--- title: "llms.txt: make your website readable for AI" description: "What llms.txt is, how it's structured and why it makes Claude, Perplexity, Cursor and Gemini more accurate about your site. With a real example for Devlights. Tech Tips, episode 5." url: "https://www.devlights.com/blog/llms-txt-make-your-website-ai-readable" language: en published: 2026-10-24T13:00:00.000Z updated: 2026-10-28T22:29:57.024Z type: article authors: ["Saul Belbey"] category: "Tech Tips" tags: ["Llms.txt", "AI", "SEO for AI", "Claude", "Perplexity", "Cursor", "Gemini", "Web development", "Tech Tips", "Web standards"] --- # llms.txt: make your website readable for AI > What llms.txt is, how it's structured and why it makes Claude, Perplexity, Cursor and Gemini more accurate when they look up information about your site. With a real example for Devlights. This article is part of the Devlights Tech Tips series on AI applied to technical work. If you missed the previous episodes:[ model, agent and tool](https://www.devlights.com/blog/model-agent-tools-ai-development) ·[ tokens and context window](https://www.devlights.com/blog/tokens-context-window-memory-ai) ·[ AGENTS.md](https://www.devlights.com/blog/agents-md-coding-agent-instructions) ·[ Skills](https://www.devlights.com/blog/ai-agent-skills-installation). 📺 [**Watch the video on Instagram**](https://www.instagram.com/reel/DadtZWuJfmr/) When an AI model tries to understand a web page, what it sees is messy HTML, scripts, ads, navigation, footers. It has to filter out all that noise before it gets to what matters. **llms.txt solves that.** ## What is llms.txt? llms.txt is a plain text file that lives at the root of your website. It holds a structured summary of your content: who you are, what you do, your main pages, your documentation, your products or services. It's the new standard of digital courtesy. If you want Claude, Perplexity, Cursor or any other model to understand your site efficiently, you hand it the map directly instead of making it crawl through all the HTML. ``` Without llms.txt → the model crawls the HTML, filters noise, infers structure With llms.txt → the model reads the summary directly and answers accurately ``` ## The problem without llms.txt To understand the value of the file, it helps to first look at what happens without it. If you ask Claude about your site and only give it the URL, the model has to crawl the HTML, interpret the page structure and separate the actual content from the navigation, scripts, banners and footer. You get an answer, but it tends to be generic: it lacks the specific details of the business, the exact services, the information that lives on internal pages. The result: more tokens consumed, more processing time and less accuracy. ## How an llms.txt is structured The structure is simple and doesn't need to be long. It needs to be precise. ``` # Devlights Software development company specializing in data, AI and cloud. We work with teams that need custom technical solutions. ## Main pages - [Services](/servicios): software development, data and AI, cloud and DevOps - [Team](/equipo): who we are and how we work - [Case studies](/casos): projects delivered with clients - [Blog](/blog): technical articles by the team - [Contact](/contacto): how to work with us ## About Devlights Devlights is an Argentine software company... [more detailed description of the business, stack, methodology] ``` Four key elements you can't leave out: **Site name and description.** One or two lines that define what the project or company is about: no empty marketing, just concrete information. **Main pages with descriptions.** The most important links on the site, each with a line explaining what's inside. The model will prioritize these when it needs to dig deeper. **Documentation or endpoints.** If you have an API, technical documentation or specific resources, listing them here lets tools like Cursor or Claude Code find them directly. **llms-full.txt (optional).** An extended version with the full content for the models that need it. The standard llms.txt is the summary; the full version is the detail. ## The same question, with and without the file A side-by-side comparison is the most direct way to see the impact. **Without llms.txt:** the model crawls the HTML, builds its own summary from whatever it can infer and answers with generic or incomplete information about the site. **With llms.txt available:** the model goes straight to the file, reads the structured content and answers accurately, with the exact services, the right description and the relevant links. No crawling the HTML. No noise. ``` Same question Fewer tokens consumed More accurate, specific answer ``` The difference isn't marginal. Any tool that hits your URL will find the file before it tries to parse the HTML. ## Who reads llms.txt and why it matters now The file is read by any model with access to your URL: Claude, Perplexity, Cursor, Gemini. If someone asks an AI about your company or your product, that file is the first thing the model will look for. It's already a standard adopted by **Vercel, Anthropic, Supabase** and hundreds of open source projects. If your site doesn't have one, the AI builds its own summary from what it can infer from the HTML, and that summary may be incomplete, outdated or just plain wrong. With llms.txt, you control the narrative. Without it, the model does. ## How to create it today It's a text file. It takes 20 minutes. The process is: 1. Create the `llms.txt` file at the root of your site (the same place a `robots.txt` would live). 2. Write the site description, the main links with one line each, and any important resources. 3. Deploy it along with the rest of the site. You can see real examples in production by opening `vercel.com/llms.txt` or `anthropic.com/llms.txt` directly in your browser. ## Coming up in the next episode With llms.txt, you show the outside world how to read your site. The next step goes deeper: connecting AI directly to your internal tools. In the next and final video of the series, we wrap up with **MCP**, the protocol that connects AI to your real tools.