Speaker Hub uses necessary cookies for secure accounts. With your permission, first party analytics help us improve signup and onboarding. We do not use advertising or cross site tracking cookies. Read the Cookie Policy.
llms.txt: what it is, and who is actually reading it · Speaker Hub
llms.txt: what it is, and who is actually reading it
A proposed standard for telling AI crawlers about your site. Before you spend an afternoon on one, here is the honest state of adoption.
•6 min read
Someone in a speaking group told you that you need an llms.txt file or the AI tools will not find you. Here is the actual state of it.
What it is
llms.txt is a proposal published by Jeremy Howard in September 2024. The idea is a markdown file at the root of your domain, at yoursite.com/llms.txt, containing a short summary of the site and a curated list of the pages worth reading.
The reasoning behind it is sound. A language model working with a limited context window cannot read your whole site. Modern web pages are mostly navigation, scripts and styling, so a crawler extracting your actual content is doing a lot of work to find a little text. A hand written summary skips all that.
The format is simple:
# Jane Doe
> Keynote speaker on burnout and staff retention in healthcare.
> Former ICU nurse, eleven years.
## Talks
- [Why experienced nurses leave](https://janedoe.com/talks/retention): the
signature keynote, 45 minutes, for hospital leadership audiences.
- [Handover](https://janedoe.com/talks/handover): workshop format, 90 minutes.
## About
- [Speaker one sheet](https://janedoe.com/speaking): topics, fees, video, contact.
Markdown, plain language, links with a line of description each.
Who reads it
This is the part that usually gets skipped, so here it is plainly.
llms.txt is a proposal. It is not a standard published by a standards body, and it is not a documented input to any major search engine. Google's guidance on how to control crawling continues to point at robots.txt, and Google has not published documentation saying it reads llms.txt.
Some developer tooling and documentation sites have adopted it, and some AI products consume it in specific contexts. What there is no evidence for is the claim that it affects how you appear in Google, or that adding one gets you cited by a chatbot.
If someone tells you otherwise, ask them for the documentation. That request usually ends the conversation.
Why it might still be worth twenty minutes
Two honest reasons.
It costs almost nothing. A speaker site has maybe six pages worth listing. That is a twenty minute job, once, and there is no downside beyond the file existing.
The exercise is useful on its own. Writing one forces you to answer, in two sentences, what you speak about and which pages matter. Most speakers find they cannot do that cleanly, which is information about the site rather than about the file.
If you write one and then notice your homepage does not say what the llms.txt says, the homepage is the problem.
What definitely works instead
If you have an afternoon for being found by answer engines, spend it here first. These are documented, not proposed.
Make the content crawlable text. Anything that only exists inside an image, a video, or a PDF is doing a fraction of the work. This is the most common failure and it is invisible, because the page looks fine to you.
One page per topic. A page about a single subject is a clear answer to a question about that subject. A page about nine subjects is a weak match for all nine, which is covered in your speaker bio.
Say the claim in a sentence. Answer engines quote sentences. A page whose central point is spread across a video, a testimonial and an implication has no sentence to quote. Write the sentence.
Structured data with sameAs. This is how you connect the person on your site to your other profiles, and it is documented. See speaker schema markup.
Get cited elsewhere. What you say about yourself is one signal. What conference sites, publications and other people's pages say about you is a stronger one, and it is not something a file at your root can manufacture.
The pattern to watch for
llms.txt is the current example of a recurring thing in this field. A plausible technical idea appears, gets repeated as a requirement, and becomes a checklist item long before anyone confirms who consumes it.
The same happened with FAQ schema, which speakers were told to add for rich results in Google right up until Google removed the feature. The post on getting quoted by AI goes through what survived that.
The test is always the same. Who reads this, and where did they say so? If the answer is a blog post citing another blog post, it is folklore, however sensible it sounds.
What to do
Add the file if you like. It takes twenty minutes and it may matter later.
Do not let it move ahead of a site that can actually be crawled, pages that each say one thing, and a bio containing a sentence worth quoting. Those work today.
A proposal by Jeremy Howard, published in September 2024, for a markdown file at the root of a site that gives large language models a curated summary and a list of the pages that matter. It is modelled on the idea of robots.txt but it does a different job, describing content rather than restricting crawling.
Does Google read llms.txt?
There is no documentation from Google saying it does, and Google's own guidance on crawling continues to point at robots.txt. Treat any claim that llms.txt affects Google rankings as unsupported until Google publishes something that says otherwise.
Should I add one anyway?
It is cheap, so it is defensible if you enjoy the tidiness. Just do not do it instead of the things that are known to work. A crawlable site with clear text, one topic per page and correct structured data is what an answer engine reads today, whether or not it ever looks at your llms.txt.
Google deleted FAQ rich results in 2026, and it never read llms.txt at all. Here is what survives once you strip out the advice that stopped being true.
