The llms.txt file format: every field, what it does, and a live example

The four fields, one by one#

The llms.txt spec defines four structural elements. Each maps to a specific signal in AI retrieval.

Field 1 — The H1: entity name

The H1 at the top of the file declares the canonical entity name for the site. This is the name AI models should use when referring to your brand in generated answers.

Fact 1: AI retrieval systems that index llms.txt treat the H1 as the preferred label for the entity. If your site operates under multiple names — legal entity, brand name, product name — the H1 is where you pick the one you want AI models to use.

Fact 2: The H1 should match the name field in your Organization schema JSON-LD. Consistency between the two signals reinforces entity identity: the AI model sees the same name in two independent machine-readable sources, raising its confidence in the entity mapping.

What to put in H1: the name as a customer would say it. Not "BrandCited.ai" or "BrandCited, Inc." — just "BrandCited."

Field 2 — The blockquote: brand description

The blockquote, placed directly after the H1, describes what the site does in one to two sentences.

Fact 1: The blockquote is the primary description field AI models use to classify your brand into a topic category. A vague blockquote ("We help businesses grow online") produces a vague classification. A specific one ("GEO SaaS that monitors ChatGPT and Perplexity for brand citations") places your brand in the right category cluster.

Fact 2: The blockquote is optional per the spec but treated as required by most AI systems that ingest llms.txt. A file with an H1 and no blockquote gives retrieval systems a name but no classification signal — the equivalent of a business card with only a name and no job title.

Keep the blockquote under 50 words. Name the engines, categories, or use cases your product addresses directly — AI models use these terms to match your brand to relevant queries.

Field 3 — H2 sections: content categories

H2 headings divide your llms.txt into named content sections. Each H2 label tells the AI what type of content follows.

Fact 1: The H2 label functions as a topic cluster signal. AI models that retrieve content for queries like "BrandCited documentation" will prioritize pages listed under a "Docs" H2 over pages listed under a general "Pages" or "Links" H2.

Fact 2: There is no canonical list of valid H2 labels in the llms.txt spec. Common categories seen in production files include: Docs, API, Blog, Guides, Pricing, About, Case Studies, Integrations, and Optional. The ## Optional H2 is a special reserved label described in Field 4 below.

Each link within an H2 section follows this format:

The colon-plus-description after the link URL is optional in the spec, but recommended. AI models use it to select which page to fetch when answering a specific question. A description like "How BrandCited calculates the 0-100 AI visibility score" pulls the scoring page to the top of retrieval for scoring-related queries. A link with no description gets treated as a generic page reference.

Field 4 — The Optional H2: lower-priority content

The ## Optional heading is a reserved signal in the llms.txt spec. Pages listed under it are accessible to AI models but marked as lower priority for retrieval.

Fact 1: Pages under ## Optional tell AI retrieval systems: fetch these if needed, but do not prioritize them when answering topic questions. Legal pages, changelog archives, and full sitemap links belong here.

Fact 2: The Optional heading is case-sensitive per the spec. ## Optional signals the reserved behavior. ## optional or ## OPTIONAL may not be parsed as the reserved label by all AI model implementations that check for it.

How llms-full.txt works alongside llms.txt#

The companion file llms-full.txt, placed at https://yourdomain.com/llms-full.txt, contains the full text content of each page listed in your llms.txt, concatenated into one document.

Fact 1: AI models using a single-fetch retrieval pattern can download llms-full.txt to access all of a site's curated content without making individual page requests. For sites with deep content archives, this reduces the number of HTTP requests a model makes from dozens to one.

Fact 2: The format for llms-full.txt is the page URL as an H1, followed by the full page markdown content, then a horizontal rule separator, repeated for each page. No proprietary format — any static site generator or CMS export script can produce it automatically.

Most sites ship llms.txt first and add llms-full.txt later, once tooling exists to generate it automatically on each deploy. Perplexity and Claude are the two AI engines most likely to ingest llms-full.txt when indexing a domain — both use retrieval patterns that benefit from a pre-concatenated content source.

A complete llms.txt template, ready to copy#

The following template covers a SaaS product with documentation, a blog, and standard legal pages. Replace placeholder content with your brand's specifics.

Save this file as llms.txt, place it at your domain root, and verify it is accessible at https://yourdomain.com/llms.txt with a 200 status code. Some CDN configurations redirect the root path in ways that break llms.txt delivery to AI crawlers — fetch it directly from your browser's address bar to confirm.

AI search news: July 2–5, 2026#

• EntityMap standard launched: Fred Laurent, CTO of InLinks, launched EntityMap on July 1 — a JSON file at your domain root that tells AI crawlers your organization's knowledge domains and entity relationships, extending what llms.txt does into structured entity graph territory. (InLinks)
• Google Search Agents live: Google launched 24/7 AI monitoring agents for AI Ultra subscribers on July 2. Agents re-query topics continuously, making citation consistency across repeat queries the new standard for AI visibility. (Google Blog)
• Ahrefs brand-mention study: Ahrefs analyzed 75,000 brands and found brand mentions predict AI search citations 3x more accurately than backlinks, with a correlation coefficient of r=0.664 for mentions versus r=0.218 for links. (Ahrefs via Design Rush)
• Claude Fable 5 globally redeployed: Anthropic redeployed Claude Fable 5 worldwide on July 1 after US export controls lifted, expanding availability to Amazon Bedrock and Google Cloud. (MarkTechPost)
• Perplexity query volume: Perplexity processes an estimated 1.2 billion monthly queries as of July 2026, up 54% from 780 million in May 2025. (Panto.ai)

How BrandCited checks your llms.txt#

BrandCited's audit engine checks for llms.txt presence, structure, and completeness as part of its technical AI visibility checks. The audit flags missing H1 declarations, absent blockquotes, H2 sections with fewer than two linked pages, and mismatches between the llms.txt H1 and the Organization schema name field. Each finding includes a specific remediation step ranked by impact on AI citation probability.

Run a free BrandCited scan at brandcited.ai to check your llms.txt against all 30+ audit criteria.

What to do right now#

1. 1Create your llms.txt today if you don't have one. Place it at https://yourdomain.com/llms.txt using the template above. This is 20 minutes of work.
2. 2Write a specific blockquote, not a generic one. Name your product category, the integrations you connect to, or the specific problem you solve. Vague blockquotes produce vague AI classifications.
3. 3Match your H1 to your Organization schema `name` field. Open your homepage source, find your JSON-LD, and confirm the name field matches the H1 in your llms.txt exactly.
4. 4Add descriptions to every link. The colon-plus-description after each link URL is optional per the spec but used by AI models to select pages for retrieval. A description makes your pages more retrievable for specific queries.
5. 5Move legal and admin pages under `## Optional`. Anything that is not a product, doc, blog, or about page belongs in Optional. Keeping your primary sections lean improves retrieval signal-to-noise.
6. 6Verify the file returns a 200. Fetch https://yourdomain.com/llms.txt directly from your browser. A redirect chain, a 404, or a 403 means AI crawlers cannot read your file — and BrandCited will flag it as a failed check.

Frequently asked questions#

robots.txt controls which pages AI crawlers and search engine bots can access. llms.txt provides structured context about what those pages contain and how they are organized. The two files serve different functions: robots.txt is access control, llms.txt is content classification guidance. A well-configured site has both.

No. AI models index and cite content from sites without llms.txt. The file improves citation probability by giving retrieval systems a faster, more accurate way to classify your brand and identify your most important content. Sites without llms.txt rely on AI models inferring structure from HTML alone — a slower and less accurate process than a direct instruction file.

The ## Optional heading is a reserved label in the llms.txt spec that marks lower-priority pages. Pages listed under Optional are accessible to AI models but are not prioritized for retrieval when answering topic questions. Legal pages, full sitemaps, and changelog archives belong here.

Each AI engine handles llms.txt differently. Perplexity fetches llms.txt as part of its real-time web retrieval loop and uses it to scope which pages to fetch for a query. Claude's retrieval layer uses the file as a topic classification signal when present. ChatGPT's browse-enabled mode can fetch and parse llms.txt during research queries. None of the three guarantee citation based on llms.txt alone — content quality and third-party citation footprint remain primary signals.

Update it whenever you add a significant new section — a new product, a new docs area, a new blog category. For sites with stable content structures, once per quarter is sufficient. Add a comment at the top of your file to track freshness if you want AI models to treat it as a maintained signal: <!-- Last-modified: 2026-07-06 -->.

llms-full.txt is a companion file at yourdomain.com/llms-full.txt that contains the full markdown content of each page listed in your llms.txt. AI models that use single-fetch retrieval can download llms-full.txt to read all curated content in one request. It is optional. Most sites ship llms.txt first and add llms-full.txt once they have tooling to generate it automatically on each deploy.

Sources#

1. 1Howard, J. (2024). llms.txt proposal and specification. answer.ai / llmstxt.org.
2. 2InLinks (2026). EntityMap standard: AI entity visibility file. InLinks.
3. 3Google (2026). Google Search's I/O 2026 updates: AI agents and more. Google Blog.
4. 4Design Rush (2026). Ahrefs Study: Brand mentions beat backlinks for AI citations. Design Rush.
5. 5Panto.ai (2026). Perplexity AI Statistics 2026. Panto.ai.
6. 6MarkTechPost (2026). Anthropic redeploys Claude Fable 5 globally after export controls lift. MarkTechPost.
7. 7Presenc.ai (2026). AI search engine market share and conversion data 2026. Presenc.ai.