one row per problem, worst first. Do not submit any form yet and change nothing.
Compare what you just found on this site with last week's table for the same client, which is attached. Tell me three things: problems that are new this week, problems that were there last week and are now gone, and anything that got worse. If nothing changed, say so in one line and stop.
Turn this week's findings across all twenty sites into three lists: what Mo fixes in a pull request, what Tess fixes as a content edit, and what needs the client before anyone can start. Each line names the client slug, the page and what a visitor loses. Sort each list by what a visitor loses, not by how quick it is.
The score and the trend
Give each site a number out of eleven for how many checks passed. Nobody cares about the number itself, but the direction is real. A site drifting from eleven to eight over two months is a client whose staff have been editing pages, and that is a conversation about training or about a bigger retainer, not a bug.
The measurable
Two numbers a week: how many of the twenty sites the pass reached, and the total number of open findings across the studio. The first should be twenty. The second should fall for six weeks and then flatten, and the week it starts climbing is the week the studio took on more sites than it can carry.