Your buyer asked AI. You weren't named.

A facility manager with a leaking roof increasingly does not open ten tabs. They ask an assistant who to call.

So we ran a test on a real commercial roofing company in Austin. We put three questions a commercial buyer would actually ask to four different AI assistants, with live web search on. Twelve answers. We counted every company named in all of them.

The company was named in none of the twelve. Forty-six other roofing companies were.

Its website is not bad. It is well built, actively maintained, and ranks fine in classic search. That is what makes it worth writing about: everything that went wrong here is a setting, not a redesign. We ran the whole audit from the outside — nothing was requested from the company, nothing was installed, and no access was needed.

Finding one — the company is never named

The three questions: who the best commercial roofing contractors in Austin are, who to call about a leaking flat roof on a commercial property, and who offers commercial roof maintenance plans.

One company was named in seven of twelve answers, by all four assistants. The company we audited was named in zero, by none.

This is not a ranking problem. The company is not placing low in these answers. It is absent from them. The next three findings are the reasons why.

Finding two — eight AI crawlers are blocked

The site tells the major AI systems not to read it. Eight crawlers are told to stay out, and between them they supply the material behind most AI assistants.

Here is the part worth sitting with: this is almost certainly not a decision anyone made. The block sits inside a section marked as managed content, inserted automatically by the site's content delivery network — a platform default applied to a large number of sites at once. Most owners have never opened the file.

What it does and does not do. It does not affect classic Google rankings and it does not remove the business from the map. It excludes the company from the material these assistants learn from and reach for when composing an answer. One qualification, stated plainly: three live-retrieval agents are not blocked, so an assistant that fetches a page mid-conversation can still reach the site. The damage is to the far larger share of answers drawn from an assistant's existing knowledge.

The fix is a toggle in the delivery network's settings, reversible the same afternoon.

Finding three — nothing on the site is machine-readable

Structured data is how a business states, in a form a machine cannot misread, what it is, where it works and what it sells. We found none — not a partial implementation, but zero records across the homepage.

For contrast, this bar is reachable. Of five commercial roofers we examined the same way, one publishes the trade classification, the business, its location, its reviews and a questions page — missing only one record of the seven. The difference is not budget. It is that the facts were declared.

Finding four — five service pages, none declared

The most expensive part of this gap is that the work is already done.

Each is a real page describing a real service. None is declared as a service an assistant can read, and none carries a defined service area — despite the homepage naming four Texas markets it serves.

So when a facility manager asks which contractors handle commercial roof maintenance plans in Austin — question three above, which this company answers on a dedicated page — the assistants named two other roofers instead.

The content exists and is invisible. That is the cheapest gap in marketing to close.

What we would do

What we would not promise. Being named by an assistant is earned, not purchased, and reputation signals outside the website contribute. What these fixes guarantee is eligibility: today the company is excluded before the question is even considered.

How this was measured

Two things this does not claim. It does not measure Google Business Profile completeness, and it does not estimate how many buyers in these markets ask these questions. Both are measurable and neither is asserted here.

We have anonymised the company. The point is not that one roofer got this wrong — it is that the defaults got it wrong for them, and the same defaults are running on a great many contractor websites right now. The four checks above take an afternoon. Run them on your own site before you decide whether this matters.

Jonathan Michael

Founder, Maverick AI