llms.txt generator: the file in ten seconds, and the truth about what it does
Type your site address. The tool reads your sitemap, then the title and description of your pages, and writes the file in the format of the llmstxt.org proposal. Nothing is made up: every line comes from a tag on your pages.
One thing before you start. As of 17 October 2026, no AI engine has confirmed reading this file, and server log studies show their crawlers request it very rarely. It costs nothing and does no harm. What counts for being cited lies elsewhere: the sources are below.
Adrien Asselineau
SEO consultant for eleven years, publisher of Suiko
Why build a tool for a file that is almost useless
I have been asked about llms.txt at almost every onboarding for a year. The honest answer fits in one sentence: to date, nobody has shown that it changes anything about how ChatGPT, Gemini or Google AI Overviews cite a site. The server log studies published in 2026 count requests for this file in the dozens, against thousands for robots.txt, and SE Ranking’s analysis of 300,000 domains finds no effect on citations.
So why this tool? Because the question keeps coming, the file does no harm, and producing it has a useful side effect: it puts the titles and descriptions of your forty main pages in front of you, as robots read them. That is often where you discover a description from 2019 or a duplicated title. Ten seconds for the file, ten minutes for that review.
What gets a page cited is something else: answering the question from the first lines, with dated and sourced facts, on a page Google already ranks. That is the groundwork Suiko has writers do, with a score and a knowledge base. The file takes ten seconds.
What the tool reads, and what it does not do
Your sitemap
The tool reads robots.txt to find the sitemap, then the sitemap itself, up to 400 addresses. If there is none, it starts from the links on the home page. It then reads up to forty pages, shallowest first, spread across sections.
Your tags, nothing else
The site name comes from the og:site_name tag or the home page title, the summary from its meta description, each line from the page’s title and description. No AI is involved: if a page has no description, the line has none.
A file to review
Sections follow your addresses (blog, products, tools…). The descriptions are those of your tags, sometimes old. Review them before uploading the file to the root of the site, and take the opportunity to fix the tags themselves.
No cookie, no tracker: your email goes to my inbox and nowhere else. The site read is not kept.
llms.txt in eight questions
What I get asked about this file, with dated sources so you can check.
What is an llms.txt file?
A Markdown text file placed at the root of a site that describes the site to AI assistants: a title, a summary as a blockquote, then sections of links with one line per page. It is a proposal published on 3 September 2024 by Jeremy Howard, co-founder of Answer.AI. It is not a standard: no body has adopted it, and only the title is required by the format.
Do AI engines really read this file?
Very rarely, according to published logs. A study across 83 sites over 12 weeks (April to July 2026) counts 7 llms.txt fetches by OpenAI’s crawler against 3,990 for robots.txt, 9 against 3,120 for Anthropic’s, 0 against 775 for Perplexity; only Meta’s crawler read it more. A single-site census (August 2026) counts 4 fetches in twenty days for Google, Anthropic and OpenAI combined, against 1,150 for robots.txt. None of these companies has confirmed using it.
Does llms.txt help you get cited, or rank on Google?
Nothing shows it. SE Ranking analysed nearly 300,000 domains (15 October 2026), 10% of which have an llms.txt, and concludes that the file has no measurable effect on how AI systems see or cite content today. On Google’s side, John Mueller explained in June 2026 that these systems cannot, by design, trust a self-declared file to choose between sites, and sees only a possible use for an agent already on your site.
So why make one?
Three honest reasons. It costs nothing and does no harm. It can help an agent already on your site find its way, the use Google concedes to it. And producing it puts the titles and descriptions of your main pages in front of you, what robots actually read, where old tags show immediately. If you are doing it for any other reason, be wary of whoever sold it to you.
Where do I put the file, and how do I check it is live?
At the root of the site, at https://yoursite.com/llms.txt, UTF-8 encoded, served as plain text. Upload it with the other site files (robots.txt lives in the same place) or through your CMS: most WordPress SEO plugins can do it. To check, open the address in a browser: the text should display as is.
Where do the lines of the generated file come from?
From the sitemap (sitemap.xml, found through robots.txt) for the list of pages, then from each page read: the title or og:title tag for the label, the meta description for the line that follows. The site name comes from og:site_name or the home page title. Sections take the first segment of your addresses. The tool reads at most forty pages; on a large site, complete the missing sections by hand.
What about llms-full.txt and .md versions of pages?
The proposal also suggests publishing each page in Markdown at the same address followed by .md, so an agent reads the text without the HTML. llms-full.txt, which puts the whole site content in a single file, is a convention that appeared later in technical documentation; the proposal does not describe it. Both take work for a benefit nobody has measured. I do not recommend them today.
What should I do to get cited by AI engines, then?
What cited pages have in common: they answer the question from the first lines, in simple sentences, with dated and sourced facts, and Google already ranks them well for the query. To see who is cited today on your keywords, the AI Overview checker is next door. To keep your facts right over time, that is the role of the knowledge base in Suiko.