AI Music Detector Accuracy
In October 2026 we ran 252 songs through the same check this site uses. It flagged every AI song and none of the songs made by people. Here are the scores, the songs that came closest, and the things this test can't tell you.
The short version
| Songs | How many | Flagged as AI | Closest to the line |
|---|---|---|---|
| New Suno songs | 64 | 64 of 64 | lowest score 92 |
| Viral AI songs | 3 | 3 of 3 | all scored 100 |
| Songs made by people | 185 | 0 of 185 | highest score 6 |
Scores run from 0 to 100. We count a song as flagged at 50 or more. The point to take away is the gap: the lowest AI song scored 92, the highest human song scored 6. Nothing landed in between.
What we tested
AI songs. 64 songs from Suno's trending list in October 2026, made with its newest model, in six languages. Plus three AI songs that made the news: Breaking Rust's Walk My Walk and two songs by The Velvet Sundown.
Songs made by people. 185 songs, picked to be hard on a detector rather than easy:
- Current pop and hip-hop: Taylor Swift, Billie Eilish, Kendrick Lamar, The Weeknd, Drake
- Heavy auto-tune: T-Pain, Cher's Believe, Travis Scott, Kanye West's Heartless
- Synthetic voices that are still made by people: Hatsune Miku and Kagamine Rin/Len (Vocaloid)
- Electronic music: Daft Punk, Skrillex
- K-pop, J-pop and other languages: BTS, NewJeans, Kenshi Yonezu, Aya Nakamura, Helene Fischer, Peso Pluma
- Classical and jazz: Yo-Yo Ma, Lang Lang, Beethoven under Karajan, John Coltrane
- Old recordings: Chuck Berry, Elvis Presley, Billie Holiday, Frank Sinatra, Ella Fitzgerald
How we tested
We used the original audio of each song and ran the same steps your browser runs here: mix to mono, convert to 16 kHz, measure the spectrum, pull out the fingerprint, score it. Same model, same settings, no tuning for this test.
Then we turned the 64 Suno songs into MP3 and into Opus (the format many streaming and voice apps use) and checked them again. All 64 were still flagged. The lowest score dropped a little, to 85 for MP3 and 87 for Opus. So a compressed file is fine.
The closest calls
These are the songs that came nearest to the line from each side.
| Song | Made by | AI score |
|---|---|---|
| Lowest of 64 Suno songsSuno (newest model) | AI, original file | 92 |
| Lowest of 64, after MP3Suno (newest model) | AI, as MP3 | 85 |
| SummertimeElla Fitzgerald | People | 6 |
| You Raise Me UpCeltic Woman | People | 6 |
| Don't Know WhyNorah Jones | People | 4 |
| Giant StepsJohn Coltrane | People | 4 |
The human songs that scored highest were mostly old or acoustic recordings. Even those stayed far below 50. That's why a song in the middle of the scale is worth a second look: our test didn't produce any.
What this test doesn't show
- Other generators. We tested Suno. The model's authors say it also catches older Udio models. We haven't tested newer Udio, ElevenLabs Music, MiniMax or Mureka.
- Songs cleaned on purpose. Some services sell processing that strips the traces detectors look for, so AI songs pass checks at distributors. Expect those to slip through.
- Half-AI songs. A real band over AI vocals, or a human singer over an AI beat. We didn't test these, and the score will likely sit in the middle.
- Rough copies. A song recorded off a speaker or ripped at low quality. We only tested clean files.
- Size. 252 songs is enough to see the pattern, not enough to put a precise error rate on it.
Other AI music detectors
You don't have to take our word for it. Here's what else is out there, as described on each tool's own site in October 2026. We haven't tested these, so we don't rank them.
- SubmitHub AI Song Checker: free, part of a music promotion platform.
- Detect.Music: free, says it's about 99% accurate.
- AI Music Checker (aimusicchecker.org): free, no sign-up.
- Deezer: labels albums it detects as AI in its app. You can't upload your own file.
- IRCAM Amplify, Pex, ACRCloud, Hive: detection APIs for labels and platforms, not tools for one song.
What's different here: the song never leaves your device, and the test above is public, close calls included. If a result matters, run the same file through two detectors.
Run the check yourself
The AI music detector takes any audio file and gives you the same 0–100 score, free for one song at a time. If you can't get the file, our guide on how to tell if a song is AI covers the artist checks that work without one.
The detector is the open-source ai-music-detector by Matěj Štágl (MIT License). We didn't build the model; we tested it before putting it on this site.
FAQ
Questions
- How accurate are AI music detectors?
- It depends on the detector and on which generator made the song. Ours flagged all 64 recent Suno songs and none of 185 human songs in our test, but we mostly tested Suno. Any detector is only as good as the generators it has seen.
- Can an AI music detector be wrong?
- Yes. A generator it hasn't learned can slip through, and so can a track that was processed to strip the fingerprint. Some sites sell exactly that to people uploading AI songs to streaming services. A low score lowers the odds; it doesn't clear a song.
- Which AI music detector is the most accurate?
- Nobody has tested them all on the same songs in public, us included. Detectors that report a number like 99% rarely say what they tested on. If it matters, run the same file through two detectors and look at the artist too.
- Does auto-tune make a song look like AI?
- Not in our test. T-Pain, Cher's Believe and Travis Scott all scored 0. The detector looks for traces of how a generator builds sound, not for a voice that sounds processed.