The AI mastering services that built themselves on streaming masters through 2022 and 2023 have spent the last two years pitching a vinyl pre-master service. By 2026 several of them have a credible product, and a few of them have a product that is best described as a vinyl-shaped streaming master. The distinction matters more than the marketing implies.

A vinyl pre-master is a different brief from a streaming master. The cutting engineer needs a file that has been prepared for the physical limits of a lacquer cut: bass that is mono below a sensible crossover, sibilance that does not chase the stylus off the groove, a side length that does not push the cut into a thinner area of the disc than the music can survive. A streaming master pushed to 0 will sound aggressive on streaming, then crackle and skip on vinyl.

The AI services that are producing usable vinyl pre-masters in 2026 are doing three things:

They are asking for the intended side length at upload. A 28-minute vinyl side is a different cut from a 16-minute side, and the AI service that ignores this is producing a master that will not cut cleanly on the long sides.

They are handling the bass mono question explicitly. The good services let you set a mono-bass crossover and apply it as part of the vinyl-pre-master path. The not-good services apply a generic streaming low end and call it done.

They are tightening sibilance with a de-essing pass that is calibrated for vinyl rather than streaming. A streaming master can survive a brighter top end. A vinyl cut on a brighter top end will eat the inner-groove tracking on side endings.

The places AI mastering for vinyl is still weak in 2026:

Loudness judgment for the whole record. The decision of how loud a record should sit relative to itself across sides is still a mastering engineer’s craft, and the AI services do not have the context. The record cut in three independent AI passes will sound like three records, not one record.

Reference fidelity. The strong human mastering engineers cut to references the artist brings in, and they understand why the reference sounds the way it does. The AI services match to a generic reference profile, and the result is competent but anonymous.

Communication with the cutting engineer. A human master gets sent with notes — “the vocal pulled up a dB on track 3 around 1:40, the bass is hot at the head of track 5” — and the cutting engineer adjusts. The AI master goes into the cutting room without context.

For an Australian independent label putting a record on vinyl in 2026, the honest read is that AI mastering is a credible option for budget-constrained projects, and a poor option for records where the artistic intent is specific. The records that work are the ones where the brief is simple, the references are conventional, and the budget for a human master is genuinely absent. The records that do not work are the ones where the production is unusual and the artist wanted the master to do specific work.

The other read is that AI mastering is changing the work of human mastering engineers. The good engineers in 2026 are spending more time on the records where their judgment matters and less time on the records where the brief was generic. That is probably healthy for the craft, even if the pricing pressure is uncomfortable.