Your Customers Are Bringing an AI With Them
More and more, people are bringing an AI along with them.
Not instead of searching — alongside it. They open an assistant and ask what they should do, how two options compare, what they ought to be asking before they call anyone. By the time they contact a business, they often have a shortlist and a view.
That is a new relationship, and your website is now feeding it. The question is whether it is feeding it anything useful, because the assistant is a different kind of reader from the one your site was built for.
We shipped version 2.5.0 of the Rootz AI Discovery plugin this week. Here is what changed, and the more useful story of why.
The plugin was working. It just never showed you anything.
Since March the plugin has been downloaded 426 times and ended up with fewer than ten active installs. Zero support threads. Zero ratings. Nobody complained — it just went quiet.
That detail is the one worth paying attention to. Nobody files a support ticket about a plugin that is not obviously broken. They activate it, see nothing happen, and deactivate it.
The plugin’s whole job is machine-facing. It publishes structured facts at /.well-known/ai so an assistant can read who you are instead of inferring it from your page layout. Which means the human who installed it saw nothing at all. No new page, no confirmation, nothing in the admin menu.
It turned out we had already built the thing that fixes this — a 120-point AI Readiness Score with a plain-English explanation of every check. It was on a tab. Under Settings. Which is the one place a new user has no reason to look, and nothing after activation pointed there.
So this was not a missing feature. It was a routing failure, which is a much cheaper thing to fix:
- AI Discovery now has its own top-level menu, with your current grade shown in the sidebar
- Activating it takes you straight to your score and names the single highest-value thing to fix next
- Every unfinished check shows what it is worth in points, so the order to work in is obvious
A fresh install scores C, 57 out of 120. Three fixes on a single tab are worth 35 of those points. That is a considerably better first thirty seconds than a blank screen.
We scanned ourselves, and it was not flattering
Before shipping this we ran all 28 of our own services against our own standard.
We averaged 30.9 out of 120. Three A grades. Fifteen D’s. Seven F’s, four of which scored exactly zero.
Our political dataset scored 28. Our freight service scored 30. These are services carrying more structured data than almost anyone, failing our own test — not for want of data, but because they published it in their own shape rather than the shape the standard defines. A generic assistant could not find the parts it expected.
We fixed six of them. The fleet average is now 49.8, and the services the standard was built to showcase sit between 113 and 118.
We are publishing that number because a scoring tool whose author will not run it on himself is not worth much. You can check us — scan politics.rootz.global or discover.rootz.global yourself and you will get the same figure we do.
Also fixed in 2.5.0
- Updating the plugin no longer reports the old version at
/.well-known/ai. A cached signed manifest was surviving upgrades, so the plugin looked as though it had not updated. We caught this on our own lab site, which was reporting 2.3.2 while running 2.4.0. - A missing
vendor/directory no longer takes your whole site down. The signing libraries were required unconditionally at load, so an interrupted upload, a partial unzip, or a security scanner quarantining one file meant a white screen on every page. Signing is an optional feature; it now degrades instead of dying. - Tested up to WordPress 7.0.
- The Adnet tab is hidden unless advertising is switched on. The feature is unchanged — it is simply not what a new install should open on.
One thing we are not claiming
None of this gets you ranked higher tomorrow. Publishing structured data does not move you up a results page, and we would rather say so plainly than let you discover it.
What it does do: when an assistant reaches your site, it reads facts you wrote rather than inferring them from your theme’s markup, and it can attribute the answer to you.
It is early
We are learning this every day, and some of what we believe today will turn out to be wrong. But the direction is not in doubt. People are going to keep bringing an assistant with them, and the businesses that give it something real to work with will be the ones it can describe properly.
Everything here is open. The standard is CC-BY, the plugin is free, there is more than one independent implementation now, and you can implement it by hand without us if you would rather.