Somewhere in the last year, “llms.txt” went from a niche idea a handful of developers were experimenting with to something showing up in client requests and Slack messages fairly regularly. Naturally, that means a wave of agencies started selling it as a must-have, and just as naturally, that means it’s worth being honest about what it actually does before you pay someone to build one.
What it actually is, in plain terms
llms.txt is a plain Markdown file you place at the root of your domain, similar in spirit to robots.txt, except instead of telling crawlers what they can and can’t access, it tells language models which pages on your site are the most important, and gives a short description of each. Think of it as a curated table of contents written specifically for a model to read, rather than the sprawling mess of navigation menus, footers, and cookie banners it would otherwise have to wade through.
The idea came from Jeremy Howard and the team at Answer.AI, proposed as a way to solve a real problem: sites are built for human browsing patterns, and that structure is often genuinely inefficient for a model trying to quickly understand what a site is and does.
What it does not do
This is the part that gets left out of a lot of the hype. There’s no confirmed, universal standard yet that says every major AI platform actively reads and prioritises llms.txt in how it ranks or cites content. Adoption is uneven. Some tools may reference it, others currently ignore it entirely. So if someone tells you it’s a guaranteed ranking factor for AI search the same way a sitemap is for Google, that’s overselling something that’s still genuinely emerging.
It also doesn’t replace good on-page structure, clear headings, or schema markup. It’s additive, not a shortcut around doing those things properly.
So why bother at all
Because the downside of having one is close to zero, and the upside, even if modest today, compounds as adoption grows. It costs almost nothing to create, it’s a single static file, and it forces you to do something genuinely useful regardless of whether any model reads it yet: sit down and clearly articulate, in plain language, what your most important pages actually are and why they matter. That exercise alone tends to expose gaps in how a site is organised.
There’s also a reasonable argument that being early to a low-cost, plausible standard is simply good practice, the same way plenty of sites added Open Graph tags years before every platform fully supported them.
What a decent one actually looks like
Keep it short. List your most authoritative pages, not every page on your site, with a one-line description of what each one covers and why a model should consider it a reliable source. Update it when your site structure changes meaningfully, the same way you’d revisit a sitemap.
Resist the temptation to stuff it with keywords or marketing language. The entire value of the file rests on it reading as a clear, honest index, not another piece of promotional copy competing for attention.
Our actual take
Build one. Don’t pay someone a large amount of money to build one, because it’s genuinely a couple of hours of work once you know your site’s key pages, and don’t expect it to move the needle dramatically on its own. Treat it as a small, sensible bet on where things might be heading, not a silver bullet you can point to instead of doing the harder work of good content and structure.
 Frequently Asked Questions
Is llms.txt an official standard supported by all AI companies?
Not yet in any formally agreed way. It’s an emerging convention with growing but inconsistent adoption across different AI platforms.
Will adding llms.txt hurt my SEO if it turns out not to matter?
No, there’s no real downside. Worst case, it simply goes unused by current AI tools while costing you almost nothing to create.
How is llms.txt different from a sitemap?
A sitemap lists URLs for search engine crawlers to index. llms.txt is a curated, human-readable summary of your most important pages, written specifically to help a language model quickly understand your site.
Should every business create one right now?
It’s low effort and low risk, so there’s a reasonable case for most sites to have one, but it shouldn’t come at the expense of more established priorities like solid on-page SEO and structured data.
If you’re not sure where to start with yours, the honest first step is simply listing the five or six pages you’d want an AI model to understand about your business, and writing one clear sentence about each.
