Myth check: "AI mastering is now as good as a human mastering engineer"

The claim: AI mastering is now as good as a human mastering engineer, so paying a person is a waste.

Verdict: misleading. AI mastering is fast and often good on a healthy mix. The public evidence does not show it matching a skilled engineer, and even the tool makers don't claim it does.

Where it spread

The claim grows out of marketing and convenience. LANDR's mastering page says its AI "is trained on the adjustments human mastering engineers make", that it is "the only online mastering service that top audio engineers and major labels trust", and that it is "trusted by Grammy Award winners and over 5 million musicians globally". Those are the vendor's own claims, with no independent test cited on the page. From there it is a short hop to "as good as an engineer".

What the evidence says

The blind test. In 2024 Benn Jordan gave a mixdown of his track Starlight to AI plugins, online services and human engineers, then had 472 people rate the randomised results double-blind, MusicTech reported. Two human engineers, Max Hosinger and Ed the Soundman, took first and second. The open-source Matchering 2.0 came third, ahead of Ozone setups. LANDR was disqualified before the final.

The reviewer. Sound On Sound's Ozone 12 Advanced review (John Walden, December 2025) calls Master Assistant "very effective" on "material in decent health", but says it "perhaps won't rescue a poor mix to the extent an experienced mastering engineer might be able to".

The maker. iZotope's Master Assistant guide says plainly: "nothing replaces a dedicated mastering engineer", pointing to spec compliance, metadata and sequencing. It also explains that the assistant matches your track's tone, width and dynamics to targets built from recent chart hits, or to a reference you upload. For more on that mechanism, see how AI mastering works.

What we think

The honest version of the claim is narrower: on a good mix, an AI master is often good enough for a single, and far cheaper. That is a real change, and it is why we'd still point most first-time releasers to the tools in our best AI mastering services guide.

But "as good as a human" is the wrong yardstick. A matcher aims at a target; an engineer decides what the target should be, hears what the mix got wrong, and handles an album as a sequence. One blind test on one track proves nothing forever, and the tools have moved on since Ozone 11. Still, it is the best public evidence we have, and it went the humans' way. Our honest look at LANDR, Ozone and an engineer puts prices next to that trade-off. Use AI for drafts and quick singles; pay a person when the record matters or the mix is in trouble.