llms.txt is a proposed convention: a plain text file at the root of your site describing what the site is and where the important pages are, written for language models rather than for browsers. Today it does very little, because no major AI system has committed to reading it. It costs about twenty minutes to add. That combination makes it worth doing and not worth being excited about.
This article exists because you will encounter people selling it as essential, and it is not.
What it is
A markdown formatted text file served at /llms.txt. It typically opens with the site name, a short description of what the organization does, and then a curated list of the most important URLs with a sentence explaining each.
The idea is reasonable. A model retrieving information about your business has to work out what you are from whatever pages it happened to fetch. A concise, authoritative statement written by you removes ambiguity. It is roughly the same argument as structured data, in prose.
What it actually does today
Honestly, very little that can be demonstrated. No major assistant has publicly committed to consuming it. There is no verified case of a citation traceable to one. Anyone telling you it drives AI visibility is describing a hope.
What can be said is that it is cheap, harmless, and positioned for a future where retrieval systems do look for a machine readable summary. The cost of being early is twenty minutes. The cost of being late, if it becomes conventional, is a period of being less legible than competitors who bothered.
Where the actual leverage is
If you have limited time, spend it here first, in this order:
- Make sure your content exists without JavaScript. This is the difference between being ranked lower and being absent entirely.
- Answer questions directly and early on the pages themselves. This is what actually determines whether a page gets cited.
- Publish a connected schema graph so your business resolves as a single unambiguous entity.
- Then add llms.txt, because by that point it takes twenty minutes and you have nothing better queued.
A perfect llms.txt on a site whose content is invisible to crawlers achieves nothing. The order matters.
Writing one worth having
Keep it short. A model reading it wants the shape of the business, not marketing copy. State plainly what you do, where you operate, who you serve and what you do not do. Exclusions are unusually useful, because they prevent a model recommending you for work you cannot take.
List your genuinely important pages with one line each. Not every page. The ten to fifteen that define the business.
Then keep it accurate. A stale llms.txt describing services you dropped is worse than none, because it is an authoritative statement that is wrong.
This site’s version
palmprojects.com serves one at /llms.txt. It is generated by the theme rather than maintained by hand, so it cannot drift out of date when a service page changes. It states what the business does, where it operates, that there are no case studies on the site, that there is no public address or Google Business Profile, and which verticals are not served.
That last part is the most useful line in the file, and it is the kind of thing a hand written marketing summary would never include.
How to tell whether it starts mattering
This is worth revisiting rather than deciding once. Two signals are worth watching.
The first is server logs. If AI crawler user agents begin requesting /llms.txt specifically, that is direct evidence the file is being consumed rather than ignored. Most managed hosts will give you log access on request, and this is a question logs answer definitively.
The second is whether the major assistants publish anything about supporting it. Right now none have committed. If one does, the convention will become worth taking seriously very quickly, and the businesses that already have an accurate file will simply be ready.
Until then, treat it as what it is: twenty minutes of insurance against a plausible future, sitting at the bottom of a list of things that demonstrably work today.
Questions
Will this hurt anything?
No. It is a text file. Search engines ignore it, it is not indexed, and it has no effect on classic rankings in either direction.
Should I list every page?
No. The value is curation. A file listing four hundred URLs is a sitemap with extra steps, and you already have a sitemap.
Is this a ranking factor?
No, and be suspicious of anyone who says it is. It is a legibility aid with an uncertain future, which is a much less exciting claim and a more accurate one.
If you want the rest of your AI search foundations checked rather than just this one file, that is the engagement.
