THE SHORT ANSWER
An llms.txt file is a plain markdown document at the root of your domain that lists your most useful pages with a one-line description of each, so a language model can orient itself without crawling everything. Support is still partial rather than universal, but it costs about an hour to produce, carries no downside, and doubles as a genuinely useful internal map of what your site actually contains.
The proposal is simple enough to explain in a sentence: give a model the same courtesy you give a human visitor, which is a curated index rather than a raw crawl of every page you have ever published.
Whether the major assistants read it consistently is still an open question. Whether writing one is a reasonable use of an hour is not.
The numbers, at a glance
Location: the root of the domain, alongside robots.txt and sitemap.xml
Format: plain markdown, one H1, optional summary, then H2 sections of linked bullets
Length: typically fifty to three hundred lines; it is an index, not an archive
Realistic benefit today: partial and unproven; the reliable win is the internal clarity of writing it
The structure that works
An H1 with your company name. Nothing else on that line.
A blockquote or short paragraph summarising what you do, in one or two sentences, written the way you would want an assistant to describe you.
H2 sections grouping your content by purpose, such as core services, pricing, answers, and about.
Under each H2, bulleted links in the form of a markdown link followed by a colon and a one-line description of what the page contains.
An optional section for material you would rather was not used, so the exclusion is explicit rather than implied.
The descriptions carry most of the value. A link with no description is a sitemap entry; a link with a precise one-line summary is a retrieval hint. Write them as though someone will decide whether to open the page based on that line alone, because that is exactly the decision being made.
What to include, and what to leave out
Include the pages that answer questions: service pages, pricing, the answers hub, case studies with numbers in them, and the about page that establishes who you are. Include anything containing an original figure, since that is what gets quoted.
Leave out blog archives, tag pages, thin category pages, legal boilerplate and anything you would not want an assistant to summarise as representative of your business. The point of a curated index is the curation. A file that lists all four hundred of your URLs has thrown away the only thing it was for.
The honest assessment of whether it works
Adoption is real but partial. Several developer-facing tools and documentation platforms read llms.txt actively. The major consumer assistants have not committed to it universally, and there is no public evidence that publishing one produces a measurable citation lift on its own.
The argument for doing it anyway is asymmetric cost. One hour, zero risk, no downside if it is ignored, and a clear benefit if adoption continues. The argument against relying on it is that a correct robots.txt, a real XML sitemap and well-structured pages do far more work today, and no llms.txt compensates for missing any of those.
The hosting catch nobody mentions
Serving a file at a root path with a .txt extension is trivial on most stacks and awkward on a few. Several visual website builders route all unknown paths to a 404 or to a rendered page, and their redirect or custom-path features sit behind a paid tier.
There are three workarounds worth knowing. Upgrade the plan if the redirect feature is the only blocker. Put a proxy in front of the site and serve the file at the edge, which is free on most CDN plans. Or publish the same content as an ordinary HTML page at a predictable path and link to it from the footer, which is not the specification but is at least discoverable.
Writing yours today
List the twenty to fifty pages that actually answer a buyer question.
Write a one-line description for each, precise enough to choose from.
Group them under four to six H2 headings by purpose.
Open with a two-sentence company summary phrased as you want to be described.
Serve it at the root, confirm it returns plain text, and link it from your footer.
Want leads like this in your pipeline?
Flock runs the campaigns, screens the enquiries and hands you only the ones that match your service area, job size and capacity. You pay per lead, not per month.
Book a 15-minute fit check | See lead package pricing
Related answers
Frequently asked questions
Is llms.txt an official standard?
It is a community proposal with growing but incomplete adoption, not a specification backed by the major model vendors. Treat it as a low-cost bet rather than a requirement.
Does it replace robots.txt or sitemap.xml?
Neither. Robots.txt controls access, the sitemap aids discovery, and llms.txt suggests priority. They answer three different questions and you want all three.
What if my platform cannot serve a root .txt file?
Use a CDN proxy to serve it at the edge, or publish the same content as an HTML page at a stable path and link it prominently. Discoverability matters more than the file extension.
How often should it be updated?
Whenever you publish something genuinely worth citing, and otherwise quarterly. A stale index pointing at removed pages is worse than none.
NEED A CLEARER PLAN?
Let’s turn your next move into momentum.
Talk to us →