Sure, but how do you build a rich search feature if you are unable to extract features from the private content?

The idea the OP put forward is a database that allows "blind" searches. So I could, for example, search for articles between dates X and Y without revealing X or Y or the content or actual date of any article. The results of the search would be encrypted.

There is disagreement in the cryptography community about the extent to which this is actually possible. I've seen a few proofs that the obvious "magic" version of this where you can directly but blindly search is impossible because you could use search criteria to reconstruct the content, but that only rules out the most obvious implementation strategies. There's a lot of research going on with this kind of thing in the cryptographic community that involves leveraging zero knowledge proofs, homomorphic encryption, and so on. Nothing practical enough to make anything like a blind SQL database though.

In any case I was arguing against the claim that this would preclude good UI/UX design if it were in fact possible.

Most search in modern projects isn’t things like date ranges though it’s often things like machine learnt mappings that identify similarity scores with words. Failing that it’s some sort of ML to learn tags and keyword matching on those tags.

When I search YouTube I’m seldom looking for all videos made on August 29th. I might instead be looking for videos of a black cat. Or videos containing music with certain lyrics that I search for. This is what I thought you meant by rich search. I contend it’s unlikely you can design a privacy respecting service that can also extract the features necessary to train a search model.