Can you detect music made with Suno?

Detection and attribution are different claims

Detection asks whether a recording looks machine-generated. Attribution asks which system generated it. Attribution is a much stronger claim and needs much stronger evidence: a model evaluated per generator, per model version, with held-out data confirming it can tell one generator from another rather than merely separating generated from human.

We have run no such evaluation, so our tool never names a generator. Any tool that does name one should be asked, politely, to publish its per-generator confusion matrix.

Why generator attribution is unusually hard here

• Model versions move. A signature learned from one release can vanish with the next update, and users can generate from several versions concurrently.
• Output overlaps. Competing systems trained on broadly similar material converge on broadly similar production characteristics.
• Post-processing intervenes. Most published tracks are exported, mastered and encoded after generation, which erodes whatever signature existed.
• No public labelled corpus. There is no widely accepted, licence-clear dataset of version-tagged output to evaluate attribution against.

What our detector will tell you about such a track

If a track was generated and exported without heavy post-production, our measurements often lean toward the generated end — typically through uniform tonal balance across sections and a low spectral ceiling. That is reported as “Possibly” or “Likely AI-generated”, never as “made with Suno”. If the track has been remixed, re-mastered or re-recorded, expect an inconclusive result.

Stronger evidence than any detector

• Provenance metadata. Check the file’s tags and any Content Credentials manifest before doing anything acoustic. Cryptographic provenance beats statistics every time.
• Publication pattern. Volume, upload cadence, cover-art style and cross-genre consistency are frequently more diagnostic than the audio.
• The creator’s own account. Many people using generation tools say so openly. Asking is underrated.
• Platform disclosure. Several distributors and streaming services now require or surface AI-involvement declarations.

Our position

We will add generator attribution only after a documented, held-out evaluation exists and is published on the research page. Until then, claiming it would be the kind of unverifiable marketing this site was built to avoid.

You can still run the track through the free detector for a general estimate, and compare with the same question about Udio.