llms.txt: does it actually work? (a measured 2026 verdict)
llms.txt is the most over-recommended, under-read file in AI SEO. Here's the honest version: it costs ten minutes, it won't hurt you, and you should not expect it to move citations.
Our measured verdict: llms.txt is worth adding but over-hyped. The file is a proposed convention — a Markdown index at /llms.txt that points models to your most important pages — and the idea is sound. The problem is the gap between recommendation and reality: as of 2026 the major AI crawlers behind ChatGPT, Gemini and Grok have not publicly committed to reading it, adoption across the web is low, and there is no credible evidence that publishing one increases how often you get cited. So the right posture is pragmatic, not zealous. Add an accurate llms.txt because it is cheap, harmless, and future-proofs you if support arrives — but do not treat it as an AEO lever, do not let it displace the work that actually earns citations, and be skeptical of anyone selling it as a growth tactic. It is plumbing on spec, not a results channel.
What is llms.txt supposed to do?
llms.txt is a proposed file you place at the root of your domain — https://yourdomain.com/llms.txt — written in Markdown. The idea is to hand language models a clean, curated map of your site: a short description of what you do, then a list of your most important pages with one-line summaries, sometimes with a companion llms-full.txt that inlines the full text. In spirit it is the AI-era cousin of a sitemap crossed with an executive summary: “if you only read a few pages, read these.”
It is a genuinely reasonable proposal. The trouble is not the concept — it is what is actually happening with it in the wild.
Does it actually work in 2026?
Here is the contrarian-but-honest read. There is currently no public confirmation from the companies behind ChatGPT, Gemini or Grok that their crawlers fetch and use llms.txt as part of how they retrieve or attribute sources. Adoption across real websites remains low. And crucially, nobody has demonstrated a credible cause-and-effect link between publishing an llms.txt and getting cited more. The case for it is “it might help later and cannot hurt now” — which is a fine reason to spend ten minutes, and a terrible reason to build a strategy around it.
Compare it to robots.txt and AI crawler directives, which the major operators do honour and which have a real, observable effect on whether you can be cited at all. That contrast is the whole story: one file is load-bearing, the other is aspirational.
Why is it so over-recommended then?
A few forces push llms.txt harder than the evidence justifies:
- It is concrete and checkable. “Add this file” is satisfying advice — easy to write, easy to audit, easy to put on a checklist. Hard, slow work like earning corroboration does not photograph as well.
- It feels like robots.txt. The naming invites people to assume it carries the same weight. It does not — robots.txt is a decades-old enforced standard; llms.txt is a young voluntary convention.
- It is sellable. A tidy deliverable is easy to package into an audit or a service. That commercial pull keeps it near the top of recommendation lists.
None of that makes it bad — it makes it overweighted. The skill is holding “do it” and “do not expect much” at the same time.
So should I add one?
Yes — cheaply, accurately, and without ceremony:
- Write a short, truthful file. One paragraph on what your site is, then a curated list of your best pages with one-line descriptions. Link the genuinely important ones, not everything.
- Keep it in sync. If your key pages change, update the file. A stale map is worse than no map.
- Do not gate strategy on it. Treat it as future-proofing insurance, then go spend your real hours on content and crawlability.
- Ignore upsells. There is nothing to “optimize” beyond accuracy. Skip anyone charging for llms.txt magic.
What actually moves citations instead?
The boring truth is that citations come from being a clean, corroborated, reachable source. That means self-contained answer blocks (semantic completeness & answer blocks), corroboration across the web, and making sure the right crawlers can reach you (do AI crawlers need to be allowed?). llms.txt sits a long way below all of those in priority.
How would I even know if it helped?
You would measure the only outcome that counts: which queries the models cite you on, before and after. That is what a reverse AI search gives you. Run the free Domain Check, note your query list, publish your llms.txt, and check again later. Our honest expectation is that the file alone will not move the list — and watching it is the fastest way to stop believing the hype and start trusting the data.
Frequently asked questions
Do ChatGPT, Gemini and Grok read llms.txt today?
There is no public commitment from the major model providers that their crawlers consume llms.txt as a ranking or retrieval signal as of 2026. Some smaller tools and docs platforms reference it, but the headline assistants have not confirmed they use it.
Is llms.txt the same as robots.txt?
No. robots.txt is a long-established, widely-honoured standard that controls crawler access. llms.txt is a newer, voluntary convention that suggests which pages models should prioritise. robots.txt is enforced in practice; llms.txt is advisory and largely unread.
Will adding llms.txt hurt me?
No — an accurate file is harmless. The only risk is opportunity cost: spending time on llms.txt that would be better spent on content and crawlability. Add it once, keep it honest, and move on.
Should I pay someone to optimize my llms.txt?
No. It is a short Markdown file you can write yourself in minutes. Anyone charging a premium to “optimize” it is selling certainty that does not exist yet.