# llms.txt for SaaS Websites: What to Include > What an llms.txt file is, how SaaS websites can structure one, which pages to include, and realistic expectations for its impact on AI visibility. Source: https://comppound.com/blog/llms-txt-for-saas-websites-what-to-include (published 2026-07-06) llms.txt is a proposed standard for giving large language models a curated map of a website. It is a plain Markdown file placed at the root of your domain, usually at /llms.txt, that lists your most important pages with short descriptions. For llms.txt on SaaS sites, the idea is straightforward: instead of an AI system crawling your whole site and guessing which pages matter, you point it to the content that best explains your product, your documentation, and your expertise. Support for llms.txt varies across AI providers and is still evolving, so it is not a guaranteed visibility boost, and anyone who sells it as one is overselling. It is, however, cheap to create, and it forces a useful exercise: deciding which pages actually explain your company well. Many SaaS teams discover during that exercise that their best explanation of the product is buried, outdated, or missing. This article covers what the file looks like, what a SaaS site should include and leave out, how to maintain it, and what to realistically expect from it. ## What llms.txt looks like The proposed format is Markdown with a few conventions. The file starts with an H1 containing the product or site name. Directly below it comes a blockquote with one or two sentences describing what the product does and who it is for. After that you can add a short paragraph of extra context about the company or product, such as its main category or the problems it solves. The rest of the file is organized under H2 headings that group related links, such as Docs, Guides, and Comparisons. Each entry under a heading is a Markdown list item containing a link and a short description after a colon, for example `- [Getting started](https://example.com/docs/getting-started): Set up your account and first project.` A section titled Optional is conventionally used for links that are useful but secondary, such as a changelog, so that systems with limited context can skip them. That is essentially the whole format. Because it is plain Markdown, it is readable by people as well as machines, which makes it easy to review and edit. ## What SaaS sites should include ### A clear product summary The summary at the top matters more than any individual link. It should state what your product does, who it is for, and the category it belongs to, in plain language. Use the same wording you use on your homepage, in your directory listings, and in your sales materials. When an AI system encounters consistent descriptions of your company across many sources, it is more likely to form an accurate picture. When your homepage says one thing, your LinkedIn page another, and your llms.txt a third, you are making that harder. Avoid marketing superlatives here. A sentence like "the leading platform for revolutionizing workflows" tells a model nothing. A sentence like "scheduling software for clinics with multiple practitioners, handling bookings, reminders, and payments" tells it exactly what you are. ### Core product pages Include the pages that describe what you sell. That usually means the homepage, the main features or product overview page, the pricing page, your key use-case or solution pages, and your integrations overview. Pricing is worth including even if your pricing is not fully public, because buyers frequently ask AI assistants about cost, and a page that explains your pricing model is better than the assistant guessing. Write each description to say what someone will learn from the page, not just restate its title. "Plans, per-seat pricing, and what is included in each tier" is more useful than "Pricing." ### Documentation If you have public documentation, it is often the most valuable content you can list. Include the getting started guide, the main feature guides, the API reference if you have one, and troubleshooting or FAQ pages. Documentation is what AI assistants need when users ask how to do something in your product, and accurate answers to those questions help both existing customers and prospects evaluating whether your product can handle their use case. For products with large documentation sites, do not list every page. List the entry points and the most-used guides, and rely on the documentation's own navigation for the rest. ### Your best educational content Choose guides that are accurate, thorough, and representative of your expertise. Do not list every blog post. A good test is whether you would be happy for an AI system to cite this page when answering a question in your area. Older posts that no longer reflect your thinking, short news items, and posts written mainly to target a keyword are better left out. ### Comparison pages Honest comparison and alternative pages can help AI systems answer questions such as how your product compares with a specific competitor. These questions are common in B2B buying, and an assistant that can find a clear, fair comparison from you is more likely to describe your differences accurately. The emphasis is on honest. A comparison page that misrepresents a competitor is a liability whether a person or a model reads it. ### Company and trust pages Include your about page, your security or compliance page, and your contact or support page. Security and compliance questions come up early in B2B evaluations, and pointing AI systems to an authoritative page reduces the chance that they repeat outdated or invented information about your certifications. ## What to leave out Leave out thin or outdated posts, login, account, and checkout pages, duplicate pages and tag archives, and internal or experimental pages. The purpose of the file is curation. Every low-value link dilutes the signal of the high-value ones, and pages that require a login are of no use to a system that cannot log in. Also leave out anything you would not want quoted. If a page contains claims you no longer stand behind, either fix the page or keep it off the list. ## llms-full.txt Some sites also publish /llms-full.txt, a single file containing the full text of key pages rather than just links to them. This can be useful for documentation-heavy products, because it lets a system read your core documentation in one request. The risk is staleness. A full-text file is a copy of your content, and copies drift out of date when the original pages change. Outdated content in a full-text file can lead to inaccurate answers about your product, which is worse than no answer. If you publish one, generate it automatically from the source content rather than maintaining it by hand. ## Keeping it maintained An llms.txt file that points to broken or outdated pages is worse than none. Build a review into your content process. Update the file when you launch major features or documentation, check its links on a regular schedule such as quarterly, and remove pages you have deprecated or redirected. Some teams generate llms.txt automatically from their CMS or documentation platform, with manual curation deciding which pages are included. That combination works well: automation keeps URLs and descriptions current, while a person decides what belongs on the list. Whoever owns your website should also own this file, so it does not become an orphan. ## Realistic expectations llms.txt is not a ranking factor in traditional search, and AI providers do not all read it. Think of it as a small, low-cost step that can help the systems that do use it, alongside work that matters much more. That more important work starts with making key content accessible in plain HTML, since many AI crawlers do not run JavaScript and will miss content that only appears after scripts load. It includes allowing the AI crawlers you want in your robots.txt, publishing clear, well-structured pages that answer the questions buyers actually ask, and earning mentions on third-party sites such as reviews, communities, and industry publications. AI systems lean heavily on what others say about you, not only on what you say about yourself. ## How to create one Start by listing your most important URLs. For most SaaS companies that is somewhere between twenty and fifty pages. If you have far more than that, you are probably not curating enough. Group them into sections such as Product, Docs, Guides, Comparisons, and Company. Write a one-line description for each link that says what the page covers. Then write the short product summary for the top, ideally reusing the wording from your homepage. Save the file as llms.txt at your domain root and check that it loads in a browser as plain text, without your site's layout wrapped around it and without redirects to a login page. If your site runs on a platform that does not let you place files at the root, look for a redirect or static file option, or ask your developer to add a route. ## A curation pass in practice (illustrative) Imagine a small project management SaaS with a few hundred blog posts, a documentation site, and a handful of product pages. The marketer drafting its llms.txt starts by exporting every URL, then cuts aggressively. Product pages and pricing stay. From the documentation, only the getting started guide, the five most-viewed feature guides, the API overview, and the troubleshooting hub make the list. From the blog, she keeps about a dozen guides that the sales team already sends to prospects, because those are the pages the company is confident are accurate. Two comparison pages stay after a quick review to confirm they still describe the competitors fairly. The process takes an afternoon. Along the way she finds that the integrations page lists a connector that was retired months ago, and that the security page mentions an audit that has since been renewed. Fixing those pages is arguably more valuable than the file itself. ## Check your robots.txt too An llms.txt file does not override robots.txt. If you block AI crawlers in robots.txt, they may not access your pages regardless of what llms.txt says. Review which crawlers you allow and make an intentional decision based on your goals. Some companies want maximum visibility in AI answers and allow everything. Others block training crawlers while allowing those used for live search and answers. Either can be reasonable, but it should be a decision rather than a default someone set years ago. ## The bottom line For most SaaS websites, llms.txt is a quick organizational win rather than a major traffic lever. Create a curated, accurate file with a clear product summary and your best pages, keep it updated, and check that your robots.txt agrees with it. Then put most of your effort into accessible, well-structured content and a strong reputation on other sites, which drive AI visibility far more than any single file. ## Frequently asked questions ### What is llms.txt? A proposed Markdown file at a website's root that lists important pages with descriptions, intended to help language models understand the site. ### Does llms.txt improve SEO rankings? It is not a traditional search ranking factor, and support among AI providers varies. ### Should every SaaS site have an llms.txt file? It is low effort and can help with AI systems that read it, so many sites add one, but it should be kept accurate and up to date. ## About Comppound Comppound runs SEO and AI search for B2B SaaS companies: a person sets the plan and reviews samples and every high-risk page, and AI speeds up production. - [Get the free report](https://comppound.com/ai-visibility-report): When buyers ask ChatGPT about your category, are you in the answer? - [See SEO Growth](https://comppound.com/services/seo-growth): Buyers compare before they book a demo. Are your pages there when they do?