llms.txt: How to Create One (With a Template)

llms.txt is a plain markdown file served at your site's root, example.com/llms.txt, that lists your most important pages with one-line descriptions. It is a companion to robots.txt, not a replacement: robots.txt tells crawlers what they may fetch, while llms.txt tells compatible clients which pages are canonical and worth reading first. No web standard requires the file, and no AI system is obligated to look for it.

Because it is ordinary markdown, it loads instantly, survives JavaScript failures, and can be read by any client that understands the format, a model, an agent, a summarizer, or a person. It sits alongside the layers most sites already have: robots.txt for access, sitemap.xml for URL discovery, and JSON-LD schema for entity facts. llms.txt adds one thing those do not: a curated reading order with context.

The cost of trying it is close to zero: a file of a few hundred bytes, no effect on page speed, and removal is a single delete. That low cost is exactly why the honest framing matters, it is easy to publish one on a Friday and mistake the activity for a result.

llms.txt earns its keep when a site is deep enough that a visitor would benefit from curation. A services firm with a dozen core pages, published pricing, an FAQ, a security page, and a knowledge base is the ideal candidate: a client landing on the root gets a guided map instead of guessing from navigation menus, and it gets that map on the first request, before any crawling happens.

The honest framing matters more than the file. Adoption sits at roughly 10% of sites, and there is no measured citation or ranking impact from publishing one, no published study shows that adding llms.txt changes how often assistants cite you or send traffic. Ship it as preparedness and curation: cheap to write, trivial to maintain, useful to any client that reads it, and neutral for clients that do not. Never ship it as an SEO hack or report it as a traffic lever.

llms.txt is a good fit when:

Keep it under one screen. The file is a map, not a mirror of your site, the depth belongs on the pages it links to. This structure works for most services firms:

# Company Name, services, pricing, and resources

> Who you are and who you serve, in one or two sentences.

## Company

- [About](/about), team, story, and how you work.

## Services

- [Core service](/services), what it is, who it is for, and what it costs.

## Pricing

- [Pricing](/pricing), packages and starting prices.

## FAQ

- [FAQ](/faq), answers to the questions buyers ask most.

## Security

- [Security](/security), data handling, access, and compliance.

## Contact

- [Book a call](https://calendly.com/chat-aiagentship/30min), a 30-minute design tour session.

Two rules keep it useful. One line per link, written as a neutral description of what the page contains rather than marketing copy. And canonical URLs only, point at the page you want described, not a campaign landing page that will be gone next quarter.

No tooling required; a competent writer can produce a useful file in about thirty minutes:

Maintenance is the part people skip. A stale llms.txt, links to retired pages, old pricing, a superseded service name, teaches clients the wrong things about you, which is worse than teaching them nothing. If nobody will own the quarterly review, do not publish the file: a small, correct map beats a large, wrong one.

llms.txt is the top layer of an agent-ready site, not the foundation. What matters more is underneath: semantic HTML that carries the real content, JSON-LD that names your entities (see schema markup for AI discovery), an accessible robots.txt, and pages deep enough to summarize. A map to a city that does not exist helps no one.

The Agent-Ready Website Audit checks exactly this stack, Discovery, Data Quality, and Actionability across roughly 40 checks, and returns a prioritized fix list with copy-ready code, including whether an llms.txt exists and whether its links still resolve. If that list is longer than your team's bandwidth, the Agent-Ready Website Build starts at $2,500 and ships the whole layer; full multi-page builds run $15,000-$25,000, see pricing for the range.

One honest limit: none of this guarantees a citation, a ranking, or a visitor. It removes the mechanical friction between your content and the systems that read it, which is the only part you control.