The short answer
An llms.txt file is a plain text file at your web root that gives AI models a curated list of your most important pages. It is a proposed, voluntary standard, not an official rule. Most sites can add it safely, but it is unproven and will not change your rankings on its own.
What an llms.txt file actually is
An llms.txt file is a plain text file placed at the root of your website that offers a curated map of your most important pages to large language models. It sits alongside robots.txt and sitemap.xml. The proposal, published in 2024, suggests one Markdown file that points AI systems to the content you most want them to read and understand.
The idea is simple. Web pages carry navigation, adverts, scripts and boilerplate that waste an AI model's limited context. An llms.txt file strips that away and gives a clean list of links with short descriptions. It is written for machines that read Markdown well. Think of it as a reading list you hand to an assistant, rather than asking it to wander your whole site.
One point causes confusion. The file is a proposed standard, not an official rule that every AI company has agreed to follow. It was suggested by Jeremy Howard of Answer.AI. Adoption is voluntary. Some tools read it, many do not yet, and no major search engine has confirmed it as a ranking signal. Understanding that distinction stops you from treating the file as a magic switch.
Does your website need one?
Probably not urgently, but it rarely hurts to add one. If you run a documentation site, a software product or a large content library, an llms.txt file helps AI systems find your key pages quickly. If you run a small brochure site with ten pages, the benefit is slight. The honest position today is that it is low cost, low risk and unproven, so treat it as optional housekeeping.
Do not expect it to change whether ChatGPT or Perplexity names your business tomorrow. Those systems mostly learn from what the wider web says about you, from pages they crawled during training and from live search results they fetch at answer time. An llms.txt file can make your own content easier to parse, but it cannot manufacture the third party mentions and citations that push a brand into an AI answer.
- Documentation and developer sites with deep, structured content.
- Software and SaaS products where accurate feature descriptions matter.
- Large publishers with hundreds of articles to prioritise.
- Small brochure sites gain little and can safely skip it for now.
What goes inside the file
The format is deliberately plain. It starts with a single H1 heading that names your site or company. A short blockquote summary follows. After that you list sections as Markdown headings, and under each heading you place links written in standard Markdown format, each with a brief note explaining what the page covers. That is the whole structure, and it is meant to stay readable by a person.
Some sites also publish a longer companion file called llms-full.txt. That version includes the full text of the linked pages in one document, so a model can read everything without following links. Keep the descriptions honest and specific. Name the product, state what each page does and avoid marketing language. A model treats your words as fact, so vague or inflated claims can lead to a poor summary of your business.
Place the file at the top level of your domain so the address ends in /llms.txt. Use absolute links, not relative ones, so any system can resolve them. Update the file when you publish or retire important pages, because a stale reading list points models at content that no longer exists. Treat it the way you treat your sitemap, as a living index rather than a one time task.
How to create one
You can build the file by hand in a few minutes, and that is often the best approach because you control what gets prioritised. There are also generators that read your sitemap and draft a first version, though you should always review the output. Below is a straightforward process that works for most sites, whether you run WordPress, a headless build or a hand coded set of pages.
List your key pages
Pick the ten to thirty pages you most want AI systems to understand, such as your product pages, pricing overview, documentation and main service pages.
Write the file
Create a Markdown file with one H1 title, a one line blockquote summary, then grouped links with a short note on each.
Save it as llms.txt
Save the file as llms.txt and upload it to the root of your domain so it loads at yourdomain.com/llms.txt.
Keep it current
Review it whenever you add or remove important pages, and add an llms-full.txt version if your content is short enough to include in full.
Test the result by opening the address in a browser to confirm it loads as plain text. Then check that every link resolves and points to a live page. There is no submission step and no dashboard to register with, because no central authority validates the file. You simply publish it and keep it current, the same way you would maintain any other file at your web root.
What matters more than llms.txt
If your goal is to be named inside AI answers, spend most of your effort elsewhere. AI systems cite sources they trust, and trust is built by other websites referencing you, by clear consistent information about your business across the web, and by content that answers real questions directly. An llms.txt file is a tidy final touch, not the foundation. Get the reputation signals right first.
Make sure AI crawlers can reach your site at all. Check that your robots.txt does not block GPTBot, which OpenAI uses, or PerplexityBot, or Google-Extended, which controls Google AI training access. Many businesses accidentally block these while trying to stop scrapers. Then structure your pages so the answer comes first, use plain headings that match how people ask questions, and back claims with facts a model can quote.
So add the file if you have the time, because it is cheap and may help as adoption grows. Just keep your expectations realistic and your priorities in order. The businesses that get recommended by AI assistants are the ones the wider web already describes clearly and often. A single text file at your web root supports that work, but it will never replace it.