Roughly 60% of the AI-search visibility questions Google publishers asked in 2026 now touch a file almost nobody had heard of two years ago: llms.txt. The proposal is simple — hand large language models a clean, prioritized map of your best content instead of making them crawl and guess. The hype around it is loud, the adoption evidence is thin, and the implementation cost is close to zero. That combination is exactly why it deserves an honest, practical look rather than a hot take.
This guide explains what llms.txt is, what we actually know about whether AI engines read it, what belongs in llms.txt versus llms-full.txt, and how to ship and sync both on a SaaS or marketing site. To see how AI-ready your own content is today, run our free 5-axis content audit.
Updated 2026-05-22 with the current state of llms.txt adoption, engine statements and tooling observed across US, UK and European sites.
- llms.txt is a Markdown index of your best content, served at the domain root — not a crawler directive. 2. No major engine has publicly confirmed it as a ranking input in 2026, so treat adoption claims with caution. 3. llms.txt is the index, llms-full.txt is the full export — ship both, generated from live content. 4. The cost is near zero and the GEO upside is real if uncertain, which makes it a reasonable bet. 5. A stale file is the only real risk — automate the rebuild in your deploy step.
What is llms.txt and how does it differ from robots.txt?
The file llms.txt is a plain Markdown document you place at the root of your domain, at /llms.txt. It gives large language models a curated, prioritized map of the content you most want them to read. The format proposed at llmstxt.org in 2024 is deliberately minimal: an H1 with your brand name, a one-line blockquote summary, then H2 sections of Markdown links, each with a short description.
llms.txt — a recommendation layer. It does not block anything and it does not change your pages. It simply says, in a format a model can parse in milliseconds, "here is my best content and what each piece covers."
robots.txt — an access-control layer. It tells crawlers which paths they may fetch and which they must skip. It is a long-standing standard honored by search and AI bots alike, and it has real teeth.
The two files do opposite jobs, which is why they are complementary rather than competing. robots.txt gates the crawl; llms.txt curates the read. You can ship both, and shipping one says nothing about the other. If you are new to the AI-search landscape, our complete GEO guide sets the wider context.
Do ChatGPT, Perplexity and Google actually read it?
This is the question that matters, and the honest answer is: the evidence is thin and mixed. As of 2026, no major engine has publicly confirmed that it uses llms.txt as a ranking or retrieval input.
Google — A Google representative stated in 2025 that the company was not using llms.txt, pointing instead to its established crawling and indexing systems. Treat any claim of a direct Google or Gemini ranking benefit as unproven.
OpenAI and Perplexity — Neither has published a clear statement confirming that ChatGPT or Perplexity reads llms.txt at fetch time. Their crawlers are documented, but documented crawling of your pages is not the same as honoring an llms.txt index.
So why ship it at all? Because the cost-benefit math is lopsided. Publishing the file takes a few hours once and can be automated forever after. Even a small probability that one engine — now or in a future version — uses it makes the expected value positive when the cost is near zero. What you should not do is expect llms.txt to lift thin content or substitute for the fundamentals covered in our guide to being cited by AI engines.
What belongs in llms.txt vs llms-full.txt?
The proposal defines two distinct files with two distinct jobs, and conflating them is the most common mistake.
llms.txt — the index. It is small, usually a few kilobytes. It contains an H1, a blockquote summary, and curated links grouped under H2 sections with one-line descriptions. Think of it as a hand-picked table of contents for a model, listing maybe 10 to 30 of your highest-value URLs.
llms-full.txt — the full export. It concatenates the actual Markdown body of every key page into one file, so a model can ingest your real content in a single fetch instead of following each link. This file can reach hundreds of kilobytes, which is exactly why you generate it rather than write it by hand.
A practical split: put product, pricing, core docs and your best cornerstone guides in the index; put the clean Markdown of those same pages in the full export. Strip navigation, cookie banners and boilerplate from the full file so a model reads content, not chrome. For SaaS vendors specifically, our AEO guide for SaaS vendors covers which pages earn citations.
How to generate it for a SaaS or marketing site
You do not need a plugin. The whole thing is a text file, and the work is curation, not code.
Inventory — List the 10 to 30 pages an engine should read first: product, pricing, core documentation, and your strongest cornerstone articles. Skip thin, duplicate or low-intent URLs.
Write the index — Start with an H1 of your brand, then a one-line blockquote of what you do. Group the shortlist under H2 sections such as Docs, Product and Guides, each link followed by a short, factual description that tells a model exactly what the page covers.
Build the full export — Pull the clean Markdown body of each shortlisted page and concatenate it into llms-full.txt with clear separators between documents. This is where automation pays off, because hand-maintaining hundreds of kilobytes is hopeless.
Serve at the root — Publish both at the root path with a 200 status and a plain-text content type. On most frameworks you expose them from the public directory or a tiny route handler. Tag inbound clicks from any AI surface with our UTM builder so you can measure referral traffic later.
How to keep it in sync with your content
A map is only useful if it matches the territory. The single biggest failure mode for llms.txt is staleness — a file that points models at a retired feature, a renamed product, or last quarter's pricing.
Generate, do not hand-edit — Build both files from your live content source, whether that is your CMS, your docs repo, or your sitemap. A hand-typed file drifts the moment anything changes.
Rebuild on deploy — Wire generation into your build or deploy pipeline so the files regenerate on every publish. This removes the staleness risk entirely: the files simply cannot fall behind the site.
Review what you expose — Because llms-full.txt contains real page content, check that nothing private, gated or unfinished slips into the export. The discipline is the same one you already apply to a sitemap.
Watch your tables and prices — Numeric content like pricing tables and feature matrices ages fastest. If a page changes weekly, make sure the generator picks up the new version, not a cached copy.
The llms.txt implementation table
Work this table top to bottom — it pairs each implementation decision with a recommendation and the reason behind it.
No major AI engine has publicly confirmed llms.txt as a ranking input as of 2026. Publishing it will not rescue thin content, fix weak fundamentals, or guarantee a single citation. The realistic case for shipping it is its near-zero cost and small possible upside — not a promised result. Build the file, automate it, and keep investing in the content quality that engines actually reward.
How llms.txt fits a broader GEO strategy
llms.txt is a layer, not a strategy. Generative Engine Optimization is the work of becoming the source an AI engine quotes, and that rests on content quality, structured data, citations and clear answers — not on a single text file.
Foundation first — Strong, well-structured, citable content is what gets quoted. Schema, FAQ markup and direct answers do far more heavy lifting than llms.txt. See how to rank in AI Overviews for the mechanics.
llms.txt as a cheap accelerant — Once the foundation is solid, the index and full export are a low-effort layer that may help engines find and read your best material faster. The keyword is may — ship it for the cost, not the promise.
Measure the referral, not the file — You cannot easily prove an engine read your llms.txt. What you can measure is downstream AI referral traffic, which is why UTM-tagging inbound AI clicks matters.
To pull all of this together — content quality, structure and AI-readiness — run the SteerAds free 5-axis audit, and tag your AI referral traffic with the UTM builder so you can see what actually moves.
Sources
Official and primary sources consulted for this guide:
-
llmstxt.org — the llms.txt proposal
-
developers.google.com — robots.txt introduction
-
platform.openai.com — OpenAI bots and crawlers
-
blog.google — Google Search announcements
FAQ
What is llms.txt?
llms.txt is a plain-text Markdown file placed at the root of your domain — at /llms.txt — that gives large language models a curated, prioritized map of your most useful content. It typically opens with an H1 of your brand, a short blockquote summary, then sections of links with one-line descriptions. The idea, proposed at llmstxt.org in 2024, is that an AI engine fetching your site can read a clean index instead of crawling and guessing. It is a suggestion to models, not an enforced directive, and as of 2026 no major engine has publicly confirmed it as a ranking input.
Does Google use llms.txt?
There is no official confirmation that Google Search or Gemini reads llms.txt as of 2026. Google has publicly said it relies on its established crawling and indexing systems, and a Google representative noted in 2025 that the company was not using llms.txt. Treat any claim of a direct Google ranking benefit as unproven. The honest position is that adoption evidence is thin across all engines, but the file costs almost nothing to publish, so the downside is negligible even if the upside is uncertain.
How is llms.txt different from robots.txt?
robots.txt tells crawlers which paths they may or may not fetch — it is an access-control file read by search and AI bots alike. llms.txt does the opposite job: it does not block anything, it recommends your best content to a model that is already reading. robots.txt is a long-standing, widely honored standard; llms.txt is a 2024 proposal with uncertain adoption. They are complementary, not substitutes — you can and should ship both, and neither one overrides the other.
What is the difference between llms.txt and llms-full.txt?
llms.txt is the index: an H1, a summary, and curated links with short descriptions — usually a few kilobytes. llms-full.txt is the full export: the actual Markdown body of every key page concatenated into one file, so a model can ingest your real content in a single fetch without crawling each URL. Use llms.txt as the lightweight map and llms-full.txt when you want models to read complete docs. The full file can reach hundreds of kilobytes, so keep it generated, not hand-written.
Is llms.txt worth implementing for a SaaS site?
For most SaaS and marketing sites, yes — on a cost-benefit basis. Generating both files from your existing content takes a few hours once and can be automated to stay in sync. The marginal cost is near zero, and even a small chance that ChatGPT, Perplexity or a future Gemini surface reads it makes it a reasonable bet. What it is not is a shortcut: llms.txt will not rescue thin content or replace solid GEO fundamentals. Treat it as one low-cost layer in a broader strategy, not a silver bullet.
Where does the llms.txt file go?
Place it at the root of your domain so the canonical URL is your-domain.com/llms.txt, exactly like robots.txt sits at the root. The optional full export goes at your-domain.com/llms-full.txt. Serve both as plain text with a 200 status and a text/plain or text/markdown content type. On most frameworks you expose them from the public directory or a small route handler. Do not nest them under a subfolder — engines that look for the file expect it at the root path.
Can llms.txt hurt my SEO?
No major engine has indicated that publishing llms.txt carries an SEO penalty, and because the file does not block crawling or alter your pages, there is no plausible mechanism for direct harm. The realistic risks are mundane: a stale file that points models at outdated prices or retired features, or a full export that accidentally exposes content you did not intend to surface. Both are avoided by generating the files from your live content and reviewing what you include — the same discipline you already apply to a sitemap.
How often should I update llms.txt?
Regenerate it whenever the content it references changes materially — a new pricing page, a renamed product, a deprecated feature, or a batch of new articles. The cleanest approach is to build the file in your deploy pipeline so it is rebuilt on every publish and can never drift from the live site. If you maintain it manually, a monthly review is a sensible floor, but automation removes the staleness risk entirely and costs less over time.