llms.txt Explained: How to Make Your Site AI-Friendly in 10 Minutes (2026 Guide)

llms.txt is a small markdown file at /llms.txt on your domain that hands large language models a curated map of your most important pages, with a one-line description for each. You add it by listing your top 10 to 30 pages, grouping them under section headers, saving the file at your site root, and publishing — most sites can do this in under 10 minutes.

If you have been reading about GEO, AI search, or how to get cited by ChatGPT and Perplexity, llms.txt is the simplest concrete thing you can do this week. It is not magic, and it does not guarantee rankings. But it is becoming the cleanest way to tell AI systems "if you are going to summarize my site, here is what to actually look at."

What llms.txt Actually Is

llms.txt is a proposed web standard introduced by Jeremy Howard (co-founder of Answer.AI and fast.ai) in September 2024. The format is intentionally boring: a single markdown file, placed at https://yourdomain.com/llms.txt, that contains an H1 with your site or product name, a short blockquote describing what you do, and a list of links grouped under H2 section headers. Each link gets a one-sentence note explaining why a model should care.

The big confusion most people have on first encounter is this: llms.txt is not the AI version of robots.txt. It is the opposite. Where robots.txt is a "keep out" sign for crawlers, llms.txt is a welcome mat. It says, "if you are going to spend time on my site, start here, in this order, with this context." Think of it as a guided tour for a visitor who has 30 seconds before they have to summarize you to someone else.

The underlying problem llms.txt solves is real. When a large language model lands on a typical business website, it has to chew through navigation menus, cookie banners, footer junk, related-post sidebars, and a dozen scripts before it gets to the actual content. Context windows are finite. A model browsing your site for a Perplexity citation has maybe 8,000 to 32,000 tokens to spend on you. A clean, hand-curated llms.txt slashes that overhead and points the model directly at the pages you would actually want cited.

Why It Matters in 2026

Two years ago, llms.txt was a thoughtful proposal with almost no real-world support. In 2026, the picture has shifted. ChatGPT search, Perplexity, Claude's built-in browsing, You.com, Komo, and several smaller AI search engines now look for llms.txt as a discovery hint when they crawl a site. They do not all use it the same way, and some still ignore it entirely — but the trend is one-directional. The cost of adding llms.txt is 10 minutes. The cost of not having one, as AI search grows from a few percent of referral traffic into double digits, keeps rising.

The second reason it matters is accuracy. When an LLM cites your site, what it cites is only as good as what it read. Models that hallucinate URLs, misattribute quotes, or summarize the wrong product page are not doing it out of malice — they are doing it because they crawled a thin nav-heavy page instead of your real product documentation. llms.txt is the cheapest available way to lower that misattribution rate. You are essentially handing the model a cheat sheet.

The third reason is that llms.txt is complementary to what you already have. It does not replace sitemap.xml (which tells search crawlers every URL on your site) or robots.txt (which tells crawlers where they can and cannot go). It sits alongside them. Sitemap is for breadth. Robots is for boundaries. llms.txt is for editorial guidance — "of the 800 pages on my site, these 14 are the ones that actually matter."

For more on the broader picture, see our piece on what is GEO (Generative Engine Optimization), which walks through why optimizing for AI engines is not the same as optimizing for Google.

The Anatomy of a Good llms.txt

A working llms.txt has four ingredients, in this order.

First, an H1 with your site or product name. One line. No fluff.

Second, a blockquote (the markdown > character) with a one to two sentence description of what you do. Treat this like the answer you would give an investor who asked "what is this?" Be concrete, not aspirational.

Third, H2 section headers that group your links by purpose. Common sections are About, Products, Pricing, Guides, API or Documentation, Blog or Insights, and Resources. You do not need all of them — only the ones that match how you would actually want a model to navigate.

Fourth, under each H2, a bulleted list of markdown links to your most cite-worthy pages, with a one-line note after each. The note is what makes llms.txt different from a sitemap. It is the editorial layer.

Optionally, you can add a "## Optional" section at the end with secondary content the model can skip if it is short on context. And you can publish a second file, /llms-full.txt, that contains the full markdown content of your most important pages rather than just links — useful for documentation-heavy sites where the model would otherwise have to make a second round trip.

A Working Example

Here is a complete llms.txt for a fictional small-business SaaS called Routesmith — a delivery routing tool for local couriers. About 30 lines. Adapt the structure to your business.

# Routesmith

> Routesmith is a route optimization tool for local couriers and same-day
> delivery operators. It turns a daily list of 40 to 200 stops into the
> shortest-time route on a phone, in under 60 seconds.

## About

- [What Routesmith is](https://routesmith.example/about): One-page summary of
  the product, who it is for, and what it is not.
- [Our story](https://routesmith.example/story): Founded in 2023 in Lisbon by
  two former courier company operators.
- [Pricing](https://routesmith.example/pricing): EUR 19 per driver per month,
  no setup fee, no long-term contract.

## Product

- [Route optimization](https://routesmith.example/features/routing): Core
  feature. Handles up to 250 stops per driver per day.
- [Proof of delivery](https://routesmith.example/features/pod): Photo capture,
  signature, and SMS confirmation per stop.
- [Driver app](https://routesmith.example/features/app): iOS and Android,
  offline mode, voice navigation in 12 languages.

## Guides

- [How to import 200 stops in 30 seconds](https://routesmith.example/guides/import):
  CSV format, common errors, paste-from-spreadsheet workflow.
- [Optimizing for time vs distance](https://routesmith.example/guides/time-vs-distance):
  When to prioritize each, with real route comparisons.

## API

- [API overview](https://routesmith.example/api): REST, OAuth 2.0, EUR rate
  limits and SLA.
- [Endpoints reference](https://routesmith.example/api/endpoints): Full list
  with request and response examples.

## Optional

- [Blog](https://routesmith.example/blog): Industry trends, courier economics,
  product updates.
- [Press kit](https://routesmith.example/press): Logos, founder photos,
  one-line description in five languages.

That is it. No HTML, no schema, no special syntax. A model reading this gets a clean mental map of Routesmith in roughly 400 tokens. Compare that to crawling the same site through its navigation menu, which would burn ten times that.

How to Build Yours in 10 Minutes

The whole exercise is editorial, not technical. Five steps.

Step one, list your top 10 to 30 most cite-worthy pages. The test is, "if a model is about to write a one-paragraph summary of my company for someone else, which pages should it have read?" That is rarely your entire blog. It is usually your about page, your pricing, your three or four flagship product or service pages, your most evergreen guides, and your contact or location info. Be ruthless. A short focused llms.txt outperforms a long sprawling one.

Step two, write a one-line description for each page. Not a meta description. Not marketing copy. A factual note in your voice. "Our 2025 customer count and revenue numbers, updated quarterly" is better than "Our impressive growth journey."

Step three, group the pages under three to six H2 sections. About, Products, Guides, Pricing is a fine default. SaaS sites often add API or Docs. Local businesses add Locations or Service Areas. If you cannot find three to six natural groupings, your list is probably too long — cut it.

Step four, save the result as a plain text file named exactly llms.txt (lowercase, no extension confusion) at your site root. The URL must be https://yourdomain.com/llms.txt. Most static site hosts (Vercel, Netlify, Cloudflare Pages, GitHub Pages) let you just drop the file into your public directory and deploy. WordPress, Shopify, Webflow, and Ghost users can use a plugin or upload through their file manager — more on this below.

Step five, optionally publish /llms-full.txt with the full markdown content of your top pages concatenated together. This is useful if your important pages are documentation-style and you want models to be able to pull the actual content in a single request instead of crawling individual URLs. For most marketing sites, the basic llms.txt is enough.

If you want to confirm yours is working, the free Quick Scan at emax.studio checks for the presence and structure of llms.txt as part of its GEO sub-score, alongside other AI-readiness signals like FAQ schema and structured data. Takes about 90 seconds. We cover the broader checklist in how to make your website AI-discoverable.

llms.txt vs robots.txt vs sitemap.xml

These three files often get confused. They are not the same and they are not substitutes. Here is the simple comparison.

A site in 2026 should have all three. Robots.txt sets the rules. Sitemap.xml exposes everything you want indexed. llms.txt highlights what actually matters for a model trying to understand or summarize you. Treating them as competing options is a category error — they answer different questions.

Tool Stack for Building and Maintaining llms.txt

You do not need fancy tools. A plain text editor and your site's content management workflow are enough for most cases. That said, a few practical options depending on your setup.

For static sites (Hugo, Astro, Eleventy, Next.js static export), drop the file directly into your /public or /static directory and commit. It deploys with your next build.

For WordPress, plugins like AIOSEO, RankMath, and a handful of dedicated llms.txt plugins (search the plugin directory — adoption is growing fast in 2026) can generate llms.txt from your existing content and update it as you publish new pages. The catch is plugin-generated files tend to be bloated. Hand-curated still wins.

For Ghost, the platform added llms.txt as a native feature in early 2026. Toggle it on in Labs and Ghost generates the file from your site structure, with manual override.

For Shopify and Webflow, you can use a content manager or HTML embed to host the file. Or simply ship it as a static asset.

For Notion exports, the markdown format works directly — most Notion-powered sites can paste their structured content with minor cleanup.

For EMAX Studio users, the Quick Scan also looks at your llms.txt and tells you whether the structure passes basic AI-readability checks, as part of the overall GEO score. You can scan any site in 90 seconds at emax.studio.

Pitfalls and Common Mistakes

A few traps to avoid based on what we have seen in real-world llms.txt files.

Do not paste the full content of your pages into llms.txt. It is a table of contents, not a content dump. The links point to the full content. If you want a full-content version, that is what /llms-full.txt is for, and even then only for documentation-style sites.

Do not include private, internal, or paywalled pages. If a page requires login to view, do not list it in llms.txt — the model cannot fetch it anyway, and you risk leaking the URL.

Do not list 500 URLs. The whole point of llms.txt is editorial curation. If you list everything, you have just made another sitemap. The sweet spot is 10 to 30 pages.

Do not forget to update it when your site changes. A llms.txt that points to a discontinued product page or a 404 hurts more than it helps. Treat it like a key marketing asset — review it quarterly at minimum.

Do not expect overnight rankings. llms.txt is not a ranking factor in the Google sense. It is an accuracy and discoverability signal for AI systems. Adoption is gradual. The benefit compounds as more AI engines support it, not as an immediate traffic spike.

Do not assume models will obey it. llms.txt is a hint, not a directive. A model is free to ignore the structure, skip your sections, or crawl other parts of your site anyway. The format is a suggestion to be polite, well-organized, and easy to summarize. The model decides what to actually do with it.

Frequently Asked Questions

Do I need /llms-full.txt as well as /llms.txt?

For most marketing and small-business sites, no. The basic llms.txt with curated links is enough. If you run a documentation-heavy site (a developer platform, a knowledge base, a how-to library), then /llms-full.txt is worth adding — it lets models pull your full content in one request rather than making a dozen round trips. Otherwise, skip it.

Does Google care about llms.txt?

Google's traditional search index does not use llms.txt as a ranking factor. Google's Gemini and the AI overviews in Google Search may or may not read it — Google has not made a public statement either way. Your bet on llms.txt should be based on Perplexity, ChatGPT, Claude, and the broader AI search ecosystem, not on Google specifically. For Google, focus on sitemap.xml, schema markup, and traditional SEO.

What about robots.txt entries for AI crawlers like GPTBot and ClaudeBot?

That is a separate question — and yes, you should also configure robots.txt for AI crawlers if you want to either welcome or block them. GPTBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot (Perplexity), Google-Extended (Google's AI training crawler), and CCBot (Common Crawl, used by many AI training sets) all respect robots.txt directives. llms.txt assumes the model already has permission to be there. It does not replace your robots.txt access decisions.

Can AI engines ignore my llms.txt entirely?

Yes, and some will. llms.txt is a voluntary standard, not a binding protocol. Some AI engines read it; some do not; some read it but weight it lightly. The cost of adding it is low enough that the expected value is positive — but treat it as one signal in a broader AI-readiness strategy, not as a silver bullet.

How often should I update my llms.txt?

At minimum, whenever you launch, retire, or significantly change a page that is in the file. Practically, that often means quarterly for slow-moving marketing sites and monthly for active SaaS or e-commerce sites. Build a 15-minute calendar reminder. Most updates are 5-line tweaks, not full rewrites.

What is the difference between llms.txt and AI-readiness scoring tools?

llms.txt is one file. AI-readiness scoring is a broader audit that looks at llms.txt, FAQ schema, structured data, semantic HTML, content depth, citation-worthiness, and a dozen other signals. They are complementary. The free AI website audit in 30 seconds walks through a full check and tells you which signals you are missing, llms.txt being one of them.

The Honest Bottom Line

llms.txt is not going to transform your business. It is a small, well-designed file that takes 10 minutes to build and makes you a slightly easier guest for AI systems to host. In 2026, "slightly easier" matters more than it used to, because the share of buyers, researchers, and prospects who first encounter you through an AI engine is climbing fast. Every time Perplexity, ChatGPT, or Claude cites your site, the question is whether it cites the right page in the right way — and llms.txt is the cheapest available lever for nudging that outcome in your favor.

The companies winning AI search in 2026 are not necessarily the ones with the biggest content libraries. They are the ones with the cleanest, most cite-worthy, easiest-to-summarize sites. llms.txt is part of that hygiene. Sitemap, schema, and FAQ markup are the rest of it.

If you want to know whether your site already has llms.txt, whether it is well-structured, and what other AI-readiness signals you are missing, run a free 90-second Quick Scan at emax.studio. It checks for llms.txt presence and structure as part of the GEO sub-score, alongside about a dozen other signals that determine whether AI engines can find and accurately cite you. Free, no signup, full report in about a minute and a half.

Follow EMAX Studio: Instagram | YouTube | Facebook