llms.txt: what it is, who reads it and is it worth adding to your site?
llms.txt is a Markdown file placed in the root of a website, containing a summary of the company and links to its important pages, written with AI models in mind. It is a proposal launched in 2024 by Jeremy Howard (Answer.AI), not a standard. The major AI providers have not publicly confirmed that they use it, and Google states that it is not needed. It is worth adding only if it costs you little.
What llms.txt is
The proposal was published in September 2024 on llmstxt.org by Jeremy Howard, co-founder of Answer.AI. The idea behind it is simple: web pages are full of menus, ads and code, and a language model has a limited context window. A clean Markdown file with the essential information and links to the details would be easier to use.
The file lives at /llms.txt, just like robots.txt. Unlike robots.txt, though, it sets no access rules. It is closer to a summary and table of contents written for machines.
The specification is open to community feedback and has been updated a few times. Some sites also publish an llms-full.txt file with their entire content in plain text, but that is a convention, not a required part of the proposal.
Who actually reads it
This is the part many articles skip. Here is what is known at the time of writing:
- Google states in its documentation on AI features in Search that no special AI files are needed to appear in Google’s AI-generated summaries (AI Overviews) or in AI Mode. There is no confirmation that llms.txt is used there.
- OpenAI, Anthropic and Perplexity document their crawlers and their robots.txt rules, but have not publicly announced that they use other sites’ llms.txt files when generating answers.
- Some AI and software companies publish their own llms.txt for their technical documentation. There, the use is clear: developers hand it straight to coding assistants to load the documentation.
- Independent server log analyses published in 2026 generally show few requests for the file from the major providers’ crawlers, compared with robots.txt.
In short: llms.txt is useful when someone deliberately gives it to a model, for example a developer working with your documentation. For a local business that wants to show up in ChatGPT’s recommendations, its effect is, for now, unproven.
What an llms.txt file looks like
According to the proposal, the structure is as follows:
- an H1 heading with the name of the site or company, the only required element;
- a blockquote with a short summary;
- optionally, a few paragraphs or lists with details, with no other headings;
- H2 sections with lists of links, each followed by a short description;
- optionally, a section called „Optional” with secondary links a model can skip.
Here is an example for a fictional service company:
# Example Heating Service
> Example Heating Service repairs and maintains gas boilers in Iași and
> the surrounding area. Same-day call-outs, annual servicing, original parts.
Hours: Monday to Friday 08:00-18:00. Area: Iași and nearby towns and villages.
## Services
- [Annual boiler service](https://example-heating.ro/annual-service/): what it includes, duration, price
- [Emergency repairs](https://example-heating.ro/emergency/): how to request a call-out, response time
## About
- [About us](https://example-heating.ro/about/): team, certifications
- [Contact](https://example-heating.ro/contact/): phone, address, form
## Optional
- [Blog](https://example-heating.ro/blog/): maintenance tips
How to write it properly
- Use the same information as on your website. Name, address, hours, services. Any mismatch creates confusion, not clarity.
- Write the summary as a direct answer. Who you are, what you do, where and for whom, in two or three sentences.
- Link only to pages worth reading. Service pages, prices, FAQs, contact. Not your whole sitemap.
- Add a short description to each link. Say what the model will find on that page.
- Update it when something important changes. An outdated file with old prices or services does more harm than having none at all.
- Publish it as plain text with UTF-8 encoding, so special characters such as Romanian diacritics display correctly.
How to add it to a WordPress site
You have three options, from the simplest to the most convenient in the long run:
- A file uploaded by hand. Write the file in any text editor, save it as
llms.txtwith UTF-8 encoding and upload it to the site’s root directory via FTP or your hosting control panel. Then check that it opens atyour-domain.ro/llms.txt. - A plugin. Some SEO plugins, as well as dedicated plugins, can generate the file from your site’s pages. Check the result: automatic generation sometimes includes too many links or empty descriptions.
- Generated by the theme or a custom plugin. The file is built automatically from your content and updates itself when you edit your pages. This is the approach we use on the sites we build.
Whichever option you choose, open the file in a browser after publishing. Special characters should display correctly and every link should lead to a page that exists.
llms.txt does not replace robots.txt
If you want AI assistants to cite you, it matters far more than llms.txt that your robots.txt file does not block their search crawlers. Their names are officially published: OAI-SearchBot for ChatGPT, PerplexityBot for Perplexity, Claude-SearchBot for Claude. The full list, with sources, is in the checklist in our GEO guide, and the definitions are in the glossary.
A misconfigured robots.txt can remove you from AI answers entirely. Not having an llms.txt, as far as anyone knows today, does not.
Is it worth adding to your site?
Arguments for
- the cost is low: an hour or two to write it, then occasional updates;
- it makes you spell out who you are and which pages matter, a useful exercise either way;
- if a provider does start using it, you are ready;
- it is useful if you have technical documentation that your customers feed to AI assistants.
Arguments against
- there is no public evidence that it influences recommendations in ChatGPT, Gemini or AI Overviews;
- it needs upkeep, or it becomes a source of outdated information;
- it can give the false impression that GEO (Generative Engine Optimization), meaning optimization for AI assistants, is taken care of, while the things that matter are still missing.
Our recommendation
Add it if you can write it properly and maintain it easily, for example by generating it automatically from your site’s content. But deal first with the things that definitely matter: proper access for crawlers, structured data, pages that answer questions directly, a complete Google Business Profile and identical contact details everywhere. All of these are explained in our article what is GEO.
If you would rather we handle it, our fixed-price service llms.txt and robots.txt for AI crawlers covers both files, written and installed, plus an access check for the main crawlers, all in a single day.