All docs
Site Health
A settings screen tells you what you asked for. It cannot tell you what your visitors receive. A caching extension can strip a tag after AI Boost writes it, another extension can serve its own sitemap at the same address, and your host can send a header that asks search engines to leave every page out. None of that shows on a settings screen.
Site Health reads your site from the outside. Your server fetches your own pages and files the way a visitor or a crawler would, and most rows report what came back. A row that reads Switched off or Nothing to check is read from your settings instead. When a fetched row and a settings screen disagree, the row describes your live site.
It is the second entry under Overview in AI Boost's sidebar. It has one numbered section, What your site is publishing, with a row for each of eight parts of AI Boost and five rows of its own. Check again and Copy report sit at the top right, beside Help.
The examples on this page come from a test site set up as a 24-room London hotel, the Aldwick Lane Hotel. The rows in the screenshots are that site's, faults included.
Running the check

Each check fetches four addresses on your site, the home page, /robots.txt, /sitemap.xml and /llms.txt. Depending on what you switched on and what those four addresses return, it also fetches:
- the picture your home page names for sharing, when Open Graph is on,
- one article, when plain-text copies are on and at least one article gets a copy, to look for the link to its copy,
- the news sitemap, when the Google News sitemap is on,
- up to five of the addresses
/sitemap.xmllists, spread across the list, to see that they answer, - every file your sitemap index lists, when AI Boost splits a large sitemap.
It waits up to six seconds for a page, a picture or a sampled address, and up to thirty seconds for a sitemap file or llms.txt, because those can be long. A whole check stops asking after 75 seconds. Most checks take a few seconds. While it runs, the section reads "Reading your site…" and Check again cannot be pressed. If the check cannot run at all, the section says "The check could not be run." with the reason under it.
Opening Site Health runs a check, unless one has run since you opened AI Boost and neither switch in the Publishing card has changed since. After that, the rows stay as they are until you press Check again here or Re-run now on the Dashboard. The line under the bold count says when the rows were read, such as "Read 6 November 2026 at 13:40:19. Nothing is re-checked until you press Check again." Old rows describe your site as it was then, so press Check again after you change something.
The top of the screen
The bold line counts the rows that need you, such as "2 problems, 2 worth a look." With none, it reads "Everything is fine."
The line on the right counts every row:
15 checks · 9 fine · 2 with nothing to check · 0 could not be checkedThe row of a part you switched off stays on the screen, and it is worth a glance. It says AI Boost does not write that thing. It does not look for anything else that might. The with nothing to check count takes in both the Switched off and the Nothing to check rows. The could not be checked count stands apart from the problems. AI Boost could not look, and that says nothing about whether the part works.
After a check, Site Health in the sidebar carries a small number, the Problem and Worth a look rows added together. On the test site it reads 4.
The word on each row
Every row carries one of six words.
| Word | What it means |
|---|---|
| Fine | What the part publishes was found on your live site |
| Worth a look | What was found is valid, but it may not be what you meant |
| Problem | AI Boost looked, and what the part publishes was missing or broken |
| Switched off | You switched the part off, so there is nothing to look for |
| Nothing to check | The part is switched on, but there is nothing to look for. Custom Code with no code saved, and Analytics & Verification with no ID entered, are two cases. While Publish nothing at all is on, every part row reads this, including the parts you switched off |
| Could not check | AI Boost could not read what the row needs, so the row makes no guess either way |
A part can hold several checks. Its row shows the most serious finding among them, Problem first, then Worth a look, then Could not check.
When a part is on but the switch inside it that publishes what the row reads is off, the row reads Nothing to check and names that switch with the label the part's own screen gives it. On Page SEO, with Write titles, descriptions and canonicals on and Enable canonical URL management off, the row names Enable canonical URL management.
Beside the word, a row has a sentence, often a second sentence and a box, on Social Sharing two links, and on Licence one. The first sentence reports what arrived. Most rows then say who wrote it:
- "AI Boost writes this, and nothing else on your page does."
- "AI Boost writes this, and something else on your site writes it too." On a row that would be Fine, this sentence takes the place of the report, the row reads Worth a look, and the box quotes the other version.
- "AI Boost serves this file." The sitemap or
llms.txtcarries AI Boost's own mark. - "AI Boost does not write this." A part row that reads Switched off says so, because anything of that kind on your site comes from somewhere else. Custom Code does not, and nor does Analytics & Verification while it still publishes your verification codes.
When llms.txt is missing, there is no mark to read. The row then says AI Boost writes the file, and that it cannot see from here whether another extension writes one too. A missing sitemap gets no such line.
When a sitemap at /sitemap.xml or a file at /llms.txt does not carry AI Boost's mark, the row's sentence reads "Another extension serves this file." and the row reads Worth a look. A Problem found by another check on the same part still comes first.
The box quotes the text the row read from your page or file, cut short when it is long, so you can judge it yourself. A row names the other extension only when what that extension wrote carries its own mark, as in "AI Boost writes this, and 4SEO writes it too." or "OSMap serves this file." An extension that is merely switched on is not named. Where it can, the box shows what that extension wrote.
The two rows about the whole site
Everything AI Boost publishes speaks for Publish nothing at all, the switch on the Dashboard in the Publishing card. It is Fine when that switch holds nothing back. It is Worth a look when the switch is on, and when no part of AI Boost is switched on at all. While the switch is on, redirect rules and the 404 log keep working, and your settings are kept.
The language your site announces reports the default content language Joomla holds, en-GB on the test site. AI Boost repeats it in two places in the structured data and, with Add og:locale tag on under Social Sharing, in the og:locale sharing tag. That tag works only on a site that has been activated, as Getting started explains. It has no way to tell whether that is the language you write in, so the row is Fine whenever it can read the language, and Could not check when it cannot. On a site that publishes more than one content language, the row says so, and each page's tags name the language of that page. Read the tag yourself, and if it is wrong, change the default in Joomla's own Language Manager, under Installed, Site.
One row per part

Each of the next eight rows carries the name of the screen it checks. Fix what a row reports on that screen, then press Check again.
Page SEO reads the canonical link on your home page, once Enable canonical URL management is on. One link that points at the page itself is Fine. None, or more than one, is a Problem. A link to another address is Worth a look, because that is sometimes deliberate. This row does not read titles or descriptions. Pages, Check addresses, Google shows those.
Structured Data reads the structured data AI Boost wrote into your home page. It is a Problem when the block is not valid, when AI Boost wrote none, and when the type you chose under Schema Type is missing from it. When AI Boost wrote none, the usual reason is an empty Organisation Name. On a site that has not been activated, a Schema Type other than Organization and LocalBusiness is published as one of those two, and the row looks for that one. It is Worth a look when another extension's structured data on the same page is broken, or when something else publishes structured data too. On the test site the row names Joomla itself, through its Schema.org plugin and its breadcrumbs module, and the box quotes the block they wrote into the home page. Decide which extension should publish structured data about your pages, and switch off the other one's version.
When the other writer publishes a type AI Boost publishes too, on your home page or the one article the check reads, the row names the type, as in "AI Boost writes Organization on this page, and so does 4SEO." When nothing marks who wrote it, the sentence ends "and so does something else on your site." The page then describes one thing twice, and the two versions may not agree. The row reads Worth a look, and the box quotes the other version, which helps you find its source when the row cannot name it. Switch that type off where the other version comes from, then press Check again. The row shows one finding at a time, Problem first, so this line can wait behind another one until that is fixed.
Social Sharing counts AI Boost's Open Graph tags in the page and, with Enable Twitter Card meta tags on, its Twitter Card tags. It checks that the page names a sharing picture, and asks for that picture. It is a Problem when none of the Open Graph tags arrived, when Twitter Cards are on and none of those arrived, when no picture is named, and when the picture's address answers that it is not there. On the test site no Default OG Image is set and the home page has no picture of its own, so a link to it would be shared with no picture. Set one on Social Sharing, then press Check again. Meta Sharing Debugger and LinkedIn Post Inspector open those tools with your home page's address already filled in.
XML Sitemap reads /sitemap.xml and counts its addresses, across every file when the sitemap is split. It is Fine when the sitemap is served and none of the faults below turns up. It is a Problem when there is no sitemap, when the address answers with something that is not a sitemap, when the sitemap lists no addresses, when it lists raw Joomla addresses, whose query carries option=com_, when a file the sitemap index lists is missing, and when any of the addresses it tried answers that it is not there. The test site's sitemap lists 107 addresses.
AI Discovery reads /llms.txt, and with plain-text copies on, it looks in one article for the link to that article's copy. It reads the link, not the copy. Plain-text copies work only on a site that has been activated, so on a site that is not, no article is fetched and the row reads /llms.txt alone. It is a Problem when llms.txt is missing or empty, and when the article carries no link. When plain-text copies are on but no article gets one, there is no page to read the link on, and that check has nothing to look at.
Robots.txt reads /robots.txt. It is Fine when the file carries AI Boost's section and the crawler rules in it match your answers. It is a Problem when there is no file, when AI Boost's section is missing, and when a rule in the file differs from what you chose. With Write usage signals into robots.txt on and an answer set, it also checks that the usage signal line is there.
Cloudflare can add lines of its own to the robots.txt it serves, and those are the lines a crawler gets. When they block a crawler you allow, a second Robots.txt row appears, as in "AI Boost lets GPTBot, ClaudeBot in, but Cloudflare's part of robots.txt blocks them. Change this in Cloudflare." It reads Worth a look, and the box quotes the line that opens Cloudflare's part. A crawler that only Cloudflare blocks does not count as a rule that differs, so it raises no Problem on the first row. AI Boost cannot change those lines. Change the setting in Cloudflare, as Sites behind Cloudflare explains, then press Check again. If you did mean to keep those crawlers out, answer Block for them on Robots.txt, and the row goes.

Analytics & Verification looks for your Google Analytics, Tag Manager and Meta Pixel IDs in the home page. The row reads the page as your server sent it. It cannot see what a browser then does, so a consent tool or a script blocker can still hold the tag back after a Fine. It is Worth a look when a Google tag loader is on the page and your ID is not, which is normal when your tags are managed elsewhere. It is a Problem when neither your IDs nor a Google tag loader are there. With Measure visits on and no ID entered, it reads Nothing to check. An ID AI Boost cannot use, because its form is wrong, is a Problem when no other ID works. The row names that ID, says nothing is measured, and sends you back to its form on Analytics & Verification. Beside an ID that works, it is Worth a look instead. The row names the ID it cannot use, and checks the page for the others. With Measure visits off and a verification code entered, the row reads Switched off and says "Measuring visits is switched off. AI Boost publishes only your verification codes." The codes themselves are checked on the Search engine verification row.
An ID that loads twice on your home page, or on the one article the check reads, is Worth a look too, because a visit can then count twice. The row reads "G-A1B2C3D4E5 loads twice on this page, so a visit can count twice. Keep it in one place." The row shows one finding at a time, Problem first, so this line can wait behind another one until that is fixed. Look for the second copy in your template, in Custom Code and in your other extensions. Keep the ID on Analytics & Verification, take the other copy out, then press Check again. A tag that Tag Manager adds in the browser is not in the page your server sent, so this check cannot see a visit counted twice that way.
On a YOOtheme Pro site with the YOOtheme extension switched on and YOOtheme's consent banner set to Opt-in or Opt-out, each of those tags your home page serves with no consent category gets a row of its own, such as Google Analytics 4 is not linked to the consent banner. It reads Worth a look. The tag still loads, and the banner cannot hold it back until a visitor answers. Waiting for consent says how to link it.
Custom Code compares what you saved in each of the three boxes with the home page, treating every run of spaces and line breaks as a single space. It is a Problem when a snippet did not arrive as you saved it, and the usual cause is a caching or minifying extension that changes it on the way out. The test site has the part switched on with nothing saved, so the row reads Nothing to check.
The four rows that belong to no screen
Search engine verification looks in your home page for the Google and Bing codes you set under Prove the site is mine. It is Fine when every code is there, as in "Your Google and Bing code is on your home page." It is a Problem when one is not, and the row names the engine, as in "Your Bing code is not on your home page, so Bing cannot confirm this site is yours." A code of the wrong form is never written into the page, so it reads the same way. With no code set, the row reads Nothing to check, and with Prove the site is mine off, it reads Switched off. Fine means the code reached the page. Only Google and Bing can say whether they accepted it. Whatever the row reads, it links to Google Generative AI performance report and Bing AI Performance, so a site verified another way still gets them. The test site has both codes set, and the row is Fine.
One site address asks for the http and www forms of your home address and reads what each one answers. It is Fine when each form that exists moves to your site's address with a permanent redirect, as in "Every form of your address leads to https://www.example.com/." It is Worth a look when a form opens your site without moving, moves with a temporary redirect, or opens a different site, and the row names that form. A form opens a different site when its page title is not your home page's, and the row then says "http://example.com/ opens a different site, so visitors who type it do not reach yours. Ask your host to point it to https://www.example.com/." When the form is the http address of your own host and Joomla's Force HTTPS is not set for the whole site, the row says "Set Force HTTPS to Entire Site in Joomla's Global Configuration." Otherwise it asks you to ask your host for a 301 redirect. A form that does not exist is not counted. When no form leads to your address and one could not be read, the row reads Could not check. The row only reports, and writes no redirect. On the test site the www form opens a different site, so its row is Worth a look.
Home page visible to search engines reads the X-Robots-Tag header your server sends with the home page, and the robots tag in the page. A header that asks search engines to leave the page out is a Problem. AI Boost does not send that header with your home page, so it comes from your host or your .htaccess file. A robots tag in the page that does the same is Worth a look.
The test site's server sends this with its home page:
X-Robots-Tag: noindex, nofollowOn a test site that is what you want. On a live site, ask your host to remove it, or look for the line in .htaccess.
Settings emptied by another screen finds two settings where one cancels the other:
- Tag pages is on for your sitemap, and Keep tag pages out of search on Page SEO takes every tag page out, so no tag page reaches the sitemap.
- Enable Google News Sitemap is on, and the news sitemap's address answers with nothing, or with something that is not a sitemap. AI Boost writes that file only when at least one article is left in it, so look at the news window and at any category answered Not listed on Pages, Overview, which takes its articles out of this file too. When the address serves a sitemap AI Boost did not write, the row says "Another extension serves this file." and names the address.
Either is Worth a look. Nothing is broken, but one of the two settings does nothing, and you decide which one you meant. With both the news sitemap and tag pages off, as on the test site, the row reads Nothing to check. With Enable XML Sitemap off, it reads Switched off.
Publish nothing at all and Hide AI Boost comments in the page source used to sit on this screen, so the note at its foot, under these rows, points to where they are now. Changing either one there throws away the rows, and the next visit to Site Health reads your site again.
The Licence row
Licence is the one row that asks nothing of your live site. It looks at the licence AI Boost has stored, so it answers a different question from the rows above it, whether the features listed in Getting started are on here at all. On the test site, activated and current, it is Fine. On a copy you have not activated yet, or after we have switched a licence off, it names each of those features, because each one publishes nothing until the site is activated. An expired licence also lands on Worth a look, but nothing on the site stops. Only updates and support do, so it is a reminder to renew rather than a fault. Every Licence reading other than Fine carries a link to Plan & Updates, where you put it right.
Copy report
Copy report puts the whole reading on your clipboard as plain text, to paste into a message to support or to your developer. It is quicker than describing the screen, and it carries the exact wording of every row. The button reads Copied for two seconds once it has worked. For the test site the report reads like this, with every row but one left out:
AI Boost: Site Health
Read at: 6 November 2026 at 13:40:19
Summary: 15 checks, 9 fine, 2 with nothing to check, 0 could not be checked
2 problems, 2 worth a look.
[PROBLEM] Social Sharing
Your page names no sharing image, so a link to it will be shared without a picture.
AI Boost writes this, and nothing else on your page does.If the browser will not copy the report, which can happen when your administrator is served over plain HTTP, the screen shows it in a box headed "Your browser would not let us copy". Select the text there and copy it by hand.
What the check cannot see
The rows read your home page, one article at most, and a handful of files. A fault that shows only on another page does not reach this screen. For one address, open Pages, Check addresses, and enter it there.