Does llms.txt actually work?

Honestly, not much, at least not for getting cited by AI engines today. llms.txt is a proposed standard: a markdown file at yoursite.com/llms.txt that hands language models a curated map of your best pages. The idea is sound, but as of mid-2026 no major AI provider, OpenAI, Anthropic, Google or Perplexity, has confirmed using it to decide citations, and people who check their server logs after adding one often see little or no crawler traffic to the file. Figures at Google have likened it to the old keywords meta tag. Where it does earn its keep is documentation sites, developer tools and emerging AI-agent workflows that read it directly. So add one if you like, it is quick and harmless, but treat it as housekeeping, not a lever. The things that actually move AI recommendations are crawlability, answer-first content and off-site authority.

What llms.txt is, and what it promises

llms.txt was proposed by Jeremy Howard of Answer.AI in September 2024. The format is deliberately simple: a title, a one-line summary, then markdown headings listing your most important links with a short note on each, saved at the root of your domain. The motivation is real, models work within a limited context window, and a normal web page is padded with menus, banners and scripts, so a clean index could help a model find the signal.

The promise to be wary of is the leap from “this could help a model parse you” to “this will get you cited.” Those are different claims, and today only the first is well supported.

The honest verdict for 2026

The balanced read is that llms.txt is a low-cost, sensible idea whose payoff for citations is currently modest and unproven. It has genuine traction where a human or an agent pastes your docs into a model, and that use may grow as site-reading AI agents become common, but it is not a shortcut around the work that earns recommendations. If a tool or agency guarantees ChatGPT citations from an llms.txt file, treat that as a red flag.

Cituna will generate one for you alongside schema and FAQ markup, then track whether your citations actually move, so you can tell a real gain from wishful thinking. Either way, spend a spare hour making your content answer-first and reachable before you spend it on a file the engines may not be reading yet.

Related questions

Do any AI systems actually read llms.txt today?

Some do, but not the ones that decide brand recommendations. Documentation platforms, developer tools and a growing set of AI agents that read sites directly make real use of it. The big answer engines that shape citations have not confirmed using it, and observed crawler traffic to these files stays low. That gap is the whole story.

If it might not work, should I bother adding one?

It is reasonable to add once your higher-impact work is done. Writing one takes minutes, it will not hurt, and it signals that you maintain your site with machines in mind. Just keep expectations calibrated: crawlability, answer-first content and off-site authority move AI visibility far more than a curated index file does today.