Anonymised and composited, per the usual rule. The pattern is exact; the business is a blend.

The business

A Tucson plumber — call it Saguaro Plumbing & Rooter — fifteen years old, two owners who still answer the phone, and a review profile that most competitors would trade a truck for: a few hundred on Google averaging 4.9, a healthy Yelp page, a BBB accreditation. In the ordinary local-search world this business was winning. It ranked in the map pack for its main terms. The phone rang.

The younger owner ran Duly Noted because a nephew told him "people use ChatGPT for this now." He assumed, reasonably, that the reviews would carry over. They did not.

The report

Overall 41. AI visibility 15 — named once, by Claude, and linked once, by Perplexity, and absent from the other ten cells. Site readiness 80, an A, which is what made the visibility number so odd. The site was in better technical shape than almost anything we see: fast, clean schema, a real FAQ, name-address-phone everywhere, a sitemap, About and Contact pages, no JavaScript nonsense.

And one failing check, weighted nine points: AI crawlers blocked in robots.txt. GPTBot and ClaudeBot, specifically, with Disallow: /.

How it got there

Nobody at the company had touched robots.txt. In 2023 the site's security plugin shipped an update that added an "AI scrapers" option and enabled it under a "recommended" preset the previous web person had applied. The plugin's dashboard showed it as a green tick — "AI bots blocked" — and a green tick reads as a good thing. Eleven months of the two crawlers that feed the most-used assistants being turned away at the door.

Perplexity, whose bot wasn't on the plugin's list, could still get in, which is why it produced the one linked result. Claude's single named result almost certainly came from third-party pages — the reviews and directories — because ClaudeBot couldn't read the site itself.

Why reviews didn't save it

This is the part worth sitting with. The business had done everything the last decade of local SEO advice said to do, and it was invisible to the new channel anyway. Reviews, it turns out, are evidence the assistants use to rank businesses they already know about. They are not, on their own, how the assistants learn that a business exists in a category and a place. That learning comes mostly from reading the business's own pages and the pages that link to them. Lock the crawler out and the reviews are a reputation with no name attached.

The competitor domains in the report told the same story. Two Tucson plumbers with far fewer reviews appeared in six and four cells. Both had ordinary sites — nothing the audit would score above a B — and both were readable by every crawler.

The fix

Fifteen minutes. Open the security plugin, find the bot-blocking tab, turn off the AI option, confirm robots.txt no longer carries the lines, add explicit Allow groups so a future update can't silently re-add them. That was the entire technical scope.

We also recommended two smaller things while we were there: adding the Google Business Profile URL to the schema's sameAs list, which the site had left empty, and publishing llms.txt with the strong FAQ pages listed. Neither was the problem. Both were cheap.

What happened next

The re-run four weeks later isn't part of this note because the point isn't the number; the point is the shape. A business can be excellent, well-reviewed, technically sound and absent, for a reason that has nothing to do with quality and everything to do with a default nobody chose. The readiness score gave that reason nine points out of a hundred. It should probably be more.

The lesson

Before you spend on anything — content, links, ads, a new site — open yourdomain.com/robots.txt and read it. If the words GPTBot or ClaudeBot appear above Disallow: /, you've found your first fix and it costs nothing. The free scan on this site does the check in a few seconds, and it's the first check it runs for exactly this reason.