Does llms.txt Do Anything? An Honest Answer
llms.txt is widely recommended and thinly evidenced. Here is what the convention proposes, what the major AI operators actually document, and when it is still worth twenty minutes.
Paste your sitemap.xml or a list of URLs and get a valid, well-structured llms.txt in seconds: an H1, a one-line summary blockquote, and your pages grouped into readable sections. But first, the thing most llms.txt tools won't tell you: Google stated in 2026 that llms.txt is not used by its AI search, and there is no public evidence that OpenAI, Anthropic or Perplexity read it for ranking either. It's cheap and harmless, some agents do fetch it, and it is not the lever it's marketed as.
The llms.txt Generator builds a spec-shaped llms.txt from your sitemap.xml or URL list, grouping pages into readable sections with humanized titles and a summary blockquote. It also states the thing most llms.txt tools omit: Google has said its AI search does not use llms.txt, and no major AI operator documents it as a ranking input. Useful for agents that do fetch it, not a ranking lever.
llms.txt is a proposed convention: a Markdown file at your domain root that gives language models a curated map of your site, a name, a one-sentence summary, and grouped links with short notes. The idea is sensible and the format is pleasant. The marketing around it has run far ahead of the evidence. Google has publicly said llms.txt plays no role in its AI search systems. None of the major AI companies have documented using it as a ranking or retrieval input, and the sites that rank well in AI answers overwhelmingly do so without one. What llms.txt genuinely does help with: agents and tools that fetch it deliberately (some coding assistants, some documentation crawlers, some internal RAG pipelines pointed at your docs), and your own clarity, the exercise of writing one sentence describing your site and picking the twenty pages that matter is worth doing regardless of who reads the output. So this generator does the job properly and tells you the truth on the same page. Paste a sitemap and it parses the XML, groups URLs by first path segment into readable sections, humanizes the slugs into titles, and caps each section so a 300-URL sitemap becomes a file a model would actually read rather than a wall of links. Write your summary, and you have a correct file. Ship it if you want the coverage. Just don't expect it to move rankings, and be suspicious of anyone who tells you it will.
Last updated
The summary becomes the blockquote: the single line an agent reads first. Say what the site is and who it's for, concretely.
Raw XML is detected and parsed automatically. Plain URLs work one per line. Everything stays in your browser.
URLs are grouped by path segment with humanized titles. Edit your inputs until the grouping reflects how you'd describe the site.
Place it at https://yourdomain.com/llms.txt. It's a static Markdown file: no server logic required.
Don't measure this file's impact on rankings. There's no evidence to expect any. Spend the effort you saved on crawler access, rendering and citable content.
What the evidence supports and what it doesn't, stated plainly next to the generator. You'll leave knowing whether you actually need this file: including if the answer is no.
The XML is parsed automatically: no converting to a list first. Plain URL lists work too, one per line.
URLs are grouped by first path segment (/docs, /blog, /guides) with slugs humanized into titles, so a large sitemap becomes a readable document instead of a link dump.
H1 name, blockquote summary, optional detail paragraph, then H2 sections of `- [title](url): note` links, matching the llmstxt.org convention.
Sections are capped at 25 links so the file stays within the length a model will actually consume. A 300-URL sitemap becomes a curated map, not a sitemap in Markdown.
Your sitemap and URLs never leave your device. The file is generated locally and downloaded directly.
Most llms.txt content is written by people selling llms.txt services. Knowing that Google has ruled it out for AI search saves you from treating a nice-to-have as a priority.
If you want the coverage for agents that do fetch it, the cost should be five minutes, not a project. Paste, write one sentence, download.
Summarizing your site in one sentence and choosing the twenty pages that matter is genuinely useful thinking, and it feeds directly into your meta descriptions and entity consistency.
Auto-generated llms.txt files that dump every URL unsectioned are worse than none. Grouping and caps produce something a model could actually use.
From quick one-off fixes to daily workflows, see how people put this tool to use.
Coding assistants and doc-focused agents do fetch llms.txt when pointed at a docs domain. If your product has developer docs, this is the strongest real use case.
"Do we need llms.txt?" comes up in every AI-SEO conversation. Now you have a sourced answer and a five-minute implementation if they want it anyway.
Turning a sitemap into grouped Markdown by hand is tedious and gets stale. Regenerate in seconds whenever the sitemap changes.
If you're building retrieval over your own site, llms.txt is a decent hand-curated index of what matters: regardless of what external crawlers do with it.
Boundaries stated plainly, with the right tool for each neighbouring job.
Accepts XML, URL list and TXT, and produces llms.txt and MD, all processed locally in your browser.
Check whether GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, Bingbot and every other AI crawler can read your site. Real RFC 9309 robots.txt matching with the deciding rule quoted, plus Cloudflare default-blocking detection. Free, instant, no signup.
One URL in, one AI-visibility report out: crawler access with Cloudflare detection, JS rendering, citability scored with published methodology, retrieval chunks, answer snippets and schema. Each finding linked to the free tool that fixes it.
Turn a list of URLs into a valid sitemap.xml with per-URL changefreq, priority and lastmod, validated against the 50,000-URL limit and ready to submit to Google. 100% in your browser.
Audit a page's JSON-LD against the seven schema types AI retrieval leans on, verify each one's load-bearing properties, and get every gap routed to the GigAI generator that fills it. A checker, not another generator.
Generate a correct robots.txt policy for every AI crawler (GPTBot, ClaudeBot, OAI-SearchBot, PerplexityBot and 11 more) with per-bot allow/block toggles and a plain-language line on what each choice costs. Free, in your browser.
Drop a server access log and see which AI crawlers actually visit: per-bot hit counts, first and last seen, most-fetched pages, status-code health, daily trend, and the pages no AI crawler has ever touched. Parsed in your browser: logs contain visitor IPs and never leave your device.
Hit a snag? Here are quick fixes for the issues people run into most.
That's the expected outcome, and it's why this page says so up front. Google has stated its AI search doesn't use the file, and no major AI operator documents it as a ranking input. If AI visibility is the goal, crawler access, JavaScript-free rendering and citable content are where the measurable gains are: run the GEO Audit for a prioritized list.
Paste the raw XML contents (view-source of sitemap.xml), not the sitemap URL. If yours is a sitemap index pointing to child sitemaps, paste one child sitemap's contents: the index only contains links to other files, not page URLs.
Grouping follows your URL structure, so the output is a mirror of your information architecture. If the grouping looks wrong, that's worth noticing about the site itself. For full control, paste a hand-picked URL list instead of the whole sitemap.
No. The point of llms.txt is curation: a map, not an index. You already have a sitemap for completeness. Include the pages you'd show someone who asked what your site is. The 25-per-section cap enforces that discipline.
Write the summary line as if answering "what is this site?" in one sentence to a stranger. That line does more work than the entire link list.
Keep it short. A curated 30-link file is more useful to a model than a 300-link dump, and far more likely to be read in full.
There's a proposed llms-full.txt convention that inlines whole page contents. It's even less established: do the basic file first, if at all.
Add notes after links (`- [Setup guide](/docs/setup): install and configure in 5 minutes`), the note is where the useful signal is.
Deploy it as a plain static file at the domain root. No headers, no negotiation, no dynamic generation needed.
Recent updates and improvements to the llms.txt generator.
Initial release: sitemap.xml auto-parsing, URL-list input, automatic sectioning by path segment with humanized titles, 25-link section caps, spec-shaped Markdown output, and on-page honest guidance about llms.txt's actual standing.
Generation happens entirely in your browser. Your sitemap, URLs and descriptions are never uploaded, logged or stored: the download comes straight from your own device.
Free, private and instant. llms.txt Generator runs right in your browser.