All docs
AI Discovery
An AI assistant that wants to understand a site quickly does not want your menus, your cookie banner or your scripts. It wants the words. AI Discovery gives it the words at addresses it knows to look for, and tells the search engines that listen the moment you publish or change an article.
It is the first screen under AI Visibility in the sidebar, and it has four numbered sections.
Be clear about what this does and does not buy. llms.txt is a young convention. Google has said publicly that it does not use it, and no search engine or major AI assistant has said that it reads it. It costs one switch and may help an assistant find the right page. It will not move you up any list of results.
llms.txt Site guide

Enable /llms.txt is off when you install, because it is a new public address. Switch it on and your site answers /llms.txt with a short plain description of the site and a list of its main pages. The boxes under it appear once either file in the first two sections is switched on.
Site Description for AI opens the file, under your business name. Write it as a flat statement of what the site is and who it is for, not a strapline. The hotel's reads "Aldwick Lane Hotel is a 24-room boutique hotel in London. This site lists our rooms, prices, reception hours, events and news." Leave it empty and the file opens with the business description you wrote on Structured Data instead. The business name comes from Structured Data too, and falls back to the site name Joomla holds. Both files open with the same name and description.
The site guide lists your home page and your menu pages. Recent articles adds your most recently changed articles, and you choose how many, None, 5, 10, 25 or 50. It starts at 10. A site that publishes rarely is served as well by 5 as by 50, and None drops articles from the guide altogether.
Include the business details and social profiles is on when you install. It adds what kind of business this is, its country, telephone and email, and the links to your social profiles, all taken from Structured Data, so you never type them twice. It adds them to both files. Switch it off if those details say nothing useful about you.
Use the XML Sitemap's choices is on when you install. It keeps the site guide, llms-full.txt and the plain-text copies to the pages your sitemap lists, so switching off Menu pages on XML Sitemap takes them out of the guide too. Switched off, they take every kind of page, whatever the XML Sitemap's switches say. Either way, a page you answered Not listed on Pages stays out.
The hotel's file, shortened:
# Aldwick Lane Hotel
> Aldwick Lane Hotel is a 24-room boutique hotel in London. This site lists our rooms, prices, reception hours, events and news.
> Base URL: https://example.com
## Pages
- [Aldwick Lane Hotel](https://example.com/)
- [Contact Us](https://example.com/contact-us)
## Recent content
- [First Blog Post](https://example.com/park-blog/first-blog-post.md)
## About
- Type: Hotel
- Country: GB
- Telephone: +44 20 7946 0321FAQs are not in this file. They are published once, in your structured data, from Structured Data, sub-tab FAQs and Events.
llms-full.txt Full index
Enable /llms-full.txt, in the same picture above, serves a second file. It works only on a site that has been activated, as Getting started explains. It lists every page the AI files draw on, grouped under the headings Pages, Articles, Categories, Sections and Tags, with the full text of each article, category and tag page under its link. It helps on a large reference site, where an assistant needs to look up one specific page. A small site does not need it.
Most articles to index caps how many articles it holds. Empty or 0 means 500, and the most is 5,000. An article saved for All languages takes one place for each language it is listed in.
AI Boost does not read every page at once. A large site fills the file in over its first few visits, as AI Boost reads each page. After that, a copy is refreshed when its page's date changes, and again after an AI Boost update or a settings change. A copy also expires on its own, after seven days, or after one day for a page that carries no date. So a second edit on the same day, or a new article on a category page, can wait up to seven days.
IndexNow URL Submission

When you publish an article, a search engine finds out on its next visit, which may be days away. IndexNow tells the engines that share it, Bing, Yandex, Seznam and others, the moment you publish or change an article. Google does not take part.
- Switch on Enable IndexNow.
- Press Generate beside API Key. AI Boost fills in a random key. To use a key of your own, type eight to 128 characters, each a letter, a number or a dash. Your site starts answering it at
/<your key>.txt, which is how an engine knows the submissions come from you. - Switch on Auto-submit URLs on article publish / update.
From then on, every article you save, publish or unpublish has its address sent at that moment. With the key empty, or auto-submit off, nothing is sent.
The line under the key says when the last addresses went out, how many, and what the engine answered. On the hotel's site it says one address went out on 20 October and the engine answered 200. It reads what AI Boost wrote in the error log on Debug, so with that log switched off the line asks you to turn it on.
It is off when you install because each address it sends leaves your site for a third party, so switch it on only if you are happy with that. It asks for a visit and does not guarantee indexing.
Plain-text copies
Switch on Offer a plain-text copy of each article, category and tag page, in the same picture, and each article, category and tag page the AI files use gets a second address, its own with .md on the end. The copy holds the same words with the design, menus and scripts taken out, and your structured data comes along as code blocks. Like the full index, the copies work only on a site that has been activated. The home page and a menu page that shows none of those get no copy. With Use the XML Sitemap's choices on, tag pages get no copy until the sitemap lists them.
The hotel's jazz article is at /park-blog/first-blog-post, so its copy is at /park-blog/first-blog-post.md. Only the .md address gets the copy. A query parameter or an Accept header does not, so a cache in front of your site never hands a visitor the plain copy by mistake. Joomla's own page cache never stores the copy in place of the page either.
Each copy carries a noindex for search engines, which keeps the jazz article's two addresses from competing. The normal page stays indexable and stays in your sitemap. It carries one line in its head naming its copy, so an assistant finds it without guessing. Shortened, it reads:
<link rel="alternate" type="text/markdown" href="https://example.com/park-blog/first-blog-post.md">If you point the page elsewhere with the Canonical URL Map on Page SEO, the line still names this page's own copy. The copy carries no such line. From AI Boost 1.1.4 it carries a Link header instead, naming the canonical address its page writes, when the page writes one, so a search engine that reads the copy knows which address is the real one. With the Canonical URL Map pointing the page elsewhere, the header names that other address.
With this switch on, the site guide links the copy of each article, category and tag page instead of the page itself. With it off, it links the page.
Check it on your site
Open /llms.txt on your domain, and /llms-full.txt if you switched it on. If you switched on Offer a plain-text copy of each article, category and tag page, add .md to the address of any article. Site Health reads /llms.txt, and checks that an article carries the link to its copy.