What llms.txt does, honestly.
A young convention, cheap to ship, and not a ranking lever. Here is what it is actually for.
What it is
A plain-text file at the root of a site that lists and describes its most important content for language-model consumers, in the spirit of robots.txt but aimed at comprehension rather than permission. It is a proposed convention rather than a standard, and adoption varies by system. Written in Markdown, it is meant to be read by a machine and to be legible to a person checking it.
What goes in it
For a roofing contractor, five things.
A file listing every URL on the site defeats its own purpose
What it does not do
It does not improve Google rankings. It does not compel any system to cite you. It does not replace pages that answer questions, and it will not make a thin site look substantial. It is a pointer, and a pointer to nothing is still nothing.
Why ship it anyway
It costs an hour, it is trivially maintainable, some retrieval systems read it, and there is no downside to having it. That is the entire case. It sits in the same category as a clean sitemap: unglamorous infrastructure that occasionally matters and never hurts.
Where the overselling happens
Anyone selling llms.txt as an AI-visibility strategy, or charging meaningfully for its creation, is overselling a text file. If a proposal lists it as a deliverable with a price attached, ask what else is in that line item. The file is a by-product of knowing what your important pages are, which is the actual work.
Keeping it accurate
An llms.txt that points at pages you have since retired or restructured is worse than not having one, because it advertises inattention to exactly the systems you wrote it for. Review it whenever the site structure changes, which for most contractors means once or twice a year rather than monthly.
What to do instead if you only have time for one thing
Write one page that genuinely answers a question nobody in your market has answered properly. That will do more for citability than any file at the root of the domain, and it will produce enquiries whether or not a single retrieval system ever reads it.
How it differs from robots.txt and a sitemap
Three files, three jobs, often confused.
It is about permission.
It is about discovery.
It is about comprehension.
A site missing the first two has bigger problems than the third.
A worked shape for a roofing contractor
One heading with the business name and a one-line description. A short paragraph naming the service area and the licence detail. Then a handful of linked sections: services, claims and storm process, city pages, guides. Beside each link, one line saying what question it answers. Under two hundred lines total, and considerably shorter for most contractors.
Who reads it today
Some retrieval and agent systems fetch it. Google has been clear it is not a ranking factor. Adoption is genuinely uneven and the honest position is that nobody can tell you precisely who reads yours. That uncertainty is the reason to spend an hour on it rather than a budget line.
Related guides.
Two acronyms for the same shift: being the source an answer engine quotes rather than a link it lists.
The retrieval and attribution mechanism, described without invented statistics.
How AI answer engines retrieve and cite sources, and what actually makes a roofing site citable.
Find out if your site is citable.
$2,500, and you see every answer a competitor is cited for and you are not.