How do you let an AI answer say "it depends" without the product feeling broken?

I'm launching Elif on Sep 21 - an ad-free app for prayer times, the Qur'an, qibla and dhikr. The part I keep rewriting is the assistant, and the problem is not the one I expected.

Religious questions have a property most product questions don't: for a large share of them the honest answer is "it depends which school of law you follow", and the four mainstream Sunni schools genuinely differ on real, everyday things. A model tuned to be helpful will flatten that into one confident answer. That is the failure mode I actually worry about - not hallucination exactly, but false confidence in a domain where people act on the answer immediately.

Where I landed so far:

Answers are grounded in a vetted corpus and surface the single source they lean on, rather than a wall of citations nobody reads.

You pick your school in settings, so the answer is given inside that frame instead of listing all four every time.

The UI states plainly that it is not a fatwa authority.

What I'm still unhappy with: when the schools genuinely disagree, saying so is the correct answer, but it reads as evasive and people bounce. Hiding it is worse. I haven't found a phrasing that is both honest and satisfying to read.

If you've built anything where "I'm not certain" is a legitimate output - medical, legal, tax, compliance, safety - how do you present uncertainty without making the product feel broken? I'd take concrete UI patterns over theory. Genuinely stuck on this one.