Around -14 LUFS integrated with true peak below -1 dBTP. What the numbers mean, and why ducking matters more than either of them.
There is a specific number you want your short-form audio to land on, it is expressed in LUFS, and almost every page explaining it is written for an audio engineer rather than a creator. Here is the short version.
Target roughly -14 LUFS integrated loudness for short-form delivery, with true peak held below -1 dBTP. That combination is the common convention across social platforms and streaming delivery, and it is what every export from ReelsAudio is normalised to.
LUFS is a perceived-loudness measurement that weights frequencies the way human hearing does, which is why it is used rather than plain RMS. -14 LUFS will not sound like -14 decibels; the two are not interchangeable.
Platforms apply their own normalisation at playback. If your master sits well below the target, the platform turns it up, which raises the noise floor in the mix and makes any low-quality voice more audible. If it sits too hot, the platform pulls it down and your own limiter has less headroom to work with.
So normalising correctly is not about being polite to the algorithm. It is about controlling what happens to your audio on someone else's device.
There is no public evidence that loudness affects reach or ranking. We are describing playback behaviour, not a distribution claim.
Integrated LUFS describes average perceived loudness across the file. True peak describes the highest instantaneous sample level. You need both, because a file can be quiet on average and still clip on a single transient, and that transient is what causes distortion when the platform pushes it back up.
-1 dBTP is the usual ceiling. Going higher leaves no room for lossy encoding, and lossy encoding can push peaks higher than the original, which is exactly when you hear clipping on playback.
A correctly normalised mix with an und ducked music bed still sounds amateur, because the music masks the voice at the frequencies the voice needs. The fix is sidechain ducking: drop the music by several dB under speech and bring it back between lines.
This is the single most audible difference between a narrated video that sounds professional and one that sounds assembled. It is a mixing decision, not a mastering one, and no loudness target fixes it.
Most digital audio workstations can measure integrated LUFS and true peak on a master. Any free meter will do; you are looking for two numbers, not a grade.
If you are exporting from ReelsAudio, the provenance record attached to each export lists the measured values, so you do not need a separate tool to verify them.
ReelsAudio renders faceless episodes with consistent voice and music, varied structure, and loudness normalised per platform.