A vendor pitched you an "AI readiness" package and llms.txt was on the list, or a marketing newsletter said every site needs one now. You want a straight answer: what is it, does anything read it, and is it worth an afternoon.
Straight answer: llms.txt is a proposed convention, not a standard, and no major AI search engine has said it uses the file. It costs almost nothing to publish and it is harmless. It is also not a substitute for the work that demonstrably matters, which is being crawlable, being fast, and stating facts about your firm plainly on your pages.
What llms.txt is, in one file
llms.txt was proposed in September 2024 by Jeremy Howard of Answer.AI. The idea is a plain Markdown file at the root of your domain (yourfirm.com/llms.txt) that gives a language model a short, curated map of the site: what the organization is, which pages matter, and where the clean text version of each lives.
The proposed format is simple. An H1 with the site or firm name, a blockquote with a one-paragraph summary, optional paragraphs of context, and then H2 sections containing bulleted links, each with a short description. A companion file, llms-full.txt, can hold the full text of the important pages concatenated together so a model can read everything in one request.
That is the whole specification. It is deliberately human-readable, and you can write one in a text editor.
Who reads it today (the honest answer)
As of this writing, OpenAI, Anthropic, Google and Perplexity have not documented support for llms.txt in their crawlers or search products. Google's John Mueller has compared it to the keywords meta tag, meaning a self-declared signal that search engines learned not to trust. Adoption so far is concentrated among developer documentation sites and the tools that generate docs, where the file is useful for coding assistants that fetch it deliberately.
For a law firm, the practical status is this: some AI agents will fetch the file if a user points them at it, some crawlers occasionally request it, and none are known to weight it in deciding which firm to cite. If you see it in your server logs, that is interesting. If you do not, nothing is broken.
We say this plainly because the file is being sold as more than it is. Publishing llms.txt does not put you in AI answers. The mechanics of how Perplexity and similar engines choose sources run on retrieval of ordinary pages, and that is where the return is.
What a law firm's llms.txt should contain
If you publish one, make it accurate and small. A workable structure for a firm:
- H1: the firm's legal name.
- Blockquote: one paragraph stating what the firm does, where it is, and who it serves (for example, a personal injury firm with offices in Richmond and Norfolk, Virginia, representing injured people in motor vehicle and premises cases).
- A "Practice areas" section linking each practice area page with a one-line description.
- An "Attorneys" section linking each bio with name and role.
- An "Offices and contact" section with the office pages and the contact page.
- An "Articles" section with the five to ten guides you most want a model to read, not the whole blog.
Write the descriptions the way you would explain the page to a new paralegal: specific, factual, no marketing adjectives. Do not include client names, case results, or anything you would not put on a public page, because the file is public. Keep the firm name, address and practice area wording identical to what is in your schema markup and on the pages themselves; a model that finds three phrasings of your firm name has learned less, not more.
llms.txt versus robots.txt versus schema
These get conflated, and they do different jobs. robots.txt controls which crawlers may fetch which paths. It is where you allow or block GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot and Google-Extended. It is honored by well-behaved crawlers and it is the file that determines whether you can be cited at all.
Schema markup, in JSON-LD on each page, describes the entities on that page in a vocabulary Google and others actually parse. It is the established way to state that this is a LegalService at this address with these attorneys.
llms.txt is a suggestion to a model about where to look. It has no enforcement and no confirmed consumer among the big search products. Put the three in that order of priority, and make sure robots.txt is not blocking the very crawlers you are trying to attract with the other two.
What to do before you bother with it
Confirm the site renders its main content without JavaScript, because a crawler that cannot see your practice area text has nothing to cite. Confirm response times are reasonable. Confirm each important page states its core facts in the first paragraph. Confirm firm name, address and phone are identical across the site, Google Business Profile, the state bar and the major directories. That list is the substance of technical SEO for law firm websites, and it is where a limited budget belongs.
Once that is in place, spend the afternoon on llms.txt if you like. Publish it at the root, link it from nowhere in particular (it does not need a menu entry), and revisit it when you add a practice area or an attorney. The broader picture of what moves the needle in AI search is on our AEO for lawyers page.
Questions we get
Can llms.txt hurt our site?
Not in any way we have seen. It is a text file that ordinary search engines ignore. The only risk is publishing inaccurate or stale information in it, which is the same risk as any page.
Should we put our full site text in llms-full.txt?
Only the pages you would want quoted. A full dump of every blog post since 2014 gives a model more noise than signal. Ten well-written practice area pages and guides make a better file than three hundred posts.
Is there a WordPress plugin for it?
Several plugins generate the file automatically from your page list. They work, but they produce generic descriptions. A hand-written file of forty lines is better than an auto-generated one of four hundred, and it takes less time than you would expect.
If you are deciding where an AI search budget should go first, tell us what you are working with and we will give you a candid order of operations.