Questions & answers

What does Spot Slop do?

You give it a song — a streaming link or an audio file — and it tells you how likely it is that the recording was generated by AI rather than performed and produced by people. It also shows you why it reached that answer.

Is it free?

Yes. Detection is free and stays free. Paid features later will be extras — batch scanning, reports — never the answer itself.

Why do I need an account?

So your history is yours, and so we can tell one person's use from another's.

Every analysis costs us real computing time. Without accounts, anyone could point a script at us and use that capacity up. Asking for an account, and for a confirmed email address, is enough of a step to keep the service working for people.

Can I change my email address?

Yes, but you need to ask us first. Your email address is both how you sign in and how we send you a password reset, so we open the change one account at a time rather than leave it as a button anyone can press.

Write to support@spotslop.app from the address on the account and we will enable it — you then make the change yourself in Settings › My data.

Can I trust the verdict?

Treat it as evidence, not proof. Every result is a probability with a confidence range, and the honest answer is sometimes that we do not know. Do not use Spot Slop to accuse anyone. A wrong AI verdict on a real musician is the mistake we care most about avoiding, and no detector can rule it out entirely.

There are three verdicts. What do they mean?

These are three points on one scale, not three separate findings. Human — the signs of AI generation are weak (0–15%). AI — they are strong (85–100%). Uncertain — in between (15–85%), and we genuinely do not know.

That is why a "human" verdict can sit beside a percentage that is not zero: the number is how strong the signs are, and "human" means they are weak — not that there were none. Uncertain is a real answer, not a soft yes: the evidence does not support a confident claim in either direction.

It said uncertain for a song I know a person made. Why?

Heavily produced music is the hard case. Precise, grid-locked, densely arranged production — much electronic and modern pop — shares surface features with generated audio, and a 30-second excerpt gives the model less to work with. Uploading the full file often helps.

What are Staccato, Legato and the Ensemble?

They are our detectors, and the name under a verdict tells you which one answered.

Staccato listens to short passages spread across a track. Legato listens to one long stretch, so it hears how a piece develops — structure that short passages cannot show. Ensemble is the two of them together, and it is the strongest answer we have: they were built separately, and they tend to make different mistakes — which is what makes the pair stronger than either one alone. Combining them roughly halves how often we wrongly flag a real musician.

Legato needs a substantial stretch of audio, so a 30-second preview is not enough for it. That is why a link usually gets Staccato, while an uploaded file can get the Ensemble when Legato is available. The version number beside the name (Ensemble v1) marks which combination produced the result — percentages only compare within the same one.

Should I paste a link or upload the file?

Upload, when you can. A link is analysed using the 30-second preview the streaming service publishes, and that clip is chosen to be catchy rather than representative. An uploaded file gives us the actual recording.

Do you keep my music?

No. Uploaded audio is analysed and discarded — we do not store your files. Being precise: the audio does touch disk briefly while it is being processed, and is deleted afterwards. We keep the result, not the recording.

What are song credits, and why don't they change the verdict?

For tracks on Apple Music we show the credits Apple publishes — producers, writers, engineers. They are context you can read, never part of the calculation. Credits cannot prove a human made a track: we have seen a confirmed AI act file seven human-looking credits.

An old result lost its percentage. Is it broken?

No. When we improve the detection model, probabilities from the old one are no longer comparable to the new one — the same recording can score very differently under each while meaning the same thing. Rather than show you a number that no longer means what it did, we hide it. The verdict itself still stands.

Why does it sometimes say “couldn't analyze”?

Because we could not get the recording, not because we judged it. Some links do not resolve to a track we can identify, and some resolve to a track with no audio we are allowed to fetch. That is a fact about our attempt, not about the music. Try another service's link, or upload the file.

It said the audio was too compressed. Why refuse instead of guessing?

Heavy compression destroys exactly the fine detail the model reads. Below a certain quality any number we produced would be noise wearing a decimal point, so we withhold it rather than publish a confident-looking guess.

Why are some songs answered without running the model at all?

If a recording matches a catalogue release dated before 2020, we call it human without analysing the audio. Generated music that could pass for human did not meaningfully exist before then. That date is a deliberately cautious threshold, not a historical fact.

How do I delete my account?

In the app: Settings › My data › Cancel account. It is immediate and cannot be undone. Your account, your email address and the record of what you analysed are deleted. The verdicts themselves remain, with nothing linking them to you — they describe recordings, not people.