AI Mastering: What It Does Well, What It Misses, and When It Is Enough
An algorithm sets a level in minutes. It does not tell you the vocal is too low.
By Hanna Eng·Audio engineer, Abbey Road Institute Paris
Mastering is the last step before a track is released: you balance the spectrum, control the dynamics and bring the level to the right loudness, around -14 LUFS for Spotify and YouTube with a true peak under -1 dBTP. AI mastering makes those settings automatically from an analysis, often of a few seconds of the loudest section. It is fast and cheap, but it does not fix an unbalanced mix and it does not know your intent: it is an average setting, not a listen.
Online AI mastering made a step once reserved for studios available to everyone, in minutes and for a few euros. For a demo or a first release, that is real progress. But many artists then find their track sounds louder without sounding better, or that the problem that bothered them is still there. I mix and master music in Pro Tools, by ear and by measurement. Here is what mastering really is, what AI mastering does, what it cannot do, and how to choose without wasting budget.
AI mastering vs a mastering engineer: what changes
| Criterion | AI mastering | Mastering engineer |
|---|---|---|
| Turnaround | A few minutes | 2 to 4 days depending on the project |
| What gets analyzed | An excerpt (in Ozone 11, at least 8 seconds, ideally the loudest section) | The whole track, and the whole record |
| Reference | A genre target or a reference track | Your references, and a conversation about intent |
| Mix problem (vocal too low, boomy bass) | Reproduced, sometimes made worse | Flagged, and fixed in the mix if needed |
| EP or album consistency | Track by track | Levels and tone aligned across the project |
| Revisions | A new automatic render | Notes listened to and applied by hand |
| Streaming target | Often -14 LUFS, -1 dBTP in streaming mode | Set for the destination and genre, checked by measurement |
Source: iZotope Ozone 11 documentation (Master Assistant); Spotify loudness normalization (official).
What is mastering?
Mastering turns a finished mix into a file ready for release. It works on overall tonal balance, so the track is neither too dark nor too bright on any playback system. It controls dynamics with light compression and a limiter. And it brings the level to the intended loudness, measured in LUFS, without exceeding a safe true peak.
On an EP or album, mastering also aligns the tracks with each other: same perceived level, consistent tone, clean transitions. Mixing balances the elements inside a track; mastering places the track in the world. You do not fix a vocal in mastering, you fix it in the mix.
How does AI mastering work?
Online services such as LANDR, eMastered or BandLab Mastering, and assistants built into plug-ins such as iZotope Ozone's Master Assistant, follow the same logic: analyze the signal, compare it with a target, then automatically set EQ, compression, stereo width and limiter to get close to it.
iZotope's Ozone 11 documentation spells it out for Master Assistant: the assistant listens to an excerpt, at least 8 seconds, ideally the loudest section, usually the chorus. It then sets the maximizer to reach the loudness of a genre target or a reference track, with a -1 dB true-peak ceiling in streaming mode. It works, but it means the decision rests on a few seconds of your track, not the whole thing.
- Analysis of an excerpt, not the whole track, depending on the tool.
- Target: a genre average, or a reference track you provide.
- Automatic settings: EQ, dynamics, width, limiter.
- Result in minutes, a new render on each try.
What AI does well
On a mix that is already balanced, AI mastering often gives a clean result at the right level for streaming. For a demo to send to a label, a track for social media, a first release on a tight budget, it is an honest option.
It is also a good comparison tool. Running your mix through automatic mastering quickly shows whether the track can take being pushed, or whether something starts to distort or hiss: a useful hint before going back to the mix.
What AI misses
A mastering algorithm processes the stereo file it is given. It cannot lift a vocal buried in guitars without lifting the guitars too, or tame a boomy bass without touching the kick. It pulls the track toward a genre average: if your intent is to step away from that average, drier, softer, wider, it pulls you back.
- It reproduces the mix's problems, and can make them worse by raising the level.
- It decides from an excerpt: a quiet verse and a dense chorus are not each treated for what they are.
- It does not know your intent or references, only a target.
- It does not think about EP sequencing: each track is mastered alone.
- It does not tell you why: you do not learn what to change in the mix.
Louder is not better: streaming loudness
Many automatic masters aim loud. But Spotify, YouTube and Apple Music normalize: they turn down tracks that are too loud to match their target. A master crushed to -8 LUFS just gets turned down, having lost its dynamics and punch along the way.
Aiming for the main platform's target, with a true peak under -1 dBTP, leaves headroom for conversion to compressed formats. It is not an absolute rule: some genres own a denser master. But it should be an informed decision, not a default.
| Platform | Normalization loudness | Advised true peak |
|---|---|---|
| Spotify | -14 LUFS | -1 dBTP (-2 if the master is louder than -14 LUFS) |
| YouTube | -14 LUFS | -1 dBTP |
| Apple Music | -16 LUFS | -1 dBTP |
Source: Spotify, Loudness normalization (official). YouTube and Apple Music do not publish a figure: -14 and -16 LUFS are the measured, industry-accepted levels.
AI or an engineer: how to choose
The question is not AI versus human, it is what you are releasing and what you are willing to fix.
- Demo, sketch, social media track: AI mastering is often enough.
- A single you will promote, an EP, an album: get the mix heard by outside ears at least once.
- A mix you made yourself: that is where an engineer adds the most, by going back to the mix instead of covering it up in mastering.
- If you test AI, give it a reference track rather than a genre target, and compare at the same perceived level, not the loudest.
How I approach a master
I listen to the whole track before touching anything, then compare it with the references you like, level-matched. If a problem comes from the mix, I tell you rather than hide it under an EQ. Processing stays light: a tonal correction, a little dynamics, a limiter set for the destination, then a check by measurement, integrated loudness and true peak.
My mixing packages include mastering, from €45 to €130 depending on track count, with a mastered stereo mix ready for distribution. Mixing and mastering stay in the same hands, which avoids exactly the back-and-forth between a locked mix and a master trying to rescue it.
Frequently asked questions
What does mastering do?
It makes a mix ready for release: a tonal balance that holds on every speaker, controlled dynamics, a loudness suited to the destination and, on an EP or album, tracks that sit together.
Is AI mastering as good as a human engineer?
On an already balanced mix and for a demo, the result can be quite good. On a mix with problems, or a release you need to stand behind, no: AI does not fix the mix and aims at a genre average, not your intent.
What level should I master to for Spotify?
Spotify normalizes to -14 LUFS integrated. Aiming around that value with a true peak under -1 dBTP keeps your track from being turned down and preserves its dynamics. Spotify advises -2 dBTP if the master is louder than -14 LUFS.
What is the difference between mixing and mastering?
Mixing balances the elements inside the track: levels, EQ, effects, the place of each instrument. Mastering works on the whole stereo track to finish it and align it with the other songs.
Can I master my own music with Ozone?
Yes, and Master Assistant is a good starting point. Give it a reference track, check loudness and true peak by measurement, and always compare at the same perceived level before committing.
How much does mastering cost?
AI mastering costs a few euros per track or a subscription. With me, mastering is included in the mixing packages, from €45 to €130 depending on tracks and revisions.