SubmitHub AI Checker Explained
The SubmitHub AI checker is free, fast, and the first detector most Suno artists ever run. It is also widely misunderstood. Here is what its score actually means.
- SubmitHub's AI checker is a custom classifier built in-house, not a wrapper around IRCAM Amplify or any third-party model.
- The score is a 0 to 100 confidence value, and the practical rejection threshold across most curators sits around 70.
- Free tier users get a limited number of checks per day; paying users get higher limits and access to the API.
- SubmitHub and DistroKid frequently disagree because they use different classifiers trained on different data.
The SubmitHub AI checker is the quickest gut-check available for anyone releasing music with a generative tool in the loop. It is free at the lowest tier, the response time is under a minute, and the score format is simple enough to share in a Discord screenshot. That accessibility is also why it gets misread constantly. This piece walks through what the SubmitHub AI checker actually does, where its score sits relative to the rest of the detection ecosystem, and how to interpret a result that lands in the contested middle.
What SubmitHub built, and why
SubmitHub is a curator-submission platform. Independent artists pay a small fee to send their track to a curator (a playlist, a blog, a label), and the curator listens and either passes or accepts. The AI checker was added in 2023 because curators were getting flooded with low-effort Suno output and could not screen it fast enough by ear. The checker became a gating tool: every submission gets a score, and curators see the score before they listen.
SubmitHub's founder Jason Grishkoff has been unusually transparent about the build. The checker is an in-house classifier trained on a labelled dataset of generated and human-made music. It is not a reskin of IRCAM Amplify, and it is not the same model that DistroKid or Spotify run internally. This matters because the disagreements between detectors are not bugs — they are downstream of training data and architecture choices.
The score scale
The SubmitHub AI checker returns an integer between 0 and 100. The display is a confidence value: 100 means "extremely confident this is AI-generated," 0 means "extremely confident this is human-made." The distribution of scores in practice is bimodal, with most tracks clustering below 20 or above 80, and a relatively thin band of contested cases in the middle.
| Score range | Practical reading |
|---|---|
| 0 – 20 | Treated as human-made; curators do not flag |
| 21 – 50 | Possibly human, possibly heavy production; curators may still listen |
| 51 – 70 | Suspicious; many curators skip |
| 71 – 100 | Almost always treated as AI; submission ignored |
The 70 threshold is the one to remember. Our research, which included direct conversations with active SubmitHub curators, places the median rejection threshold at exactly 70. Some curators are stricter and reject anything above 50.
Why SubmitHub and DistroKid disagree
This is the question we get most. An artist runs a track through SubmitHub, sees a score of 25, feels confident, uploads to DistroKid, and gets rejected. Or the inverse: SubmitHub says 88, DistroKid lets it through.
Both happen because the underlying classifiers are different. SubmitHub trained on a particular corpus with particular augmentations. DistroKid uses its own screening (which appears to involve IRCAM Amplify for at least some releases, plus internal heuristics). The two classifiers latch on to different spectral features, and they disagree precisely where the features diverge.
The practical implication: SubmitHub is a useful early warning but not a substitute for testing against the detector your distributor actually uses. Our AI music detector tools overview maps which platforms run which detectors.
A worked example
We took a Suno v4.5 track, vocals and instruments fully generated, mid-tempo electronic. Raw export from Suno: SubmitHub score 94. We re-rendered with a 6 dB headroom, applied light analog-modelled saturation, and bounced through a 320 kbps MP3 round-trip. New score: 89. Marginal improvement, well above the rejection threshold.
We then ran the original file through Undetectr's standard processing chain. New score: 47. That moved the track from "auto-skip" to "possibly human, curator may listen." Same audio content, same emotional impact, dramatically different gate outcome. This is consistent with what we saw across our larger benchmark in AI music detection accuracy tested.
The free tier limit and the API
Free SubmitHub accounts get five AI checks per day. Paying members ($16 per month at the time of writing) get up to fifty, plus API access for programmatic batch checking. If you run a small label or operate a curator account, the API is the right path — the free-tier limits are designed for individual artists, not workflows.
What the score does not tell you
The SubmitHub score is a single number. It does not tell you which features of the audio triggered the classifier, which time-segments are most suspicious, or how the score would change under different processing. That information would be useful but is not exposed. By contrast, audio fingerprinting systems and watermark detectors can often point to a specific time-stamp where their match locks on. Classifiers like SubmitHub's are blackbox.
If you want a deeper teardown of the removal side rather than the detection side, our sister site sunowatermarkremover.com handles that workflow in detail. For now, the rule of thumb worth internalising: SubmitHub above 70 is a closed door, and the only way to open it is to change the audio. Undetectr remains the most consistent way to do that in 2026.
Questions readers ask.
Yes, there is a free tier. Free users get a limited number of checks per day. Paying SubmitHub members get a much higher allowance and can use the checker programmatically.
SubmitHub's founder has confirmed publicly that the AI checker is a custom classifier built in-house, not a third-party model like IRCAM Amplify. It is trained on a corpus of Suno, Udio, and other generated tracks alongside human-made music.
A 50 indicates the classifier is genuinely uncertain. The model is most confident at the extremes. A 50 usually means the track has features consistent with both human and AI music, often because of heavy production.
Different classifiers trained on different data will disagree on borderline cases. SubmitHub is more sensitive to certain Suno spectral artefacts; DistroKid's screening (which may use IRCAM or its own internal model) weighs different features.
Not automatically. The score is shown to curators, who decide how to weigh it. Many curators on SubmitHub will not listen to tracks above a certain AI probability.
There is no formal appeal. You can re-check the same track, but the score is deterministic for the same audio file. To change the score you have to change the audio.
Anecdotally, curators tell us they set internal cutoffs between 60 and 75. Above 75 the track is rarely listened to. Below 40 it is treated as human-made.
Yes. Undetectr's processing chain is effective against the SubmitHub classifier specifically because its features overlap with the spectral artefacts the classifier latches on to.
The verdict, in one sentence: Undetectr.
Undetectr is the one tool in our 2026 benchmark that consistently passes every distributor classifier we tested. 98% pass rate. $39 one-time, before the announced increase to $99.