Pass TuneCore's AI Check

TuneCore distributes to Spotify, Apple Music, Amazon, YouTube Music and dozens of other stores, and every one of those platforms now has its own stance on AI-generated audio. To protect its relationship with those partners — and to keep fraudulent, mass-generated uploads off its pipeline — TuneCore screens incoming tracks for signs of AI generation before they ever reach a store. It's a gatekeeping step, run automatically on upload, that most artists never see until a release gets held.

The screen isn't there to punish anyone for using AI tools. It exists because the streaming ecosystem is flooded with machine-generated content, some of it used to game royalty payouts, and distributors are on the hook for what they deliver. If your track sets off that screen, it lands in the same bucket — even if you wrote the topline, arranged it yourself and only used a generator for part of the production. The classifier doesn't know your intent; it only measures the audio.

TuneCore's AI screening is a statistical classifier, not a person listening for something that sounds "fake". It has been trained on large sets of human-made and machine-made audio and has learned the fine-grained features that separate the two. When your file goes through it, the system extracts those features and returns an AI-probability score — a confidence level, usually expressed as a percentage, that the audio was AI-generated.

What it's reading is the acoustic fingerprint of how the audio was built, not whether the song is good. Neural synthesis leaves consistent tells: spectral energy distributed a little too regularly across frequency bands, phase relationships that are unnaturally coherent or smeared, micro-timing and transients that are reconstructed rather than performed, and a noise-floor texture that's synthesized rather than captured by a mic. You can't hear any of it, but the classifier measures all of it at once and rolls it into a single number. Cross a platform-set threshold and you're flagged as high-risk.

This is also why the flag has almost nothing to do with quality. A polished, well-mixed AI track and a rough one can score equally high, because the signal the detector keys on lives in the synthesis, not the songwriting. You can see exactly where your own track sits on that probability scale with the free AI Checker before you do anything else.

Being flagged isn't a small inconvenience. Depending on the score and the release, the consequences stack up quickly:

• The release is held or rejected. Your delivery stops before it reaches stores, sometimes with a generic notice and no clear path to fix it.
• Your release date slips. A held track can blow a coordinated launch — a music-video drop, a playlist pitch, a press cycle — because the audio isn't live when everything else is.
• Revenue and eligibility are at risk. A track that never goes live earns nothing, and repeated flags can put your standing with the distributor or with downstream platforms under scrutiny.
• Manual review takes time. Appeals and re-reviews aren't instant, and there's no guarantee the outcome changes if the underlying audio hasn't.

The frustrating part is that none of this depends on whether your song is actually any good. It depends entirely on a number the screen assigned to your audio — which means the fix is to change that number, not to argue about the music.

The instinct after a rejection is to bounce the track again, convert it to a different format, normalise it, or run it through a master and re-submit. It almost never works, and it's worth understanding why. The fingerprint the screen scores on is baked into the samples themselves. When you re-export, you're repackaging the exact same synthesized waveform in a new container — the deep statistical features the classifier relies on are still right there, so the probability barely moves.

The same goes for the usual "tricks": adding a touch of noise, nudging the pitch, slapping a limiter on the master. Those change the surface of the audio, but not the specific characteristics the model measures — and some of them can even introduce new artifacts the same model reads as suspicious. To actually lower the score you have to process the signal to break up those statistical regularities while leaving the music intact. That's a targeted problem, which is exactly what the AI Cleaner is built for.

The order you do things in matters. The most reliable route from a flagged file to a release-ready one looks like this:

• Check. Run the file through the free AI Checker first to see your AI-probability score. If it already reads low, you may not need to clean at all — don't fix what isn't flagged.
• Clean. Send it through the AI Cleaner. Start from the highest-quality source you have — a lossless WAV or FLAC gives a better result than a low-bitrate MP3.
• Re-check. Run the cleaned file back through the checker and compare the before/after score. Most tracks drop well below the high-risk line here; a minority stay higher depending on how the source was generated.
• Master and finalise metadata. Do your EQ, compression and loudness after cleaning, then get your title, artist, ISRC and credits right so nothing else trips a manual review.
• Disclose where required. If TuneCore or a downstream store requires you to flag AI use, do it. Cleaning changes what a scanner measures; it doesn't change the rules you agreed to.

If you want to line up your key and tempo for the master, the free BPM & Key finder reads both straight from the file.

Cleaning a finished stereo mix works, but cleaning stems works better. When every element is baked into one file, the processing has to treat vocals, drums, bass and synths together. Split them out and each part can be handled on its own terms — which matters because artifacts aren't spread evenly. Vocals, in particular, often carry the strongest AI signature of any element, so isolating and cleaning them individually usually moves the score the most.

If you have stems, upload them as a ZIP and you'll get each one back cleaned, ready to re-balance into your own mix before you master and redeliver. If you only have the flat mix, cleaning that is still effective — it's just a coarser tool. Either way, always re-check the result so you have proof the score dropped before it goes back into TuneCore.

One honest note, because it matters. artefactFX removes the acoustic artifacts that an automated screen scores on — it does not remove any legal or platform obligation you have, and it is not a tool for passing off machine content as something it isn't. Where TuneCore, a store or a label requires you to disclose that a track uses AI, you should still disclose it, and you should keep your use of any generator within its own terms. This is about removing an inaudible synthesis signature that gets legitimate releases throttled, not about evading a policy or committing fraud.

Used that way, the tool does exactly what a working artist needs: it stops a statistical fingerprint from holding up a release you're entitled to put out, while you stay compliant with the platforms you publish on. If your worry is broader than one distributor, our guides to DistroKid AI detection, distributor AI checks and Spotify AI detection walk through the same path for those platforms, and you can compare plans anytime on pricing.

artefactFX was built by people shipping real releases, not a generic audio utility. Detection uses professional AI analysis, cleaning targets the hidden fingerprint without wrecking your sound, and every result comes with a before/after score so you are never guessing whether a resubmission will hold. Check for free, clean only when you need to, and redeliver with confidence.

It's also honest about its limits. We won't tell you every track will magically pass TuneCore's screen — most drop well below the high-risk line after cleaning, a minority stay higher depending on the source, and mastering afterwards lowers the risk further. You see the real numbers at every step, on your own files, with the free AI Checker and the AI Cleaner.