How to Create an llms.txt File for AI Citations

By UpGeo · 2026-09-02

Direct answer: To create an llms.txt file for generative engine citation optimization, place a Markdown file named llms.txt at your site root. Start with an H1 site name and a one-paragraph summary in a blockquote, then list only your most citation-worthy URLs with a short description for each. That gives ChatGPT, Perplexity, Google AI Overviews, Gemini, and Copilot a clean, structured map of your content.

What is an llms.txt file?

llms.txt is a plain-text Markdown file served at https://yourdomain.com/llms.txt. It follows the emerging llms.txt standard to make web content easier for language models to read, summarize, and cite. An XML sitemap helps traditional search bots discover URLs; llms.txt works as a curated content map instead. It tells an AI which pages matter, what each page contains, and how to describe them.

The file stays intentionally small, human-readable, and machine-friendly. You can also publish an optional /llms-full.txt file with full clean text of your priority pages, but the main llms.txt acts as the citation index.

Why generative engines need a separate file

Generative engines don't need to parse megabytes of HTML, JavaScript, cookie banners, and ad scripts. They need answer-ready claims, definitions, prices, and source URLs. A compact llms.txt reduces token overhead and gives the crawler a clear signal about your most citable pages. For more on the strategic layer behind this, see what is GEO.

A marketing page can exceed 1.5 MB in HTML/CSS/JS before images. A curated llms.txt for a 200-page site can stay under 20 KB. That's roughly a 98% reduction in text the crawler must process, which makes it far more likely your entire priority map gets ingested.

Core structure of an llms.txt file

Use UTF-8 encoding and absolute URLs. A minimal structure looks like this:

# Brand or Site Name

> A two- to three-sentence summary of what the site covers, who it is for, and why it is authoritative.

## Core

- [Pricing](https://example.com/pricing): Current plans, limits, and billing FAQs.
- [API Docs](https://example.com/docs): Endpoint reference with authentication examples.

## Articles

- [2025 Industry Report](https://example.com/reports/2025): Primary data on adoption, spend, and benchmarks.

The H1 is the site name. The blockquote is the summary. The ## sections group URLs by type, and each bullet is a link plus a one-line description.

Key rules

• Use absolute URLs: https://example.com/pricing, not /pricing.
• Keep descriptions under 20 words. Make them factual, not sales copy.
• List only URLs that are useful as citations or answers.
• Place the file at exactly /llms.txt, not in a subfolder.
• Add optional /llms-full.txt if you want crawlers to access full clean text at scale.

Step-by-step: create and publish an llms.txt

1. Audit your top pages. Pull the 20–100 highest-value pages: pricing, docs, case studies, reports, FAQs, and editorial guides. Base the list on analytics, conversions, and existing search rankings.
2. Choose citation-worthy URLs. Prioritize pages with unique data, definitions, pricing specifics, or original research. Skip login pages, cart flows, thin category filters, and duplicate legal pages.
3. Write one-line descriptions. Say what a reader or model will find instead of repeating the title. For example: “Current pricing tiers, features included, and billing limits,” not “Pricing page.”
4. Group URLs under H2 sections. Simple headers like ## Core, ## Guides, ## API, and ## Research work well. Three to six sections are easier to parse than 20 flat bullets.
5. Create the file at root. Save the file as llms.txt in your domain root, UTF-8 encoded with no BOM. Check that it is served as text/plain; charset=utf-8 or text/markdown.
6. Validate the syntax. Run a Markdown linter or the llms.txt generator to catch broken links, relative URLs, or malformed bullet lists.
7. Allow AI crawlers. In robots.txt, make sure you are not blocking GPTBot, PerplexityBot, Google-Extended, ClaudeBot, or Copilot. Current bot names and user-agent patterns are listed in an AI crawlers list.

Example section mapping

Common mistakes that reduce citations

• Listing hundreds of URLs. That dilutes priority. A generative engine may treat every page as equally important, or simply skip the file. Curate aggressively.
• Using relative links. /pricing cannot be cited without a base URL. Always use fully qualified URLs.
• Forgetting to update the file. Stale prices or retired pages train models to avoid your site. Add an Updated: YYYY-MM-DD note in the summary or descriptions.
• Writing ad copy. “Best-in-class platform” is low value. “Pricing starts at $49/user/month for the Pro plan” is citation-ready.
• Blocking AI crawlers in robots.txt. The file is useless if GPTBot, PerplexityBot, Google-Extended, and others cannot fetch it.

Measuring the impact

Set up server-log monitoring for /llms.txt requests. A rise in fetches from GPTBot, PerplexityBot, Google-Extended, ClaudeBot, and Copilot often precedes increases in generative citations. Then track brand mentions in ChatGPT, Perplexity, and AI Overviews for your priority topics. Focus on answer-level metrics: Is your page being cited for queries that matter, and does the citation include your key fact? If not, sharpen the descriptions and reorder your most authoritative URLs under ## Core.

Need to generate and validate a file quickly? Use the llms.txt generator. For a deeper explanation of the strategic layer, read what is GEO.