Skip to content
EnterraHost
Products
Tools
Hosting Email Support Client Area Find a domain

llms.txt for Joomla

llms.txt is a proposal, not a standard. It was published as a convention for telling a language model what a site is and where the useful parts are, and no standards body has adopted it. That matters for how much work it deserves. A file almost nobody reads is not worth a week, and a file that takes an afternoon is worth having before the crawlers that do read it arrive.

The convention lives at llmstxt.org. The short version is a Markdown file at the site root with an H1, a one line summary and links grouped under headings. There is no schema, no validator, and no required field beyond the title.

Joomla has no llms.txt of its own, and no sitemap either. That is the whole reason FindSignal for Joomla exists. Both files are built from the content Joomla already holds, and neither needs an account or a third-party service.

The whole FindSignal dashboard in the Joomla administrator, top to bottom. The header carries the Dashboard and Settings tabs. The score card reads 100 out of 100, agent readiness, readable, 23 of 23 agents allowed, with bars for agent access at 40, content signals at 15, llms.txt at 20, sitemap linked at 15 and Markdown at 10. The files card lists the llms.txt, llms-full.txt, Markdown and sitemap addresses for the site and marks robots.txt as a physical file. The environment card shows the template, the sitemap owner, the physical robots.txt and the robots setting in Global Configuration. Below them the Search visibility card lists Global Configuration setting robots to noindex, with a Make this site searchable button. What an agent sees offers an article picker and a Render Markdown button above a dark panel of rendered Markdown. Schedule and Content reach show nine pages and 553 words. Agent activity lists GPTBot, Amazonbot, Applebot-Extended, Bytespider, CCBot, ClaudeBot, Google-Extended, OAI-SearchBot and PerplexityBot with request counts, above a longer request log. Can these agents fetch the files carries a 23 row table with a column each for the root, llms.txt and llms-full.txt, every row Allowed. Help and tools closes the page with eight links.
The whole dashboard on a Joomla 6.2 site. The score is built from the same five measurements the checker on this site uses, and everything below it is the evidence for that number.

What the extension publishes

Each switch saves itself the moment you change it, and a small line under the buttons says Saved. There is a Save settings button as well, and both of them rebuild the files.

Nothing is published until you turn it on. Every switch that writes a file starts off, because a site may already have a sitemap extension or an llms.txt plugin in place, and two components publishing the same file is a conflict nobody asked for. Turn on what this site wants, save, then press Rebuild the files.

Languages

The interface ships in eleven languages. English in both UK and US spelling, German, Spanish, French, Korean, Portuguese in both Brazilian and European forms, Russian, and Chinese in Simplified and Traditional.

Joomla uses the language of the site or the administrator. When a language has no files it falls back to English, so an English site needs no configuration at all.

A translation becomes active when its language pack is installed. Joomla only copies language files into a folder that already exists, so nothing has to be switched on, and nothing arrives before the pack does.

There is no separate Cantonese option. Cantonese is spoken, and it is written in Traditional Chinese, which is the Chinese Traditional entry.

Four files are written into the site root once those switches are on.

  • /llms.txt carries the site name and description, then every public page with one line each.
  • /llms-full.txt carries the same pages as full Markdown, in one file.
  • /sitemap.xml is a valid sitemap, and it steps aside automatically when OSMap, JSitemap or Aimy is installed and serving its own.
  • /ai.txt is a pointer file, the same one FindSignal for WordPress serves. It names the two text files and nothing else, and deliberately carries no contact address, because a public root file is the first place an address harvester looks.

It also answers a Markdown request for any page. Add ?format=md to a page URL, or send Accept: text/markdown. A browser still gets HTML, because the negotiation is on the header and not on a separate address.

Why the files are written to disk

Most Joomla extensions serve a file like this through a component route, which produces something like index.php?option=com_something&file=llms.txt. That works, and it is wrong for this file, for two reasons.

The first is that a reader has to know the route. The convention says the file is at the root, so a checker, a crawler or a person will fetch https://example.com/llms.txt and nothing else. A route that needs URL rewriting answers 404 on a site with rewriting off, which is the default on a new Joomla install.

The second is that this file describes the site, so it should survive the component being disabled. A physical file is served by the web server without starting PHP.

Where the site root is not writable, the same four files are served dynamically at index.php?findsignal=llms.txt, and the dashboard says which mode is in use. That fallback is a real one, not a placeholder.

What Joomla does not give you

Joomla controls search visibility in two places, and they are not the same thing. Global Configuration → Metadata → Robots emits a robots meta tag on every page. A host can send an X-Robots-Tag header from .htaccess, which outranks both the meta tag and robots.txt. A site can therefore look fine in Joomla and still be hidden.

The component's Search visibility card reads both and lists what is keeping the site out of search, including a header it cannot remove because the host owns the file. On a site with no .htaccess at all, which means nginx or a default install, it reports nothing rather than claiming it cannot tell.

Per-article control is Joomla's own. The Metadata tab on an article sets its robots value, and the access level decides who may read it. Both are honoured, which is the next section.

The same rule robots.txt follows

An llms.txt that lists a page a visitor cannot open is a file that leads a crawler into a wall. So the files carry only what a guest could reach. That excludes four kinds of page.

  • Articles that are unpublished or trashed.
  • Articles scheduled for a future date, and ones past their removal date.
  • Articles at an access level a guest does not hold, such as Registered, and articles in a category that is unpublished or restricted.
  • Articles whose Metadata field sets noindex.

The Markdown path applies the same gate, so a restricted article is refused there too. Joomla 5 answers 404 where 6.2 answers 403. Neither serves the text, and the difference is Joomla's rather than the extension's.

The Agent activity card in the Joomla administrator. A heading reads Agent activity, with a note that requests to the files and the Markdown URLs in the last seven days are matched to a known name and that the raw user agent is never stored. A summary table lists GPTBot and Other at eleven requests each, then Amazonbot, Applebot-Extended, Bytespider, CCBot, ClaudeBot, Google-Extended, OAI-SearchBot and PerplexityBot at three each, with the time each was last seen. Below it a filter set to All agents with a Filter button, then a log of individual requests with a timestamp, the agent name and the requested path, every path shown as /md, ending at Page 1 of 4 with 46 requests recorded.
Which agents asked, and when. Every request here is a Markdown fetch, because the root files are served by the web server and never reach PHP, so the plugin can only see the requests it answers itself.

robots.txt, and the line worth adding

robots.txt and llms.txt do different jobs, and confusing them is common. robots.txt is a crawling rule with a specification behind it, RFC 9309. It decides what may be fetched. llms.txt is a courtesy note that changes what a reader gets after it decides to fetch something.

The extension can add two lines to robots.txt. The first is a Content-Signal line saying whether the content may be used for search results, as input to an answer, and for training. The second is a Sitemap line, because Joomla's own robots.txt does not have one.

Both are written between two marker lines and nothing outside them is touched. That write is deliberately manual. robots.txt is a file a host may own, and an extension that edits it on install is an extension that can change crawling rules without being asked.

The whole Settings tab in the Joomla administrator. What the site publishes carries five switches, for llms.txt, llms-full.txt, sitemap.xml, serving a page as Markdown, and declaring a Content-Signal line, all set to Yes, then three selects for Search, AI input and AI training set to yes, yes and no, and a switch to keep the activity log when the extension is removed, set to No. Advanced carries a rebuild interval of 24 hours, reading rendered pages set to Yes, twenty rendered pages per rebuild, three excluded article IDs, a Sitemap line switch set to Yes, and a schedule switch set to No. The robots.txt card says the block is in robots.txt, shows the Content-Signal line reading search equal yes, ai-input equal yes, ai-train equal no, and carries a link to see the full robots.txt. The switches save themselves, so the only button at the bottom is Cancel, and Save settings appears only when JavaScript is unavailable.
The settings tab. The three Content-Signal choices are separate because search, answering and training are separate permissions, and the robots.txt card is the only place that writes a file the host may own.

How to check what a crawler receives

Three commands answer most questions, and they are worth keeping in a terminal.

curl -sI https://example.com/llms.txt
curl -s https://example.com/llms.txt | head -20
curl -sI -H 'Accept: text/markdown' https://example.com/some-page

The first should be 200 with content-type: text/plain. The second shows the index itself. The third should be text/markdown and not text/html.

The dashboard does the same checks from inside Joomla and scores them, using the same weighting as the AI visibility checker on this site. It also lists, per known agent, whether robots.txt allows each of the three files.

When it does not work

/llms.txt answers 404. Check the plugin is enabled, then that the site root is writable by PHP, then that the files have been rebuilt since install. The dashboard's Environment card names the mode it is serving in.

The files are there but stale. A save rebuilds them. An import that writes to the content tables directly does not fire a save event, so the component compares a fingerprint of what is public on each page view and rebuilds when it has moved.

Content-Signal does not appear in robots.txt. The block is only written when the setting is on and the button is pressed. The robots card says which of the two is missing.

The dashboard says the files were never built, and they exist. The component's cache directory is not writable by PHP, so the statistics cannot be saved. Give cache/com_findsignal to the PHP user, or make the site cache directory group writable.

The site is hidden and the card says so. Clearing it is one button, but a header set by the host has to be removed by the host, and the component says that rather than pretending it succeeded.

Where to go next

Tick only the package row, the one whose type reads Package. The component and the plugin are its children, and Joomla refuses them on their own with a message saying they belong to a package. Selecting all three still removes everything, but it reports two errors on the way, which reads worse than it is.

The FindSignal for Joomla page has the requirements, the install steps and the changelog. The same plugin exists for WordPress, and the WordPress article covers the file itself in more depth. If you want to test a site before installing anything, the robots.txt tester and the sitemap checker run in the browser.

Summarize this guide with

You need an account with the assistant you choose. It fetches the guide itself, and you send the prompt, not us.

Every guide in this category is on the FindSignal page.

Something here wrong, or out of date? Tell us at editorial@enterrahost.com.