How Much Human Editing Makes an AI Song Copyrightable?

Why there is no magic percentage, what human authorship actually means, and how to document your creative contribution.

A suspiciously confident claim keeps appearing in conversations about AI-generated music:

Change 20% of the song and it becomes copyrightable.

Sometimes the number is 10%. Sometimes it is 30%. Nobody seems to know where the percentage came from, but it usually arrives wearing a serious face and no legal source.

The U.S. Copyright Office does not use a magic human-editing percentage.

It does not ask whether you changed 20% of the waveform, replaced three instruments or spent exactly seven hours moving clips around Ableton.

The real question is whether the final work contains original human-authored expression and whether the human creator exercised meaningful creative control over the elements being claimed.

Purely AI-generated material is not protected under the current U.S. Copyright Office approach. However, a work containing AI-generated material may still receive copyright protection for human-written lyrics, human-composed music, recorded performances, original arrangement choices and sufficiently creative modifications. The analysis is made case by case.

So how much human editing is enough?

The honest answer is not a percentage.

It is authorship.

There Is No Legal Human-Editing Percentage

The U.S. Copyright Office’s report, Copyright and Artificial Intelligence, Part 2: Copyrightability, does not create a numerical threshold for human involvement.

Instead, it says that:

• copyright protects original expression created by a human author
• purely AI-generated material is not protected
• insufficient human control over expressive elements is not enough
• human contributions must be assessed case by case
• creative selection, arrangement and modification may be protected
• prompts alone usually do not provide sufficient control under current generative AI systems

That is the framework. No calculator. No mystical 20% rule. No plugin that turns green when the song becomes legally human.

Changing a limiter ceiling from -0.1 dB to -1 dB is not a meaningful transformation of the musical authorship.

Neither is:

• converting the file from MP3 to WAV
• adding a fade-out
• making the master louder
• changing the sample rate
• cutting two seconds from the intro
• adding basic EQ
• renaming the file MY ORIGINAL HUMAN SONG FINAL.wav

Those actions may affect the file, but copyright is concerned with protectable creative expression, not merely whether somebody touched the audio.

The amount of time spent is not decisive either.

You can spend eight hours removing clicks and still contribute less original authorship than someone who writes a new chorus in twenty minutes.

Copyright law is not billing by the hour.

What Human Authorship Actually Means

Human authorship means that a person created original expressive elements rather than merely requesting, selecting or technically processing material produced by a machine.

For music, that expression may exist in different parts of the work.

This is important because a recorded song may contain two legally separate works:

1. The musical composition, meaning the music and any accompanying lyrics.
2. The sound recording, meaning the particular recorded performance and production.

The songwriter may contribute authorship to the composition. Performers, producers and recording engineers may contribute authorship to the sound recording. The two copyrights are separate and can have different authors and owners.

An AI-assisted track could therefore contain a copyrightable human-written lyric even if part of the instrumental was generated.

It could contain a copyrightable human performance even if the underlying composition includes unprotectable generated material.

It could also contain protectable human arrangement or production choices without making every original AI-generated element separately copyrightable.

Copyright does not always arrive as one giant blanket covering the entire WAV.

Sometimes it protects specific human contributions inside a hybrid work.

Human-Written Lyrics and Composition

Human-written lyrics can qualify as human-authored expression.

The same applies to music created by a human composer, including original melody, rhythm and harmony.

For example, imagine that an artist:

• writes the lyrics
• creates the lead melody
• writes the chord progression
• establishes the song structure
• uses an AI system to generate an initial backing track

The human-written lyrics and original musical elements may be protectable even if the generated backing material is not.

The important question is whether those musical or lyrical elements genuinely originated from the human creator and remain perceptible in the final work.

A prompt saying:

Write a sad alternative pop song about leaving home.

is an idea or instruction.

Actually writing the verses, chorus, melody and harmonic movement is expressive authorship.

Those are not the same contribution.

Recorded Performances

A real recorded performance may support human authorship in the sound recording.

This could include:

• a human lead vocal
• live guitars
• played drums
• recorded synthesizer performances
• spoken sections
• instrumental improvisation
• original backing vocals

The U.S. Copyright Office explains that sound recording authorship may come from performers, producers or both. A sound recording protects the particular series of fixed sounds, while the underlying composition remains a separate work.

Suppose an AI system generates the chords and basic instrumental, but you write and record a completely original lead vocal.

That human vocal performance may contribute protectable authorship to the final sound recording. Your human-written melody and lyrics may also support a separate claim in the musical composition.

That does not necessarily grant you copyright over every generated instrument underneath it.

It means the final track contains identifiable human contributions that can be analysed separately.

Creative Arrangement and Selection

Creative selection, coordination and arrangement can also qualify for protection.

The Copyright Office recognises that a person may combine human-authored and AI-generated material in a sufficiently creative way so that the arrangement of the overall work contains protectable authorship. However, protection generally applies to the human arrangement, not automatically to the underlying AI-generated elements.

In music, that could include original decisions about:

• which generated sections are used
• how different passages are combined
• where the verse and chorus occur
• which vocal takes appear
• how contrasting elements interact
• where silence, transitions and breakdowns are placed
• how several generated sources are shaped into a coherent work

Simply choosing one complete result from four generated versions may not demonstrate much authorship.

But taking fragments from different sources, restructuring them, composing connecting sections, creating new transitions and building a deliberate arrangement may be much more significant.

The distinction is between merely selecting a meal from the menu and entering the kitchen to create the final dish.

Pressing regenerate seventeen times is still mostly ordering.

Substantial Musical Modification

Human modifications can also receive protection when they contribute original authorship.

The Copyright Office says that a person may modify AI-generated material to such a degree that the human modifications meet the originality standard. Copyright would then cover the new human-authored material, not necessarily the underlying generated content.

In practical music production, meaningful modification might include:

• rewriting the melody
• replacing the chord progression
• creating a new bass line
• composing additional instrumental sections
• changing the rhythm and groove
• rebuilding the chorus
• writing a new bridge
• replacing generated vocals with an original performance
• substantially changing the structure
• creating new countermelodies and harmonies
• reconstructing major portions of the production

The focus is not whether the final waveform is mathematically different.

The focus is whether the human added original musical expression.

Pitching the entire master down by one semitone changes every sample.

It does not necessarily create a new human-authored composition.

A new melody might.

Why Prompts Alone May Not Be Enough

The U.S. Copyright Office currently takes the position that prompts alone generally do not give the user enough control over the expressive details of an AI output.

A prompt can describe the desired result, sometimes in great detail. But the generative system still decides how those instructions are expressed in the final output.

The Office compares prompts to giving instructions to another creator. You may explain what you want, but that does not necessarily make you the author of the resulting expression if you did not control how the expression was executed.

This does not mean prompts are worthless.

A sufficiently original written prompt could potentially be protected as text in its own right. But copyright in the wording of the prompt does not automatically produce copyright in the generated song.

For example:

Create a slow psychedelic rock track with a baritone vocal,
a descending guitar motif, funeral drums and lyrics about
watching your old life disappear through a train window.

That may communicate atmosphere, instrumentation and subject matter.

It does not necessarily determine the exact melody, chord voicings, rhythm, lyrics, vocal performance, arrangement or recorded sound.

The AI system still makes many of those expressive decisions.

The same principle applies whether the tool is Suno, Udio or another generative music platform.

The name of the generator does not determine copyrightability. The question is what human-authored expression exists in the final song and what expressive control the human actually exercised.

What Parts of a Hybrid Song May Be Protected?

A hybrid AI-assisted song does not always receive a simple yes-or-no answer.

Different parts may have different copyright status.

Imagine a track containing:

• AI-generated instrumental material
• lyrics written by a human
• a melody written by the same human
• a real vocal performance
• human-recorded guitars
• a manually rebuilt arrangement
• human mixing and production choices

Possible protectable elements may include:

• the human-written lyrics
• the human-composed melody
• original human harmonic or rhythmic material
• recorded human performances
• sufficiently original production authorship in the sound recording
• creative selection and arrangement of the different materials
• substantial human modifications

The unchanged AI-generated parts may remain outside the copyright claim.

This is why asking whether “the song” is copyrightable can sometimes be too vague.

Which part?

The composition?
The lyrics?
The recorded vocal?
The complete master?
The arrangement?
The generated instrumental?

Under U.S. registration guidance, applicants using AI-generated material should identify the human-authored contributions and exclude more-than-minimal AI-generated content from the claim.

That is not the most romantic part of making music.

Neither is naming stems correctly, but civilisation survives through such sacrifices.

Editing vs Authorship

Not every edit is authorship.

A useful distinction is whether the work merely improves the technical presentation of existing material or adds original musical and expressive choices.

Technical editing can be valuable, skilled and commercially important without necessarily creating enough authorship to claim the whole underlying work.

A mastering engineer may perform excellent work on a song without becoming the songwriter.

Skill and copyright authorship are related sometimes, but they are not identical.

Basic Mixing and Mastering

Basic mixing and mastering normally involve decisions about:

• level
• equalisation
• compression
• limiting
• stereo balance
• loudness
• tonal consistency
• translation

These decisions can involve judgement and craft.

Producers and engineers can contribute authorship to sound recordings where they creatively manipulate, process and fix the final sounds. However, whether a particular contribution is sufficiently original remains fact-specific.

The safe practical conclusion is this:

Basic processing should not be treated as a guaranteed method of converting an entirely generated song into a fully human-authored work.

Uploading an AI song into Ableton, putting Pro-Q on it, adding a limiter and exporting it again is still mostly the same song.

The limiter has worked hard.

It has not composed anything.

More original and detailed production choices may contribute protectable authorship to the resulting sound recording, but they do not retroactively make the generated composition human-written.

Rebuilding the Arrangement

Rebuilding the arrangement is much closer to traditional creative authorship.

This may involve:

• removing entire sections
• composing a new introduction
• changing the verse structure
• creating a pre-chorus
• rewriting the chorus entry
• adding a human-composed bridge
• combining fragments from several generations
• creating new transitions
• changing the final emotional arc

Creative selection and arrangement can be protected where the choices meet the originality standard. The Office makes clear that this remains a case-by-case question.

A useful test is whether the final structure reflects your own creative conception or whether it is still essentially the complete structure delivered by the generator.

Moving the second chorus eight bars earlier may be an edit.

Reconstructing the entire progression of the song may be authorship.

Context matters.

Replacing Vocals and Instruments

Replacing generated parts with original performances can create clear human contributions.

Examples include:

• recording a new lead vocalist
• playing new guitar parts
• replacing the generated bass
• programming an original drum performance
• recording live percussion
• creating new synthesizer parts
• adding original backing vocals

These contributions may affect both the musical composition and the sound recording, depending on what was created.

A newly written bass line may contain compositional authorship.

A particular performed and recorded bass part may also contribute to the sound recording.

Replacing elements can therefore do more than clean the source. It can build a genuinely human production around the original AI-assisted concept.

Rewriting Melody and Lyrics

Rewriting melody and lyrics is one of the clearest ways to add human authorship.

If you replace generated lyrics with your own writing, the new lyrics may be protected.

If you compose a new lead melody, that melody may form part of a protected musical work.

If you rewrite the chorus, alter the harmonic progression and create a new bridge, those human-authored elements may be claimed even where other generated material remains.

This is very different from changing one generated word or moving one note.

Again, there is no percentage.

A short but original chorus may be more legally meaningful than hours of cosmetic editing across the entire master.

How to Document Human Contribution

The U.S. Copyright Office requires registration applicants to disclose AI-generated content and briefly explain the human contributions being claimed. More-than-minimal AI-generated content should generally be excluded from the claim.

That makes documentation useful.

You do not need to film yourself moving every MIDI note while holding today’s newspaper.

But you should preserve enough material to explain how the song developed.

Session Files and Version History

Keep the DAW project.

Do not preserve only the final stereo master.

Useful records include:

• dated Ableton, Logic, FL Studio or Pro Tools sessions
• automatic backup versions
• arrangement revisions
• exported rough mixes
• revision notes
• track freeze and bounce history
• collaborator messages
• production change lists

A version history can demonstrate that the final track did not simply arrive from one prompt and receive a louder export.

It can show the arrangement developing, generated sections being replaced and original parts being introduced.

Use normal, understandable file names.

final_final_REAL_12_USE_THIS_ONE.als may be emotionally honest, but it is not ideal evidence management.

Raw Recordings and MIDI

Keep raw human recordings.

This may include:

• vocal takes
• guitar recordings
• bass recordings
• percussion
• piano or synthesizer performances
• spoken material
• tuning and comping sessions

Also preserve original MIDI for parts you composed or performed.

MIDI can help show:

• melody construction
• chord voicings
• rhythmic changes
• arrangement development
• original human programming
• revisions over time

Do not flatten everything into audio and immediately delete the source.

Storage is cheaper than regret.

Lyric Drafts

Keep lyric drafts, especially if the lyrics were written or heavily revised by you.

Useful records include:

• notebook photographs
• text documents
• phone notes
• emails
• dated cloud documents
• alternative verses
• rhyme lists
• handwritten corrections
• recordings of early melody ideas

A finished lyric does not need to emerge fully formed like Moses returning from the mountain.

Drafts are normal evidence of human writing.

Generated Source Material

Keep the original AI-generated files too.

That may sound backwards, but it helps show the difference between the source and the final work.

Preserve:

• original generator exports
• stems downloaded from the service
• prompts
• generation dates
• alternate versions
• tool and model information
• source file names
• original tempo and key where available

Then keep the human-edited versions alongside them.

A clear before-and-after record can show which parts were retained, replaced, rewritten or rebuilt.

Do not hide the source.

Document the transformation.

Copyright, Platform Acceptance and Commercial Rights Are Different

These concepts are often thrown into one bucket, where they immediately begin fighting.

They are separate questions.

1. Is the human contribution copyrightable?

This is a legal authorship question.

Under the current U.S. Copyright Office approach, purely AI-generated material is not protected, while identifiable human-authored expression may be protected in whole or in part.

2. Do the generator’s terms allow commercial use?

This is a contractual question.

A music generator may give a user certain commercial permissions under its terms.

That permission does not necessarily prove that the output contains copyrightable human authorship.

A contract can define what the company permits you to do in its relationship with you.

It cannot simply rewrite the U.S. Copyright Office’s authorship standard.

3. Will a distributor or streaming platform accept the release?

This is a platform-policy question.

A distributor may have its own rules concerning:

• fully AI-generated music
• ownership evidence
• voice cloning
• artist impersonation
• commercial licences
• AI disclosure
• mass-generated uploads
• fraudulent streaming

A track can be contractually licensed for commercial use and still face distribution restrictions.

It can be accepted by a distributor and still contain limited copyrightable authorship.

It can also be copyrightable in part while violating another person’s rights.

“Uploaded successfully” is not a legal opinion.

The green distributor button has many gifts.

A law degree is not one of them.

Registration Does Not Mean Claiming Everything

In the United States, human-authored copyright protection generally begins when the original work is fixed, such as when the song is recorded or written down. Registration provides additional benefits, including a public record and, for U.S. works, access to federal courts for infringement actions.

For hybrid AI works, registration should describe the human authorship honestly.

The Copyright Office instructs applicants to:

• disclose the presence of AI-generated material
• describe the human-authored contribution
• claim the human-created elements
• exclude more-than-minimal AI-generated material
• avoid listing the AI system as an author or co-author merely because it was used

The Office may register the human-authored parts without extending protection to the generated material itself.

This is not necessarily a failed registration.

It may simply be a limited claim reflecting what the human actually created.

For a commercially important release, speak with a qualified U.S. copyright lawyer before filing. A studio article can explain the map. It should not pretend to be the courthouse.

When an AI Track Rebuild Makes Practical Sense

Sometimes the copyright question reveals a production problem.

You may look at the track and realise:

• the lyrics came entirely from the generator
• the melody came entirely from the generator
• the vocal is synthetic
• the arrangement is almost unchanged
• the human work consists mainly of mastering
• there is little original material to point to

That does not automatically mean the track cannot be released under the provider’s terms.

But it may mean there is very little human authorship in the actual song.

An AI Track Rebuild can make practical sense when the idea is worth keeping but the human creator wants to contribute genuine musical authorship.

That might involve:

• writing new lyrics
• composing a new topline
• rebuilding the chords
• recording a real vocalist
• replacing generated instruments
• creating a new arrangement
• adding human performances
• producing a new mix from rebuilt parts

The goal should not be to perform a cosmetic legal ritual.

The goal is to turn the generated idea into a real human-led production.

Do not replace one snare and declare victory over jurisprudence.

Rebuild the parts that matter.

The Honest Answer: Prove the Work, Not a Percentage

There is no reliable answer such as:

• change 20%
• replace three stems
• add one real vocal
• spend ten hours editing
• use 51% human audio

Copyrightability is not measured by project duration, file count or percentage of altered samples.

The stronger practical questions are:

• What did the human write?
• What did the human compose?
• What did the human perform?
• What expressive decisions did the human control?
• What material did the human select and arrange creatively?
• What generated elements were substantially modified?
• Which claimed elements remain clearly identifiable in the final song?
• Can the creator document that process?

A prompt can begin the work.

It does not automatically establish authorship of the output.

Mixing can improve the work.

It does not automatically create authorship of the underlying composition.

A real human contribution must exist in the expressive result, not only in the story told after release.

At Unsaid Records, AI Track Rebuild work is not about applying a legal-looking coat of paint to a generated master. It is about adding real production, arrangement, performance and creative decisions where the song actually needs them.

Because the best evidence of human authorship is not a percentage.

It is the work itself.

Legal note: This article provides general information based primarily on the current U.S. Copyright Office position. It is not legal advice and does not establish how copyright law applies in every country or to every individual release. Copyrightability, ownership, registration, contractual rights and platform acceptance are separate issues. Obtain advice from a qualified lawyer for a commercially important or disputed work.