Can this evidence prove a track was AI-generated?

No. It can support a review, but origin and authorship require provenance and context. Do not use one automated or acoustic observation as proof.

What should I do when the evidence conflicts?

Preserve the disagreement, inspect file quality and production context, and request documentation if the decision matters. Inconclusive is a valid outcome.

Guide / Field notes

How to Check a Song for AI: An Evidence-Led Workflow

Start with a decision, not a tell.

3 August 2026 · 9 min read

Choose the standard before you inspect

A playlist choice and a contract dispute do not require the same evidence. Write down the decision first. For low-stakes curiosity, a listening pass and automated report may be enough. For a consequential action, plan to collect production records and speak to the creator.

Climb an evidence ladder

Begin with the best available file. Note audible anomalies before opening a detector, then read the complete report rather than one percentage. Move next to metadata and release context. Finish with provenance: dated drafts, stems, project files, contributors and a coherent account of the session.

Know when to stop

An inconclusive result is a stopping point, not an invitation to shop for a harsher tool. Compression, mastering and hybrid production can all confuse classification. If the evidence cannot support the decision you planned to make, record that uncertainty and do not escalate the claim.

Key takeaway

The short version

Detection is a triage process: define the stakes, review independent evidence, and let provenance carry more weight than an acoustic estimate.

Run a free detection

Frequently asked questions

No. It can support a review, but origin and authorship require provenance and context. Do not use one automated or acoustic observation as proof.