What 9,483 small business websites actually serve to an AI crawler

Gaash — Sun, 25 Oct 2026 11:32:07 GMT

I pointed a crawler at 11,060 small business websites the way an AI assistant reads them: no JavaScript, three pages per site. 9,483 answered. Most of them give an assistant very little to work with.

What the crawler found

Finding	Share of sites
No question-and-answer content at all	61.5%
No review signal a crawler can read	64.7%
Name, address or phone inconsistent	43.1%
No postal address a machine can parse	34.1%
No schema.org markup at all	17.8%
Behind a bot wall that served a challenge page	7.2%
AI crawlers blocked on purpose in robots.txt	3.2%

The pages aren't thin. The median homepage has 880 words. The words just aren't in a shape a machine can quote, attribute or check.

Four things stood out

Questions without answers. Half the sites (50.7%) have a heading phrased as a question. They just don't answer it in a way a machine can lift out.

Reviews nobody can see. Most of these businesses have dozens of Google reviews. Only 8.9% put rating markup on their own site, so the assistant reading it never learns about them.

Schema about the website, not the business. 82% have some markup, mostly WebSite and BreadcrumbList that the page builder added. LocalBusiness shows up on about a quarter.

Blocking is mostly accidental. Only 3.2% block AI crawlers on purpose. 7.2% block them by accident, behind bot protection nobody configured for it.

Worst trades for missing schema: real estate agents (33.6%), photographers (32.7%), tax advisors (31.0%).

Does the assistant name them?

I asked Claude, with web search on, one buyer question for each of 9,240 of these businesses ("Who are the best electricians in Brampton?").


Business named	14.9%
Named, when the answer was a clear list	17.7%
Names that were directories like Yelp	0.6%

That last row matters. Google's first page for these questions is mostly directories. The assistant's answer almost never is, so a small business is competing on what its own site says. It's one question, one assistant, one day, so read it as a direction rather than a rate.

What to fix, in order

One name, address and phone, written the same way on every page
LocalBusiness markup with address, phone, hours and area served
Reviews on the page as text, plus AggregateRating
The five questions customers ask on the phone, as headings with real answers
A check of what a crawler actually receives (7% are behind a wall they don't know about)

The free visibility check at gaash.ai runs this same fetch on your site.

Method: user agent gaash-audit/1.0, no JavaScript, homepage, contact and about pages, robots.txt and llms.txt. The sample comes from businesses found per metro and trade, so percentages describe this sample. Full methodology. Originally published at gaash.ai.