Or the ISPs database could simply be of IP addresses connected to, and the purchaser of that database could apply identification based on which other IPs were connected to around a similar time.
If it's your argument that mapping IPs to "websites visited" is not 100% accurate, then of course you're correct... So what we need to do is figure out how accurate it is. Because if the answer to that question is 95%, then the whole value proposition of DoH flys out of the window. Why massively centralise DNS to just make a tiny dent in the problem?
If the answer to that question is 5% rather than 95%, then I guess that would mean we've centralised the web so far already behind gatekeepers like Cloudflare+Google+AWS+Microsoft, that it doesn't really matter if we go and centralise DNS for web usage in the same way, as the web is already fucked.
$ telnet minjiexin.com 80
Trying 208.80.153.224...
Connected to minjiexin.com.
Escape character is '^]'.
GET / HTTP/1.0
HTTP/1.1 400
[...]
<!DOCTYPE html>
<html lang="en">
<meta charset="utf-8">
<title>Wikimedia Error</title>
[...]
Big mystery.
minjiexin.com resolves to the same IP. So, you see an HTTPS connection to 208.80.153.224, what's the user doing?