Ask ChatGPT or Claude to recommend a vendor and it will pull an answer together from whatever it can read. Your website is part of that reading list. The question is whether the machine doing the reading actually understands what you do.
llms.txt is a small, plain text file that tries to help. It sits at the root of your domain and gives AI systems a short, curated introduction to your site. We publish one on citamark.com. This post covers what the file is, what belongs in it, how to write one, and what it can and cannot do for you.
What llms.txt is
llms.txt is a convention proposed in late 2024 by Jeremy Howard, a co-founder of the AI research lab Answer.AI. The idea is simple. You put a Markdown file at yourdomain.com/llms.txt. Markdown is a lightweight text format that both humans and language models read easily.
The file answers three questions for an AI system. Who is this site? What matters here? Where should it look first?
You already know two files that live in the same neighborhood. robots.txt tells crawlers what they are allowed to access. sitemap.xml lists every URL you want indexed. llms.txt does a different job. It curates. Instead of listing everything, it points to the handful of pages that explain your business, with a one-line note on why each one matters.
Why the web is hard for machines to read
A modern web page is mostly not content. Navigation menus, cookie banners, pop-ups, scripts, footers stuffed with links. A human filters that clutter without thinking. A language model has to work through it, and it does that work inside a limited context window, which is the amount of text a model can consider at once.
So when an AI assistant fetches your homepage to answer a question about your company, a large share of what it reads is plumbing. The signal is buried. llms.txt removes the digging. One clean file, no clutter, structured exactly the way models like to read.
What goes in the file
The proposed format is short:
- One H1 heading with the name of your site or company. This is the only required element.
- A blockquote directly under it with a one or two sentence summary of what the site is.
- A few optional paragraphs of context if the summary needs support.
- H2 sections, each containing a list of links. Each link gets the page name, the URL, and a brief description after a colon.
- An optional H2 section literally titled "Optional" for secondary links a model can skip when it is short on room.
A skeleton looks like this:
# Company Name
> One or two sentences on what the company does and who it serves.
Brief additional context if needed.
## Services
- [Service page name](https://example.com/service): What this page covers.
- [Another page](https://example.com/other): What this one covers.
## Optional
- [Blog](https://example.com/blog): Articles on related topics.
That is the whole thing. No tags, no code, no special tooling.
How to write yours
Six steps.
1. Write the summary first. Pretend you have two sentences to explain your company to a smart stranger. No slogans. What you do, who you do it for.
2. Pick pages, do not dump pages. Ten to twenty links is plenty for most businesses. This is a highlight reel, not a sitemap. Choose the pages that explain your services, your process, and your proof.
3. Write descriptions for the machine, not the market. "Pricing and plans for the scheduling product" beats "Discover flexible options built around you." Plain and factual wins here.
4. Use full URLs. Absolute links, including the https part. Do not make the reader guess the domain.
5. Move the nice-to-haves under Optional. Blog archives, press pages, careers. Useful, but not the first thing a model should spend its attention on.
6. Keep it current. Review the file whenever you add or retire a major page. An introduction that describes services you no longer offer does you no favors.
There is also a companion convention worth knowing: llms-full.txt, a bigger file that includes the full text of your key pages rather than links to them. Documentation sites use it most. For a typical business site, the standard file is the right starting point.
What it will not do
Here is the honest part. llms.txt is an emerging convention, not a ratified standard. No standards body governs it. The major AI companies have not committed to fetching it on any reliable schedule, and some people in the search world are openly skeptical that it matters yet. Some AI crawlers do request the file today. Some documentation platforms generate it automatically. Adoption is real but uneven, and nobody can promise you a traffic bump the week you publish one.
So why bother? Three reasons.
First, the cost is roughly an hour and the risk is zero. The file harms nothing.
Second, the file is useful beyond crawlers. Paste it into an AI assistant yourself when you want help writing about your business. Hand it to a contractor as an orientation document. It earns its keep even if no crawler ever reads it.
Third, writing it is a clarifying exercise. If you cannot summarize your site in two sentences and fifteen links, that usually says something about the site, not the file.
If the convention takes hold, you were early at almost no cost. If it fades, you spent an hour producing a summary you needed anyway.
We publish one on this site
We do not recommend things we skip ourselves. citamark.com/llms.txt is live. It introduces the company, links to the service pages, and stays in sync with the site.
That is part of how we think about websites in general. A site has two audiences now: the person deciding whether to trust you, and the machines that increasingly sit between that person and your front door. We build sites wired for both.
The bigger picture
llms.txt is one small piece of a larger question: when someone asks an AI assistant about your industry, do you show up, and are you described accurately? That practice is called GEO, or generative engine optimization, and it covers much more than one file. Clean page structure. Structured data that spells out who you are in a format machines parse directly. Consistent facts about your business across the web. Pages that answer the questions buyers actually ask.
If you want to know where your site stands, we run a fixed-scope AI discoverability audit that looks at exactly this.
See how your site looks to a machine
The Performance Audit is where we start with every engagement. We look at how your marketing actually runs, websites and search included, and hand you a dollar-weighted plan you keep whether or not you work with us. If AI assistants are starting to matter in how your buyers find vendors, and they are, it is worth knowing what those systems see when they look at you.