Should audio feedback be continuous or on-demand?

I'm building Cogniear - an audio-based AI interview coach that whispers guidance during meeting sessions.

Design dilemma: Should the AI:

A) Whisper continuously as it detects areas for improvement

B) Stay silent unless you press a button to request feedback

C) Wait for natural pauses, then whisper briefly

Option A feels more like a real coach, but might be distracting.

Option B gives control, but breaks the "invisible coach" feeling.

Option C is what we're testing now.

What would work better for you? Curious to hear from anyone who's done interviews lately or public speaking practice.

Thanks!