The short answer: normalisation
Most listening platforms measure how loud your episode sounds on average and adjust playback to match their own internal target, in real time, on the listener's device. The file you upload isn't exactly what people hear; it's the input to a process each platform runs independently, with its own target level.
Different platforms, different behaviour
It helps to separate three things: what a platform officially requires, what it recommends, and what people have simply observed it doing. Very little of this is a requirement.
Apple Podcasts is the most explicit: its creator documentation recommends an overall loudness of about -16 dB LKFS, within roughly a decibel, and true peaks no higher than -1 dBFS (LKFS and LUFS are the same measurement). Around -14 LUFS is widely used as a practical working target for video and music platforms, but that figure is common practice rather than something YouTube publishes as a required upload specification.
Practically, anywhere between about -16 and -14 LUFS with a decibel of true peak headroom behaves well everywhere. Picking one consistent level for all your output is easier than maintaining a different master per platform, and normalisation absorbs the small difference. The detail is in podcast loudness standards explained.
What normalisation can't fix
This is the part people miss: normalisation adjusts your whole file up or down as one block. It has no way to know that your co-host was speaking quietly in one segment; it can only turn the entire episode down to hit its target, which makes an already-quiet voice quieter still relative to everything else. Uneven levels within an episode are a mixing problem, not something loudness normalisation will ever solve for you.
What's actually in your control
- Balance speakers against each other before you master, not after.
- Master to a sensible target rather than the loudest you can push it.
- Leave true peak headroom so normalisation and encoding don't introduce distortion.
- Keep it consistent episode to episode, since that's what listeners actually notice.
The loudness war was never worth fighting
Because every major platform normalises, mastering hotter than the target gains you nothing. The platform just turns it back down, and all that's left is the squashed dynamics heavy limiting introduced along the way. Aim for a sensible, well-known target and keep it steady; that's the whole game.
One consistent target, handled for you
Optivox levels speakers first, then masters to a broadcast-standard loudness target with peak headroom intact, so the file behaves predictably no matter which platform's normalisation touches it next.