How do you make users trust an AI feature when money is involved?

I'm building an AI layer over Indian stock market news — it

scores each headline for how much it's likely to actually move

a stock. Launching on PH soon.

The hard part hasn't been the AI. It's been trust.

If my model says a headline is "high impact" and the stock

doesn't move, I've cost someone credibility — maybe money. So

I've been going back and forth on how much to show:

• Show the score only? Clean, but feels like a black box.

• Show the reasoning? More honest, but nobody reads paragraphs

while the market is open.

• Show confidence levels? Users seem to either ignore them or

over-read them.

I've landed on short reasoning + a visible "this is information,

not advice" boundary, but I'm not convinced it's right.

For anyone who's shipped AI features in a high-stakes domain

(health, legal, finance) — how did you handle this? Did you

expose the model's uncertainty, or did that just make people

trust it less?