Editorial standards

Who writes this

AI Music Detector is independently operated. Content is written and maintained by its operator rather than a rotating pool of contributors, so one person is accountable for every claim on it. Contact details, including the publisher identity disclosure, are on the contact page.

How claims are sourced

• Claims about our own detector are checked against the running code before publication. A single canonical description of how audio is handled lives in the codebase, and every page describing processing, privacy or capability is written from it, so pages can't quietly contradict each other.
• Claims about platform policy — streaming services, distributors, contest rules — come from the platform’s own published documentation and carry a date, because these things change often.
• Claims about detection science draw on published work in synthetic-audio and deepfake-audio detection, and are pitched at the confidence level that work actually supports. Where the field genuinely doesn't know something, the page says so instead of filling the gap.
• Claims about other products are limited to what those vendors state publicly. We don't publish comparative accuracy numbers for competitors, because we haven't measured them.

What we refuse to publish

• An accuracy percentage for our own detector. No controlled evaluation has been completed, so precision, recall and error rates remain unmeasured. See accuracy and limitations.
• Generator attribution. The system doesn't identify Suno, Udio or any other platform, and no page is allowed to imply otherwise.
• Forensic framing. A result is an automated estimate of resemblance, not proof of authorship, and never evidence for a copyright, employment or academic decision.
• Invented testimonials, ratings, user counts or credentials. Where a number hasn't been measured, no number is given.
• Claims that the detector runs locally. It doesn't; the file is uploaded for classification, and the privacy policy spells out exactly what that involves.

Review and maintenance

Every page carries a last-updated month. When the detection pipeline changes, the pages describing it are revised in the same change, and the changelog records what moved. Changelog entries describe behaviour, never accuracy improvements, because an accuracy improvement can only be claimed from a measured benchmark run.

Pages covering platform rules or generator capabilities are the most perishable content on the site and get re-checked against source documentation on every revision. If a page looks stale to you, the fastest fix is telling us which sentence.

Advertising and independence

The site carries advertising to cover the cost of running a free detector that requires no account. Advertising is bought programmatically; advertisers have no input into what's written, no advance look at it, and no ability to have content changed or removed. We publish no sponsored posts and no paid reviews. Where a page compares tools, the comparison sticks to publicly verifiable facts and states our own product’s limitations in the same terms as everyone else’s.

Corrections

If something here is wrong, we'd rather know about it. Send the page URL and the specific sentence through the contact page. Factual errors get corrected directly, and where a correction changes what a page means, the update month gets bumped so the change is visible. Requests to remove accurate, sourced information are declined.

Use of AI in producing this site

Drafting assistance is used, as it is almost everywhere now. That doesn't change accountability: every published page is read and edited by the operator, technical claims are checked against the code or the cited source, and anything that couldn't be verified is cut rather than softened. We hold ourselves to the same standard we ask of musicians disclosing AI use — see the disclosure guidance.