llms.txt: the new file every site should ship in 2026.

An llms.txt file is a short plain-text file you place at the root of your website that tells AI models what your business does, which pages matter, and where to find your best content. It sits at yoursite.com/llms.txt, uses simple Markdown, and reads more like a guided index than a technical config file. If you have ever wished you could hand ChatGPT or Perplexity a one-page briefing on your firm, that is close to what this file is for.

The idea is new. The problem it solves is not. AI answer engines now sit between you and the people searching for you, and they need a clean way to understand your site. llms.txt is the community's first serious attempt at giving them one. It is not a rule yet. But shipping one in 2026 costs you an afternoon and risks nothing, so we think most sites should.

What is llms.txt and why did it appear?

llms.txt is a proposed standard for a Markdown file that summarizes your site in a way large language models can read quickly. It was proposed in late 2024 by Jeremy Howard, and the format is deliberately plain: a title, a short description, and a set of linked sections pointing to your most useful pages.

The reason it appeared comes down to how AI models read the web. A model has a limited context window. It cannot swallow your entire site, your navigation menus, your cookie banners, and your footer legalese and still reason clearly. Your homepage HTML is full of code the model has to wade through to find the sentence that actually matters. llms.txt cuts that noise. You write the summary yourself, in your own words, and point the model straight at the pages you want it to cite.

Think of it as robots.txt's friendly cousin. robots.txt tells crawlers where they may not go. llms.txt tells AI models where they should look and what they will find when they get there. One is a fence. The other is a welcome mat.

What does an llms.txt file actually contain?

An llms.txt file contains a heading with your business name, a blockquote summary, and lists of links to your key pages with a short note on each. That is the whole shape of it. The format rewards restraint, not volume.

Here is what a version for a San Diego family law firm might look like in plain terms. The top line is an H1 with the firm name. Below it, a one-line blockquote: "A San Diego family law firm handling divorce, custody, and support for clients across the county." Then a section headed Services, listing the divorce page, the custody page, and the support page, each with a five-word description. Then a section headed About, pointing to the attorney bios and the fees page. Then an Optional section for blog posts and guides the model can pull from if it needs more.

The descriptions matter more than the links. A model reading "Divorce — contested and uncontested filings, timelines, and what to expect in San Diego County" understands your page before it fetches it. A bare URL tells it nothing. Write the descriptions the way you would explain each page to a new client on the phone. Short. Concrete. No jargon.

There is a companion file some sites ship called llms-full.txt. That one holds the actual content of your key pages in one long Markdown document, so a model can read everything in a single fetch. Most small firms do not need it. Start with the index file. Add the full file later if your content is worth reading end to end.

Does llms.txt actually work yet?

No major AI engine has publicly confirmed it reads llms.txt as a ranking or retrieval signal in 2026. That is the honest answer, and you should hear it clearly before you spend a weekend on this. Google has said it does not currently use the file. OpenAI, Anthropic, and Perplexity have not committed either. Adoption is growing on the publishing side faster than on the reading side.

So why ship one now? Three reasons. First, the cost is near zero and the downside is nothing — a static text file cannot slow your site or break your pages. Second, standards get adopted after a critical mass of sites already support them, not before; the sites that shipped robots.txt and sitemap.xml early were ready when crawlers started reading them. Third, the act of writing an llms.txt file forces you to state, in plain English, what your business does and which pages matter. That clarity is useful whether or not a machine ever reads the file.

We treat llms.txt the way we treat structured data. We do not promise it moves the needle tomorrow. We ship it because it is cheap, honest, and pointed in the right direction. If AI engines adopt it, you are already there. If they do not, you spent an afternoon writing a clear index of your own site, which you needed anyway.

How llms.txt fits the rest of your foundations

llms.txt is one small file that only pays off when the pages it points to are fast, crawlable, and clearly written. The file is a signpost. If the road it points down is broken, the signpost does not help. This is where a lot of the AI-visibility conversation goes wrong — people chase the new file and skip the boring work underneath it.

AI engines still read your actual pages. They still care whether those pages load. A model or its crawler that hits a page taking four seconds to render will often move on before it gets what it needs. We have written before about why your slow site is a sales problem, not an IT problem, and the same logic applies to AI crawlers. Slow pages lose citations the same way they lose customers. If you want the mechanics of what "fast enough" actually means, the three numbers that decide if Google bothers covers the metrics that matter.

The order runs one way. Get your pages fast. Get them crawlable. Write them so a human and a machine both understand them in one read. Then ship the llms.txt file that points to the best of them. That order is the whole of our Search Foundations work — the file is the last step, not the first. Foundations first.

When we rebuilt the technical base for McShanes Solicitors, the win came from clean pages, fast loads, and clear service descriptions — the exact material an llms.txt file would index. The file is only as good as what it lists. Fix the pages, and the file writes itself.

How to ship your first llms.txt in one afternoon

You can write and publish a working llms.txt file in under two hours if your key pages already exist. Here is the order we use.

First, list your money pages. For most firms that is three to six: your core service pages and your about-or-fees page. Do not list everything. A file that points to forty pages tells a model nothing about which four matter.

Second, write one sentence for the whole business. This goes in the blockquote at the top. Say what you do, who you serve, and where. "A downtown San Diego dental practice offering general, cosmetic, and emergency care." One line. No adjectives you would be embarrassed to say out loud.

Third, write one line per page. For each money page, a short description that states what the page covers. Match the wording to how clients describe the problem, not how you file it internally. Clients search "tooth pain" not "endodontic consultation."

Fourth, add an Optional section. This holds your guides, blog posts, and FAQs — the material a model can pull from for deeper answers. Anything under this heading is explicitly lower priority, which tells the model where to look first and where to look second.

Fifth, save it as llms.txt and upload it to your site root so it lives at yoursite.com/llms.txt. Confirm it loads in a browser. That is the ship. Update it when you add or retire a major page — twice a year is plenty for most firms.

One mistake to avoid: do not stuff the file with keywords. It is not a place to game a model. Write it like a helpful index, because that is what it is, and because a model reading forced keyword soup will trust it less, not more.

Where this breaks down

llms.txt does not fix a site that AI engines already struggle to read, and it will not manufacture authority you have not earned. If your pages are thin, slow, or contradictory, a tidy index file just points machines at your weakest work faster. The file is also unratified — no engine owes you a citation for shipping it, and that may not change in 2026. We recommend it because it is cheap and honest, not because we can promise a result. If someone sells you an llms.txt file as an AI-visibility guarantee, keep your wallet closed.