TL;DR
llms-full.txt is the complete-text companion to llms.txt. Where llms.txt is a curated index with links and one-sentence descriptions, llms-full.txt is the whole library — every page you want AI agents to know about, concatenated into one Markdown file.
Most small business sites do not need it. If your site has fewer than 15-20 substantial content pages, a well-written llms.txt covers you. llms-full.txt is worth the effort when you have genuinely dense reference content: a deep guide library, API documentation, legal explainers, or a large blog corpus where an AI agent answering detailed questions needs to read the full text, not just a summary.
What Is llms-full.txt?
The llms.txt spec defines two files:
| File | What it contains | Typical size |
|---|---|---|
| llms.txt | Navigation index — site summary + links to key pages | Under 10KB |
| llms-full.txt | Complete text of all linked pages, concatenated | 100KB to several MB |
llms.txt answers the question “what pages does this site have and what do they cover?” llms-full.txt answers “give me everything.” For an AI agent with a large enough context window and a specific enough question, full-text access removes a round-trip: instead of fetching llms.txt, then deciding which page to read, then fetching that page, the agent can search the entire content corpus at once.
The practical difference shows up when a user asks something like “what’s the step-by-step process for X?” against a site with 40 detailed how-to posts. An agent working from llms.txt has to pick a page and fetch it. An agent working from llms-full.txt can scan all 40 posts simultaneously and synthesize the most accurate answer.
Who Actually Needs llms-full.txt
Be honest about your site before building it:
Likely worth it:
- Software documentation sites (Mintlify, Fern, ReadMe users) — these platforms generate it automatically
- API reference libraries with dozens of endpoint descriptions
- Comprehensive legal, financial, or technical guide sites (20+ substantive long-form posts)
- Knowledge bases where users frequently ask detailed, multi-step questions
Probably not worth it:
- A local service business with 5-10 service pages and a thin blog
- An e-commerce store where product descriptions are the main content
- Any site where most pages are thin (under 500 words) or largely duplicate each other
- Sites updated so frequently that maintaining a manual llms-full.txt would lag behind
The SEOPulse audit checks for llms.txt presence as part of its GEO readiness score — llms-full.txt is a follow-on step, not the starting point.
The Format
llms-full.txt is plain Markdown. The structure is straightforward: a section header for each page, followed by the full page content, separated by horizontal rules.
# Site Name — llms-full.txt
> Brief site description (same as your llms.txt summary)
---
# How to Do X: Step-by-Step Guide
Source: https://yourdomain.com/blog/how-to-do-x
## Introduction
[Full page content in clean Markdown — headings, paragraphs, lists, tables]
---
# The Complete Guide to Y
Source: https://yourdomain.com/guides/complete-guide-y
## What Is Y?
[Full page content continues]
---Critical formatting rules:
- Remove all HTML tags — clean Markdown only
- Strip navigation, headers, footers, sidebars, cookie notices
- Keep the body content intact: headings, paragraphs, lists, code blocks, tables
- Include the source URL for each page (this lets AI engines cite back correctly)
- Use
---to separate pages so agents can parse boundaries
Sizing Guidance
File size matters because it must fit within an AI agent’s context window to be useful end-to-end.
| Site type | Typical llms-full.txt size | Notes |
|---|---|---|
| Small blog (15-30 posts) | 150-300KB | Manageable for most agents |
| Medium content site (50-100 pages) | 400KB-1MB | Consider curating to top 40 pages |
| Large documentation site (500+ pages) | 5MB+ | Auto-generation required; most agents will truncate |
A reasonable rule of thumb: under 500KB for hand-curated files. Beyond that, most AI agents will read only the beginning of the file, which means your most important content should come first — not be buried at page 80.
For reference, some major adopters have published much larger files (Cloudflare’s developer docs runs into millions of tokens), but those are auto-generated by documentation platforms and intended for specialized ingestion tools — not a pattern to emulate manually.
What to Include, What to Skip
Include:
- Your best how-to guides and step-by-step posts
- Comparison pages with specific, factual data (product specs, side-by-side tables)
- FAQs with detailed, accurate answers
- Any page where specific numbers, dates, or procedures matter
- Your most-cited reference pages (check Google Search Console for top organic landing pages)
Skip:
- Thin pages under 400 words
- Pages with duplicate or near-duplicate content
- Legal boilerplate (privacy policy, terms) unless your site is specifically a legal resource
- Paginated archive pages and tag/category pages
- Checkout flows, account pages, and any transactional content
The test for inclusion: “Would an AI agent looking at this page be able to extract a specific, verifiable, useful fact?” If the answer is no, leave it out.
Generating It Automatically
For simple sites, hand-building llms-full.txt from your source Markdown files is practical — a shell script that loops over your posts/ directory and concatenates them takes about 10 lines.
For larger sites:
- Mintlify and Fern generate both llms.txt and llms-full.txt automatically during deployment — no manual work if you’re already on those platforms
- Astro and Next.js can generate it via a custom static route that reads your content collection and concatenates pages at build time
- WordPress users can write a simple PHP function that exports post content or use a plugin that outputs Markdown
If you’re not on a docs platform, a build-time script is the right approach. Manual maintenance of a large llms-full.txt will drift out of sync with your actual content within weeks.
How It Fits Alongside llms.txt
The two files serve different agent behaviors:
- An agent doing a quick site orientation reads llms.txt to understand what you cover, then fetches the specific pages it needs
- An agent doing deep research or answering a complex multi-step question reads llms-full.txt to get everything at once
Having both is the best-practice setup. Start with llms.txt, verify it’s accurate and well-organized, then add llms-full.txt once your content base is substantial enough to warrant it.
Neither file replaces the core GEO signals that actually drive AI citation: a direct answer in the first paragraph, question-based headings, specific verifiable facts, and consistent claims across your own pages. llms-full.txt surfaces your content to agents — it’s the quality of that content that determines whether they cite it.
Run a free GEO readiness audit to see how your site currently scores across schema markup, answer-first structure, FAQ coverage, and llms.txt presence.