Guide
Published · By Shoaib Mustaque· 5-minute read
llms.txt is a plain Markdown file served at /llms.txt that tells an AI system what a site is and which pages matter, in a format proposed by Jeremy Howard on 3 September 2024. It is optional, no major engine has committed to reading it, and it takes about an hour to write. Citemark's own file is the worked example below.
The spec
The llms.txt specification asks for a Markdown file with sections in a fixed order: an H1 with the site's name, which is the only required part; a blockquote summary; detail paragraphs; and H2 sections containing lists of links, each a Markdown link with an optional note after a colon. A section named “Optional” marks links an agent can skip.
# Citemark. The only required line.
One paragraph starting with >. What the site is, for whom, where. Written so it can be quoted whole.
Plain Markdown, no headings. Facts an agent should know before following links: what you do not do, how you price, your timeline.
## Services, ## Company, ## Learn. Each item is name: a one-line note. Absolute URLs.
An H2 literally named Optional, for secondary links an agent may drop when context is short.
The current version of the spec, revised 10 August 2026, also allows a file at a subpath covering the URLs beneath it, and suggests serving a Markdown copy of each page at the same URL with .md appended. We do the first and not yet the second.
Candidly
As of 17 October 2026, none of the crawler documentation published by OpenAI, Google, Anthropic or Perplexity mentions llms.txt, and Google's guidance on AI features says there are no additional requirements and no special structured data. So the honest position is: write it because it is cheap and forces clarity, not because an engine has promised to read it.
Worked example
Our file has five parts: the name, a blockquote any engine can quote as a definition of the company, a paragraph stating what we do not guarantee and how we price, and link sections for services, company pages and these guides, with a facts list at the end. Every line is also true on the page it links to.
# Citemark
> Citemark is a Generative Engine Optimization (GEO) and Answer Engine Optimization (AEO) agency in Bengaluru, India. It gets brands cited by ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, and ranked on Google. The team has run SEO since 2013 and is made of engineers who fix rendering, schema and indexing in the client's codebase. Tagline: "Engineered to be cited."
Citemark does not guarantee rankings or citations. It publishes its timeline (first AI-engine signals in 2–4 months; competitive prompts in 6–12 months) and reports monthly with screenshots of each citation. Plans are month to month and include SEO, AEO and GEO together; the audit is free.
## Services
- [Generative Engine Optimization](https://citemark.in/services/generative-engine-optimization): entity engineering, retrievability (llms.txt, Bing indexing, crawler access), citation-worthy content, mentions and digital PR, monthly citation tracking.
- [Free AI visibility audit](https://citemark.in/ai-visibility-audit): what each engine says about a brand today, screenshot by screenshot, and the three fixes first. Delivered in five working days.
## Learn
- [What is Generative Engine Optimization (GEO)?](https://citemark.in/learn/what-is-generative-engine-optimization): definition, the 2023 research paper the term comes from, how engines build answers, the six workstreams, timelines.
- [llms.txt: what it is and how to write one](https://citemark.in/learn/llms-txt-guide): the specification, who reads it today, this file annotated, six steps, mistakes.
## Facts
- Based in Bengaluru, Karnataka, India. Serves India and international clients.
- Engines covered: ChatGPT (including search), Google AI Overviews, Gemini, Perplexity, Claude, Microsoft Copilot.
- Email: hello@citemarkhq.comHow to
Writing an llms.txt takes six steps: agree one paragraph that describes the company, list the ten to twenty pages that matter, write a one-line note for each, add the facts a buyer checks, serve the file at the root as plain text, and add a reminder to update it whenever a page changes. The hard part is the first step.
Category, city, what you deliver, since when. If the founder and the sales lead describe the company differently, settle it here; the engines will otherwise pick one at random.
Services, pricing, about, contact, the three or four guides you would hand a prospect. Not every blog post.
After the colon: what the page says, in the page's own numbers. “Three monthly plans and a sprint; prices exclude GST” beats “Our flexible pricing”.
City and country, engines or products covered, industries, a contact email. Nothing you would not put in your schema.
Static file at /llms.txt, served as text/plain or text/markdown, UTF-8, no redirect. In Next.js, a file in the public folder does it.
Add a line to your release checklist: new page, new route entry, new llms.txt line, same commit.
Mistakes
The common failures are a marketing paragraph instead of a description, placeholders left in from a template, a file that lists every URL on the site, notes that repeat the page title, and a file written once and never updated so it contradicts the pages it links to. Each makes the file less useful than having none.
Engineering
The file is useful while it matches the pages. We keep ours in sync: every public route is registered in one list that generates the sitemap, and the release checklist adds a matching llms.txt line in the same commit. A quarterly check reads each note against its page. If they disagree, the page wins and the file is fixed.
If you want the file written for you, it is part of the entity setup in every Citemark plan, and the free AI visibility audit records what each engine currently says about you, which is the description the file should match.
Sources
Every factual claim above about how an engine works, or about what a company publishes, links to the source it came from. Each URL was opened and checked on the date shown at the top of this guide. If a source later changes or disappears, the claim should be re-checked, not assumed.
Keep reading
The free audit runs your category's prompts on ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews and shows who each engine cites today. Five working days, screenshots included.