English Version

llms.txt is a root-level Markdown file that tells AI models like ChatGPT, Claude, and Perplexity which pages on your site to read and cite. This five-step guide covers everything from deciding whether you need one to writing descriptions that actually get referenced accurately.

Step 1: Decide Whether Your Site Needs an llms.txt

llms.txt is an open specification proposed by Answer.AI co-founder Jeremy Howard in September 2024. The concept is straightforward: place a curated Markdown file at your site's root (/llms.txt) that tells large language models which clean, high-value pages to prioritize when reading your site. It is not an access control file, and it is not a crawler discovery file. It is, at its core, a reading prompt.

According to GEO Wiki, approximately 10% of sampled domains have already deployed llms.txt, with documentation platforms beginning to generate the file automatically and adoption rates continuing to rise.

When it makes sense to create one:

• Your site has defined brand information, product descriptions, or specialist content you want AI to cite correctly
• You have more than 5 core pages — if AI only crawls your homepage, it misses critical context
• You are already working on GEO (Generative Engine Optimization) and need a complete machine-readable signal chain

When you can safely skip it for now:

• A purely personal blog with no brand consistency requirement
• A site still under construction whose core pages are not yet stable

If your positioning is clear and your content is mature, site-now.app's llms.txt guide recommends that small and mid-sized brands in 2026 treat llms.txt as a baseline AI-visibility configuration — not an optional extra.

Step 2: Understand How llms.txt Differs from robots.txt and sitemap.xml

All three files exist at the infrastructure layer of your site, but they serve entirely different readers and solve different problems. They cannot substitute for each other.

"llms.txt is the machine-readable company brief for the GEO era — analogous to robots.txt for crawlers. It tells ChatGPT / Claude / Perplexity: if you want to cite my brand, here are the sources I've authorized." — Aijentra

robots.txt and sitemap.xml have existed for years and primarily serve traditional search-engine crawling and indexing workflows. When an AI model reads pages during inference, it does not rely on sitemap.xml to discover links, nor does it strictly honor robots.txt crawl restrictions. What it needs is a manually curated, semantically clear content guide — and that is the gap llms.txt fills.

According to AI SEO Hacker's 2026 syntax guide, the entire llms.txt setup process can be compressed to 3 steps, and the ongoing maintenance cost is substantially lower than running an automated sitemap.xml pipeline.

The key distinction to internalize: robots.txt governs access permissions; llms.txt governs reading priority and content comprehension. They operate at entirely different levels.

Step 3: Build the Standard llms.txt Skeleton

llms.txt has an officially recommended file structure. The core components break down as follows.

Required elements:

1. H1 title — your site or brand name, one line, unambiguous
2. Summary paragraph or blockquote — 1–3 sentences describing what the site is, who it serves, and what value it provides
3. H2 sections with link lists — pages grouped by theme, each link accompanied by a one-sentence description

Recommended elements:

1. Optional block — supplementary information such as contact details, social media handles, or legal notices that do not affect primary comprehension

Optional elements:

1. Link to llms-full.txt — if you maintain a full-content version, point to it here

itlibra's 2026 format guide defines three tiers of information: required (brand name, primary description, core page links), recommended (product categories, FAQ links, contact details), and optional (social accounts, partnership information). The entire file should stay under 50 lines — longer files reduce AI parsing efficiency.

A baseline skeleton looks like this:

> One-sentence description: what this site is, who it's for, what problem it solves.

## Core Pages

- [Product Overview](https://example.com/product): Describes product features and use cases
- [About Us](https://example.com/about): Company background, team, and mission
- [Pricing](https://example.com/pricing): Plan comparison and audience fit

## FAQ

- [FAQ](https://example.com/faq): Answers to the 10 most common user questions

## Optional

- Contact: hello@example.com

The governing principle here: every link must carry a functional description that tells the AI what problem the page solves — not just what the page is called.

Step 4: Choose Which Pages Belong in llms.txt

llms.txt is not sitemap.xml. It should not list every page on your site. It should contain only the high-value core pages that survive deliberate editorial curation.

Inclusion criteria:

• The page can independently answer a user's complete question
• The page contains key brand facts (product names, features, pricing, contact information)
• The page content is stable and unlikely to change frequently due to promotions or seasonal updates

Page types to exclude:

• Paginated list pages (page 2, page 3, etc.)
• Functional pages such as login, registration, and shopping cart
• Low-quality or duplicate pages such as tag archives and date-based archives

"Before you chase mentions on other people's sites, make sure an AI model can actually crawl, rank, read, and attribute your own." — BrandGEO

BrandGEO's AI visibility audit guide introduces the concept of the Four Gates of Citation: can an AI crawl your page, can it rank it, can it read it, can it attribute it? Pages you include in llms.txt must pass all four gates. Pages that cannot be read or attributed dilute the file's signal quality rather than enhancing it.

A practical filtering workflow:

1. Export all site pages from sitemap.xml
2. Score each page on whether it can independently answer a complete user question (1–5 scale)
3. Keep pages scoring 3 or above — this typically yields 10–20 pages
4. Remove functional pages from that set; the final llms.txt list should contain 5–15 pages

This filtering logic aligns with the "required information" principle in itlibra's guide: fewer entries with higher information density beats an exhaustive list every time.

Step 5: Write Descriptions That Help AI Understand Quickly

The link descriptions inside llms.txt are the most overlooked component of the file — and they have the most direct impact on citation quality. Description quality determines whether AI chooses to cite a page and whether that citation is accurate.

Three elements of a good description:

1. State what problem the page solves, not what the page is named
2. Use the language your users actually ask in, not internal product terminology
3. Keep it to one sentence, targeting roughly 15–30 words

Weak vs. strong descriptions:

The relationship between llms.txt and llms-full.txt:

llms-full.txt is the extended version of llms.txt. Where llms.txt lists links and short descriptions, llms-full.txt embeds the full content of each page inline — so AI can read the complete text without a second request.

According to GEO Wiki, the two files suit different scenarios: llms.txt works well for most sites' day-to-day maintenance, while llms-full.txt suits content-dense documentation sites, knowledge bases, or situations requiring offline access. If your site has more than 20 core pages, consider maintaining both files and noting the llms-full.txt path in a single line at the bottom of llms.txt.

AI SEO Hacker's syntax guide provides a real-format example of llms-full.txt: each link entry is immediately followed by the full Markdown content of that page, with each page's content block separated by a horizontal rule.

Verification steps after deployment:

• Visit yourdomain.com/llms.txt in a browser to confirm the file is publicly accessible
• Check that the Content-Type header returns text/plain or text/markdown
• Use BrandGEO or a comparable tool to run an AI visibility audit and confirm the file is being read correctly

FAQ

Q1: Is llms.txt an official standard or a community convention?

It is a community-driven open specification proposed by Jeremy Howard in September 2024. It has not been ratified by W3C, IETF, or any other standards body. Support levels vary across AI platforms, but adoption is rising. For technical details, see GEO Wiki.

Q2: If I don't have an llms.txt, will AI stop citing my site?

Not necessarily. AI models will still cite your content through crawling or training data. What llms.txt improves is the accuracy and consistency of those citations — steering AI toward the pages you want cited, not random pages. Skipping llms.txt means missing one proactive guidance signal, not losing all AI visibility.

Q3: How often should I update llms.txt?

Update it whenever an important page goes live or comes down. For sites with relatively stable content, a quarterly review is sufficient. The file is lightweight and updating it carries almost no overhead — no dedicated automation pipeline is needed.

Q4: Does llms.txt affect robots.txt or Google's crawling behavior?

No. llms.txt is an independent file. It has no effect on robots.txt crawl rules and no effect on indexing settings in Google Search Console. The two files serve different readers (crawlers vs. AI models) and do not interfere with each other. See AI SEO Hacker's breakdown for the full comparison.

Q5: Should descriptions in llms.txt be in English or another language?

Match your target audience and the AI platforms you primarily want to reach. If your site mainly serves a non-English audience, write descriptions in that language. If you want to be cited by English-language AI tools as well, add a brief English equivalent inline. Aijentra's template includes a pattern for listing both language variants alongside brand aliases, which is worth referencing.

Q6: Does a small site (fewer than 10 pages) need llms.txt?

Yes — and the effort is minimal. Fewer pages mean a more precise file. Listing 5–8 core pages with accurate brand descriptions takes roughly 20 minutes. For a small site, llms.txt is one of the highest-return AI visibility actions available precisely because the scope is so manageable.

Q7: How do I verify that AI is actually reading my llms.txt?

There is no official verification tool yet. The practical approach: ask ChatGPT or Perplexity a question about your brand and check whether the response cites pages listed in your llms.txt. Alternatively, use BrandGEO to generate an AI visibility report and compare citation rates before and after deployment.

Sources: site-now.app llms.txt Guide · Aijentra Writing Template · AI SEO Hacker Syntax Tutorial · itlibra Format Guide · BrandGEO Four Gates of Citation · GEO Wiki

Who should use this guide?

It is for teams evaluating llms.txt Complete Guide: How to Write a Site Guide for AI who need clear steps, evidence, and risk boundaries.

What should I confirm before acting?

Confirm the target audience, public evidence, citable site pages, and the structured content that needs attention first.

How do I tell whether the work is effective?

Track brand mentions in AI answers, cited sources, indexed pages, structured-data status, and the content quality-gate results.

What common mistakes reduce reliability?

Do not present unverified claims as facts. Keep a reachable source for every material statement and re-check accessibility after publishing.

How often should I re-check?

After deployment, verify accessibility and structured data first, then track citations and indexing on a regular cadence.