No single test can prove that a record was made by people. What works in 2026 is a trail of evidence that is hard to fake all at once: honest AI credits through your distributor, a verified artist profile, named people and rooms in your credits, dated session files and stems, and video of the band playing the songs. A filmed take, played live in one room with no overdubs, is the hardest part of that trail to fake.
Tapetown Studio is a mixing, mastering and live-recording studio for indie and alternative music, founded by Rasmus Bredvig in 2011. Since 2016 we have recorded 75 Tapetown Sessions with 56 artists, all played fully live with no overdubs. Tapetown has studios in Aarhus and on Djursland, Denmark. This post is for bands who have been asked "is this AI?", and for bands who want the answer to be obvious before anyone asks.
Why bands are being asked
Half of the new music is now generated. In June 2026 more than half of all new music delivered to Deezer each day was fully AI-generated, about 90,000 tracks a day. Those tracks drew only 1–3% of streams, and Deezer found that up to 85% of the streams on fully AI-generated tracks in 2025 were fraudulent (Deezer, July 2026).
The platforms are sorting. Spotify says it has removed more than 75 million spammy tracks from the platform and more than 50 million slop and spam tracks from recommendations this year, and that tracks with a high probability of being AI-generated are ten times more likely to fall below its 1,000-stream royalty threshold (Spotify, October 2026).
The law now asks AI to label itself. The EU AI Act's transparency rules (Article 50) have applied since 2 August 2026. AI systems that generate audio must mark their output as artificially generated in a machine-readable way, and deepfakes must be disclosed. Systems already on the market before that date have until 2 December 2026 to add the marking (Jones Walker, AI law blog).
Voices and performances are getting protection. In Denmark, a bill introduced in October 2026 would ban sharing realistic digital imitations of a person's face or voice without consent, including imitations of artists' performances. The culture minister hopes it will take effect at the start of 2027 (Euronews).
The result: listeners, curators, labels and festival bookers now ask a question they never used to ask. A band that can answer it with one link has an advantage.
What the platforms show today
Signal | What it tells people | What it does not tell them |
|---|---|---|
Spotify AI credits (beta since April 2026) | Which roles used AI, such as vocals, instruments, lyrics or production, when the artist discloses it through a distributor that supports it | Spotify says the absence of an AI credit doesn't mean AI wasn't used |
Verified by Spotify (since April 2026) | A real, identifiable artist is behind the profile: steady listener activity, good standing, a presence on and off the platform (concerts, merch, linked social accounts), reviewed by people | It verifies the artist, not each recording; profiles that mainly represent AI-generated or AI-persona artists are not eligible |
Spotify AI Persona labels | The artist identity itself is AI-generated; such profiles are left out of recommendations unless a listener follows them | Nothing about human artists |
Deezer AI tags | A track was detected as fully AI-generated; it is removed from algorithmic recommendations and editorial playlists | A track without a tag has no certificate either |
EU AI Act, Article 50 | AI tools must mark generated audio; published deepfakes must be disclosed | Nothing asks a human band to prove anything |
Sources: Spotify's AI credits help page, Verified by Spotify and One Year In; Deezer.
So the platforms label AI and verify artists. None of them certifies that a given recording was played by people. That part is up to you, and it is easier than it sounds.
Why a detector score is not enough
An AI detector estimates from the audio alone. It can flag a real recording, especially a heavily processed or lo-fi one, and it can miss careful AI output. If a platform or a label asks for a detector result, give it to them, but don't build your case on it. A detector says what the audio resembles. Your evidence says what happened.
The evidence a band can keep
Start now, also for records that are already out:
The session files. The DAW sessions, multitracks and stems, with their original dates. Multitracks from a band playing together are hard to generate: every instrument leaks into every other microphone, the way it did in that room on that day.
The takes you didn't use. Keep the false starts, the takes that broke down and the talkback between them.
Demos, voice memos and lyric drafts, with their dates.
Photos and video from the sessions. Phone footage is fine. Wide shots of the room, with the band playing together.
Credits that name people and places. "Recorded by [name] at [studio], mixed by [name], mastered by [name]." Put them on Bandcamp, in your distributor's credit fields and on your website. Named engineers and studios can confirm the work. In our own press archive, only 3% of the articles about our sessions named an engineer (the data), so put the credits where people look: in the video description and in the release credits.
Paperwork. Studio bookings, invoices and emails with the people who worked on the record.
Registration under your own names with your collecting society: Koda in Denmark, PRS in the UK, ASCAP or BMI in the US, SOCAN in Canada, APRA AMCOS in Australia.
Content credentials where your tools support them. C2PA is an open standard for attaching a tamper-evident record of how a file was made (c2pa.org).
Be honest about the parts that did use AI. If a tool generated a sound or a lyric line, disclose that part in the AI credit. A precise, honest credit is more convincing than a blanket claim, and it protects you if the question ever gets serious.
Why a filmed live take carries the most weight
A video can be faked. What is hard to fake is everything a live session brings at once: the band in a room, playing the song from start to end; the bleed and the room sound in the audio; the small mistakes that stay in; the named people who recorded and mixed it; the date and the place; and often independent press that wrote about it when it came out.
That is how we make the Tapetown Sessions:
The band plays the song together, and that performance is the recording.
Nothing is added afterwards: no overdubbed parts, no re-sung vocals.
No default tuning, quantizing or cleanup. Imperfections that carry feeling stay in.
We mix from the room and audience mics outwards, with the bleed, the crowd and the dynamics left in.
Every session's credits name who recorded it, who mixed it and where (session credits).
Our sessions and records have been covered by more than 120 independent music publications in 19 countries since 2015 (Tapetown in the press). In 2026 a live session does a second job: it is press material, booking material and a public answer to "is this real?" that anyone can check. More on the format in Ten Years of Tapetown Sessions.
If someone says your music is AI
Don't argue in the comments. Reply once, with one link.
Make that link a page with the evidence: a session video or photos, the credits with names, and a short note on how and where the record was made.
If a platform flags or removes a track, contact your distributor and send the session files, the credits and the paperwork.
Ask the studio or the engineers to confirm in writing who recorded and mixed the record, and where.
FAQ
Can an AI detector prove my music is human-made?
No. A detector estimates from the audio and can be wrong in both directions. Use it as one item next to your session files, credits, photos or video, and paperwork.
Does Spotify label music as human-made?
Not track by track. Verified by Spotify, since April 2026, shows that a real, identifiable artist is behind a profile, and AI credits show where AI was used when the artist discloses it through a distributor that supports it. Profiles that mainly represent AI-generated artists are not eligible for verification.
Does the EU AI Act require bands to prove their music is not AI?
No. Article 50, which has applied since 2 August 2026, puts the duties on AI providers, who must mark generated audio in a machine-readable way, and on those who publish deepfakes, who must disclose them. A band that plays its own music has nothing to file.
Is a live session proof that the music is not AI?
It is the strongest public evidence a band can have: a filmed performance of the song, played together, with named engineers, a date and a place, and multitrack audio with bleed between the microphones. Together with your session files and credits, it is hard to dispute.
What should go in my release credits?
Who wrote the songs, who played what, who recorded, mixed and mastered the record, and where. Add the credits on Bandcamp, in your distributor's credit fields and on your website, and disclose any AI-assisted part honestly.
Make it obvious
Real music is made by people in rooms, and in 2026 it pays to show it. Keep the files, name the people, film the band playing. If you want a record or a session made that way, read how we work: live sessions, mixing live recordings and online mixing. For the sound side of the same question, see How to Make Your Music Sound Human in the AI Era.
