This whole thing seems to me like it isn't any AI by the common understanding, but just a few hardcoded scores and criteria.
Now, thats probably wrong, and the AI is super neat, but I think it doesn't look very good when it does that.
I dont think the chances are high that so many different websites all score exactly 72%. Not 71, not 73, 72.
Its weird. Now the whole site looks nice and all, I saw the papers referenced, too.
But again, when I read that it's "97% accurate", I wonder how that is even measured? I understand that there are measures of how easy to use a site is, etc. So I wonder what cases those 3% were, where your AI was "wrong"? Did you hand-review them, or what?