How to add /llms.txt to your site — the LLM-friendly standard, with a real example

Last updated: May 24, 2026. Read time: 9 minutes. What you’ll learn: What /llms.txt is, the exact Jeremy Howard spec, real examples from Anthropic / Vercel / 7minai, how to add one to your own site in 10 minutes (Astro / Next.js / Hugo / any static stack), and whether the SEO claim actually holds up.

In the past 6 months, the AI builder web has quietly adopted a new convention. Visit docs.anthropic.com/llms.txt or vercel.com/llms.txt and you’ll see a structured markdown file describing the site to LLMs. The standard has a name — /llms.txt — proposed by Jeremy Howard on September 3, 2024.

The thesis is simple: robots.txt tells crawlers what they can crawl; sitemap.xml lists every URL; /llms.txt is the human-and-LLM-readable map of what’s actually worth reading. It’s optimised for inference-time consumption, not training — i.e. when ChatGPT or Claude is answering a user’s question and needs to know what’s on your site right now.

This article walks through the spec, shows you 7minai’s own /llms.txt (we added one before writing this — eat your own dogfood), and gives you a copy-pasteable implementation in 10 minutes.

1. The exact spec — what /llms.txt is and isn’t

From llmstxt.org (Jeremy Howard, September 2024):

  • It is a markdown file at the site’s root, served at /llms.txt
  • It is human-readable (so you can edit it by hand) and machine-parseable (so LLM tooling can read it with fixed processing)
  • It is not a replacement for sitemap.xml, robots.txt, or RSS — it’s an additional, curated, LLM-targeted layer
  • The spec is CC-BY / open. No vendor controls it

The required structure

# Site Title (H1, required)

> One-line description in a blockquote (optional but recommended)

Optional longer paragraph providing context

## Section name (H2, repeating)

- [Link title](https://url): Optional notes about what this page is
- [Another link](https://url): More notes

## Optional

- [Link title](https://url): Secondary/skippable content goes here

Every section is a markdown bullet list. Each bullet is a markdown link with optional : notes. The ## Optional heading is a convention for “skip this if you’re rate-limited.”

/llms.txt vs /llms-full.txt

The original spec is just /llms.txt. A second convention has emerged in practice (used by Anthropic, Mintlify, and others): /llms-full.txt is the same structure but with the full markdown content of every linked page concatenated into one file. This lets an LLM read your entire site in one fetch.

The trade-off: /llms.txt is small (~3-10KB) and fits easily into any LLM context window; /llms-full.txt can be megabytes and is meant for tools that have the context budget to ingest a whole site.

For most sites — including 7minai — only /llms.txt is necessary. We’ll add /llms-full.txt later if we see actual demand for it.

2. A real example: 7minai’s own /llms.txt

Before writing this article, we added /llms.txt to 7minai itself. You can view the live file at https://7minai.com/llms.txt. Here’s the structure — copy and adapt for your own site:

# 7minAI

> Fresh, hands-on tutorials for every new AI release. We test new AI tools and models, screenshot real outputs, and turn each one into a 7-minute guide you can actually use — never a press release rewrite.

7minAI is an English-language SEO site for AI builders. We cover new AI model releases (Gemini, Claude, GPT, Qwen, DeepSeek), AI coding tools (Cursor, Composer), and infrastructure shifts. Each tutorial is real hands-on testing — every screenshot is captured by us, every benchmark cited is from vendor docs or independently verifiable.

## Tutorials (money pages)

- [7 Minutes to Master DeepSeek V4 Pro](https://7minai.com/deepseek-v4-pro/): 75% permanent price cut math, cache-hit savings, when V4 Pro beats Claude Haiku 4.5
- [7 Minutes to Master Qwen3.7-Max](https://7minai.com/qwen-3-7-max/): Alibaba's agent-frontier model, real Qwen Chat screenshots
- [7 Minutes to Master Cursor Composer 2.5](https://7minai.com/cursor-composer-2-5/): Opus 4.7-level coding benchmarks at ~1/4 the cost
- [Gemini 3.5 Flash vs Claude Haiku 4.5](https://7minai.com/gemini-3-5-flash-vs-claude-haiku-4-5/): Agent-tier comparison using only vendor-published numbers

## News (funnel pages with builder lens)

- [DeepSeek's first $10B+ funding round](https://7minai.com/news/deepseek-10b-financing/): Why the permanent price cut + funding round are the same story
- [Microsoft cancels internal Claude Code licenses](https://7minai.com/news/microsoft-cancels-internal-claude-code/): What it does (and does not) mean for builders

Three design decisions worth noting:

  1. The blockquote opener is the LLM’s “what is this site for” answer. When ChatGPT cites you, the blockquote is what it’ll summarise. Make it a one-sentence pitch.
  2. The notes after each bullet are the LLM’s “should I cite this page” decision. Be specific. “Real Qwen Chat screenshots” tells the LLM this is a hands-on review, not a press release rewrite.
  3. Only the things worth citing belong in the list. This is not sitemap.xml. We omit the /about/ page and most listing pages because they’re not what an LLM would cite when answering a user question.

3. Who’s actually adopted it (May 2026)

This is not a theoretical spec. Major AI infrastructure companies have shipped /llms.txt:

  • Anthropic — docs.anthropic.com/llms.txt — full multilingual structure, lists 1,541 English pages
  • Vercel — vercel.com/llms.txt — covers product docs, framework integrations, deployment guides
  • Cursor — docs.cursor.com/llms.txt — IDE docs, settings, model selection
  • Mintlify — auto-generates /llms.txt for all hosted docs sites
  • fast.ai / Answer.AI — most nbdev-based projects ship with /llms.txt since the spec was proposed

The pattern: anything that sells API docs or developer tools is shipping /llms.txt in 2026. Anthropic’s adoption is the strongest signal — they would not invest in a vanity standard.

4. How to add /llms.txt to your own site (10 minutes)

The spec is so simple that “implementation” is mostly “write the file and put it at the root.” Below are concrete recipes per stack.

Astro (what 7minai uses)

# 1. Create the file in public/
touch public/llms.txt

# 2. Write content following the spec template above

# 3. Build and verify
npm run build
ls dist/llms.txt

Astro automatically serves files in public/ at the site root, so public/llms.txt becomes yoursite.com/llms.txt. No config needed.

Next.js

# Same as Astro — put it in public/llms.txt
touch public/llms.txt

Next.js serves public/ at root by default.

Hugo

# Put it in static/llms.txt
touch static/llms.txt

Hugo serves static/ at root.

Vercel / Cloudflare Pages / Netlify

Same as your framework — put llms.txt in whatever directory is served at root (public/, static/, out/).

WordPress

Upload llms.txt to your site root via FTP or a plugin like File Manager. Or use the llms-txt plugin for auto-generation.

Auto-generation tools (if you have many pages)

For sites with >50 pages, manually maintaining /llms.txt is painful. Tools:

For Astro / Next.js / static sites, no plugin needed today (the file is hand-curated, which is the right granularity for an editorial site like 7minai).

5. Does it actually help SEO? Honest answer.

This is the part most “llms.txt guide” articles get wrong. Let me state it directly:

What /llms.txt does:

  • Helps LLM-driven traffic discover you when they answer questions. When ChatGPT, Claude, or Perplexity is answering “What’s the cheapest LLM API for production?”, a tool consuming /llms.txt files has a much easier time finding 7minai’s DeepSeek V4 Pro tutorial as a cite.
  • Gives you control over how LLMs summarise your site. Without /llms.txt, an LLM scrapes your homepage and decides. With it, you supply the description.
  • Signals to LLM ecosystem builders that you’re up-to-date. Including it is “table stakes” in AI builder content.

What /llms.txt does NOT do:

  • It does not directly affect Google Search ranking. Google does not (publicly, as of writing) use /llms.txt as a ranking signal. Your sitemap.xml + robots.txt + content quality + backlinks are still what Google looks at.
  • It does not guarantee LLM tools will read it. The spec says: “this spec does not include any particular recommendation for how to process the llms.txt file, since it will depend on the application.” Consumption is up to each LLM/tool vendor. Adoption is growing but not universal.
  • It will not save a site with low-quality content. If your articles are AI slop, listing them in /llms.txt just makes the slop easier for LLMs to find and not cite.

The honest framing: /llms.txt is the AI-era equivalent of writing a good meta description. It’s a small, low-cost, high-signal thing to do. It will not flip your traffic curve overnight. It will gradually compound as LLM-driven traffic grows.

6. Practical checklist

A 10-minute /llms.txt for a content site:

  • Write a 1-sentence blockquote describing the site’s “what / for whom”
  • Write 1-3 paragraphs explaining what makes you a credible source (editorial standard, sources, hands-on testing, etc.)
  • List your money pages (the things you’d want an LLM to cite)
  • List your news / latest content (so LLMs see freshness)
  • Skip your /about/, /privacy/, listing pages (LLMs won’t cite these)
  • Save to public/llms.txt (Astro/Next) or static/llms.txt (Hugo)
  • Build + deploy
  • Verify: curl https://yoursite.com/llms.txt | head -20 should return your file
  • Optional: add to your robots.txt as # LLMs: see /llms.txt (some tools look for the cross-reference)
  • Optional: ship /llms-full.txt only if you see real downstream usage

That’s it. The whole spec is so simple that the longest part is deciding what’s worth listing.

7. Gotchas — what to expect

After shipping /llms.txt for 7minai and reviewing several large-site implementations:

  1. Don’t list every URL. The temptation is to dump your sitemap. Resist — /llms.txt is editorial curation, not site index. 20-30 entries is normal; >100 reduces the file’s signal value.
  2. Markdown headings matter. ## section headings are how LLM tools group your content. Use real section names, not generic ones like “Pages” — “Tutorials” / “News” / “Reference docs” are better signals.
  3. Use absolute URLs in links. Relative links like [X](/foo/) work for humans but break when LLM tools fetch your /llms.txt and try to follow.
  4. Update it when you publish. A stale /llms.txt listing old articles is worse than no file. Either commit to manual maintenance or use a build-time auto-generator.
  5. The blockquote is the one thing most LLMs will quote. Treat it like a press-release headline. We rewrote 7minai’s blockquote 3 times before shipping.
  6. CC-BY license on content matters. If your site requires attribution, mention the license terms in the optional details so LLMs citing you do so correctly.

8. The bigger picture

/llms.txt is part of a broader shift: the web is no longer just for browsers, it’s also for LLM agents. Sites that ship machine-readable structure today (llms.txt, schema.org JSON-LD, OpenAPI specs, AGENTS.md) will be over-represented in LLM-cited answers tomorrow. The cost is low; the upside is exposure to a growing channel.

If you’re publishing technical content in 2026 and you don’t have /llms.txt, you’re invisible to the part of the web that’s growing fastest.

Sources