← Back
EmergingIndie 5 min read

AI Mastering: When It Works, When It Kills the Song

A sober guide to LANDR, iZotope, and the human mastering engineer — and the songs where the answer changes.

by · · 5 min read release-strategy

AI mastering is one of the genuinely useful tools the last few years have produced. It's also one of the easiest places to quietly hurt a record you spent six months on. The right answer isn't "always use it" or "never use it" — it's a per-song decision, made with eyes open about what the algorithm preserves and what it flattens.

The decision, song by song

The question is never "is AI mastering good?" It's "is AI mastering good for this song, on this release plan." The honest answer changes track by track.

Use AI mastering

Cheap, fast, fine

  • Loosies, alt-versions, instrumentals, sped-up edits
  • Loud, dense genres — drill, hyperpop, EDM, trap
  • Demo masters for pitching and playlist consideration
  • Catalog refreshes where the original master is the reference
  • Anything you need turned around in under 24 hours
vs

Send to a human

Pay the $200

  • Singles you're actively promoting and pitching
  • Dynamic genres — singer-songwriter, jazz, ambient, classical-leaning
  • EPs and albums where cohesion across tracks matters
  • Mixes with unusual character or stereo image you want preserved
  • Anything that has to sound right on a car system or in a venue
Per-song decision. If the left column matches your release, ship the algorithm. If the right column matches, pay the engineer.

AI mastering is fine for the songs you're shipping. Pay the human for the songs you're betting on.

What the algorithm flattens

LANDR, iZotope's auto-master, and the Bandlab equivalent all do the same broad thing — they pull every song toward a target average. On dense music with little dynamic range, the target is close to where the song already lives, and the result is fine. On music with quiet verses and loud choruses, the target compresses the gap, and the song stops breathing. The chorus that should hit twice as hard now hits 1.3x as hard. The whisper in the second verse is no longer a whisper.

A human engineer makes a hundred small calls the algorithm can't — leaving the verse quieter on purpose, tucking a sibilant frequency that's distracting on phone speakers, widening the stereo image in the bridge so it lifts when the second chorus arrives. None of these are dramatic. All of them are why the record feels different.

Budget proportionally to the mix

A useful rule of thumb: spend on mastering proportional to what you spent on the mix. A $40 algorithmic master on a $4,000 hand-mixed record is the audio equivalent of putting cheap tires on a sports car. A $200 master on a bedroom mix is overspending in the wrong direction — fix the mix first.

The decision is rarely about the algorithm itself. It's about respecting where the song is in its life and what the next twelve months ask of it.

Frequently asked

What does AI mastering actually do?
It analyzes your track, picks a reference profile, and applies EQ, compression, limiting, and stereo widening to hit a target loudness and tonal balance. LANDR, iZotope Ozone's auto-master, and Bandlab's mastering all work this way. The result is technically competent — loud enough for streaming, tonally balanced — but every track gets pulled toward the same average, which is where dynamic music suffers.
When is a human mastering engineer worth $150–$400?
When the song has wide dynamic range, when the song is a single you're actually promoting, when the mix has unusual character you want preserved, or when you're putting together a body of work that needs sonic consistency across tracks. A skilled engineer also catches mix issues the AI quietly compresses past — a kick that's clashing with the bass, a sibilant lead vocal, a stereo image that's too narrow for the chorus.
Can I A/B test AI vs human master myself?
Yes, and you should — at least once. Send the same mix to LANDR and to a human engineer, listen on three systems (phone speaker, decent headphones, car), and pay attention to where the energy sits in the chorus and whether the quiet moments still feel quiet. If you can't hear the difference, ship the AI master next time. If you can, you've learned something important about your ears and your songs.
What about AI mastering plus a human for the singles only?
That's the working answer for most indie artists. Use AI mastering for everything except the two or three singles you're actively promoting — and on those, pay for the human. It controls the cost while protecting the songs that have to land.