What I found in 27 local-business websites
By Joshua Hopkins · Published June 16, 2026 · ~4 min read
In June 2026 I reviewed 27 independent local-business websites across three US metros — Chattanooga, Boise, and Des Moines — to see how ready they were for AI search. The businesses were ordinary home-services, dental, and salon shops, not tech companies. The pattern was consistent enough to write down, so here it is.
Six of the 27 sites were telling AI assistants to stay out
Six of the 27 sites blocked AI crawlers outright in their robots.txt file, refusing GPTBot (ChatGPT), ClaudeBot (Claude), and Google-Extended (Google's AI). In every case this looked like a website-template default, not a decision the owner made — the block was simply shipped with the site and never reviewed.
Two of those six had a welcome mat on a locked door
Two of the six blocking sites also published an llms.txt file — a plain-text summary written specifically to help AI assistants read the site. They were inviting AI in through the front door while their robots.txt bolted it shut. That contradiction is the clearest sign of unreviewed vendor work: two tools added at different times that flatly cancel each other out.
A second, hidden blocker lives at the server firewall
Blocking does not only happen in robots.txt. On four more sites whose robots.txt was wide open, the server's security layer (Cloudflare and similar) challenged or refused requests that identified themselves as AI crawlers. Counting both layers, 10 of the 27 sites turned at least one major AI assistant away somewhere on the path to the page. These four firewall results were measured with AI crawler labels sent from a single machine; a firewall that verifies crawlers by their real network address may still admit the genuine bots, so I treat firewall findings as "challenges unverified AI traffic" and confirm them before naming a specific assistant.
Seven sites gave AI no facts to work with
Seven of the 27 sites carried no structured data identifying the business — six had none at all, and a seventh published only a generic "this is a website" tag. Structured data is the machine-readable label that tells an AI who the business is, where it operates, and what it sells. Without it, an assistant has to guess from raw page text, and it often guesses wrong or skips the business entirely.
When one template is wrong, every business on it is wrong
Two dental practices in the same metro ran an identical website template — down to the same structured data and the same AI crawler block, character for character. When a vendor template carries a buried default like that, every business on it inherits the same invisibility without anyone choosing it. The problem scales silently across a vendor's whole client list.
The takeaway
Almost none of this was a decision. It was defaults — template settings and vendor leftovers nobody reviewed. That is the good news: defaults are cheap to fix. Most of the sites above could be made readable to AI with a two-line change to one file, which is exactly why my first check is free.
Methodology: a hand-run review of 27 independent local-business websites (home services, dental, salons and restaurants) across Chattanooga TN, Boise ID, and Des Moines IA, conducted June 2026 over plain HTTP. This is a field sample of 27 sites, not a formal study, and the metros were chosen for size and low agency saturation rather than at random. I'm happy to walk through the raw results. Background on the standards involved: the llms.txt specification and schema.org structured data.