One detector classified 38.5% of July releases as fully or partly AI-made

SubmitHub says its SH Labs detector analyzed more than one million tracks released in July 2026. It classified 23.2% as fully AI-generated and another 15.3% as containing generated audio that had been modified or processed by people.

Combined, that produces the widely repeated 38.5% figure. It is useful as a signal of scale, but it should not be presented as a census of all recorded music.

Detection is part of the claim

The categories were assigned by SH Labs, a commercial detection system operated by SubmitHub. The public result therefore measures what that detector classified inside the analyzed sample, not a ground-truth declaration from every artist.

MusicRadar reports an important conflict in the same dataset: some artists whose tracks were flagged said they had not used AI. That could reflect nondisclosure, false positives, disagreement about what counts as AI assistance, or some combination of the three.

The more durable takeaway is not the exact percentage. Distributors and stores are increasingly making policy decisions from automated classification, while creators and listeners may use different definitions of fully generated, assisted and processed music. Any platform publishing an AI share should disclose its sample, categories, detector and appeal process alongside the headline number.