llms.txt Best Practices
Most llms.txt files fail in the same few ways: they list too much, describe too little, or go stale. The practices below are the habits that separate a file a model can actually use from one that adds noise — paired with honest caveats about what an llms.txt can and cannot do.
The ten practices
Curate, don't catalog
Include the pages that best represent what you do, not every URL you have. A shortlist signals priority; a full dump erases it. If a section runs long, you are including too much.
Write descriptions that add meaning
The note after each link is where the value lives. Say what the page is and why it matters in plain language. A described link tells a model something a bare URL never could.
Lead with a precise summary
Your blockquote is the first and most-read line. Make it a plain description — what you do and for whom — not a slogan. “Fee-only planning in the Pacific Northwest” beats “we put clients first.”
Use absolute URLs
Always link with full https URLs. A model reading your llms.txt out of context may not know your base domain, and relative paths become meaningless.
Order by importance
Put your most important sections and links first. A model scanning top-down encounters your priorities in order, and any truncation hits the least important material last.
Keep it fresh
A stale file pointing at moved or deleted pages is worse than none. Re-check after structural changes and prefer generating the file from the same source as your site so it stays in sync.
Serve it as plain text
Make sure /llms.txt returns text/plain, not HTML. Confirm it loads at the exact root URL and is not intercepted by your framework's routing.
Never treat it as access control
llms.txt is advisory and public. It cannot block, gate, or hide anything. Use robots.txt or server rules for access, and never link private material from llms.txt.
Match it to your robots.txt intent
If you disallow certain AI crawlers in robots.txt, do not undercut that by highlighting the same areas here. Keep your access decisions and your curation consistent.
Set realistic expectations
A clean llms.txt improves how easily your content can be understood. It does not guarantee ingestion, citation, or ranking. Treat it as good hygiene, not a growth lever.
A good file, illustrated
The practices above produce something like this — lean, described, absolute-URL, ordered by importance:
# Meridian Dental > Family and cosmetic dentistry in Asheville, NC, > accepting new patients. ## Services - [General dentistry](https://example.com/general): Cleanings, fillings, and exams. - [Cosmetic](https://example.com/cosmetic): Whitening, veneers, and bonding. - [Emergencies](https://example.com/emergency): Same-day urgent care. ## Patients - [New patients](https://example.com/new-patients): Forms, insurance, first visit. - [Contact](https://example.com/contact): Hours, location, and booking.
The honest caveat
It would be easy to imply that a perfect llms.txt buys you AI citations. It does not. The llms.txt proposal is young, adoption across AI platforms is still emerging as of mid-2026, and there is no enforcement mechanism or guarantee that any particular system reads your file. What these practices do is make your content as clear and easy to consume as possible for whatever does come looking — which is the most any advisory, curation-focused file can promise. Do the work because it is good hygiene, not because it is a lever you can pull.
Frequently Asked Questions
What is the single most important best practice?
Curate ruthlessly. The value of llms.txt comes from restraint — a short, ranked list of your most representative pages. If you include everything, you communicate nothing about priority.
How often should I update llms.txt?
Whenever your site changes materially — new services, renamed pages, a redesign, a hosting migration. Quarterly review is a reasonable floor for stable sites; automated regeneration on publish is ideal.
Should I stuff keywords into the descriptions?
No. Keyword stuffing reads as noise to a language model and adds no meaning. Write plain, accurate descriptions. Clarity outperforms density here.
Can best practices guarantee I get cited by AI?
No. Following these practices makes your file clean, current, and easy to parse, which improves your odds. But adoption is still emerging and there is no guarantee any AI system reads or uses your file. Treat it as one signal among many.
Is a longer llms.txt better than a short one?
Usually the opposite. A focused file of well-described key pages tends to serve a model better than an exhaustive one. Length is not a virtue; relevance is.
Put the practices to work
Generate a lean, well-described, absolute-URL llms.txt for your domain — best practices baked in.