If you make songs with Suno, Udio or any other generative tool and you want to register them in the United States, the question isn't really "can AI music be copyrighted?" It's "which parts of this song did a person make, and can I show it?" The US Copyright Office's answer is clear enough: it will protect the human-authored parts and nothing else. The hard bit is proving which parts those are, months later. This guide sets out a simple way to build that record as you go, so the evidence exists before anyone asks for it.
A quick note before we start: we're one human songwriter and Petunia, our AI collaborator, not lawyers, and this isn't legal advice. If a song matters commercially, or you're unsure what to claim, talk to a copyright lawyer before you file. If you're releasing in the UK, the law is different; this guide is about the US standard.
What the Copyright Office Actually Protects
In January 2025 the Copyright Office published Part 2 of its report on copyright and AI, covering copyrightability. As the Copyright Alliance's summary sets out, the key points are these. Material that is wholly AI-generated isn't protected. Using AI as a tool doesn't, on its own, stop a work from being protected. Prompts alone don't give a person enough control over the result to make them its author, and writing prompt after prompt doesn't change that: the Office compared it to re-rolling the dice, and rejected the idea that effort alone counts.
What does count is human expression you can point to. The report names three routes. First, your own creative input that stays perceptible in the output, such as lyrics you wrote. Second, original changes you make to AI output. Third, original selection, coordination and arrangement of material into a larger work. The Office's earlier registration guidance from March 2023 adds the paperwork side: applicants must disclose AI-generated content that is more than minimal, briefly explain what the human contributed, and exclude the AI material from the claim.
For a song, that usually means your lyrics, any melody or chords you wrote yourself, and real edits and arrangement decisions may be claimable, while a vocal or instrumental generated from a prompt is the part you disclose and exclude.
Why 600 Prompts Weren't Enough
The case most people point to is Allen v. Perlmutter. Jason Allen made the image "Théâtre D'opéra Spatial" in Midjourney, prompting it at least 624 times by his own account, then adjusted details in Photoshop and upscaled it. It won first prize at the Colorado State Fair. The Copyright Office refused to register it, and its Review Board said his only contribution to the Midjourney image itself was entering the text prompt. Allen sued in federal court in Colorado. The most recent reporting we found, from May 2026, described the case as still waiting for a decision, so check its status before relying on it.
Two lessons matter for songwriters. The Office said Allen could have sought a narrower registration covering his own later edits while disclaiming the AI-generated parts. And the dispute shows how little a pile of prompts proves on its own. A related case, Thaler v. Perlmutter, ended in March 2026 when the Supreme Court declined to hear it, leaving in place the appeals court ruling that a work needs a human author. Aaron Moss at Copyright Lately notes that no court has yet set out what evidence proves human authorship, which is exactly why a clear, dated record is worth having.
Step by Step: Build the Paper Trail
You don't need special software. A folder per song, a plain text log and the files you already make are enough. The rule is to record things on the day, because this is the part you can't rebuild honestly later.
Step 1 — Open a song folder before the first session. Name it like 2026-10-31_working-title. Inside, make four subfolders: 01-lyrics, 02-generations, 03-edits and 04-final, plus a file called session-log.txt.
Step 2 — Save your lyric drafts as separate dated files. Write the lyrics yourself where you can and save each meaningful revision as a new file rather than overwriting, for example lyrics_v1_2026-10-31.txt, lyrics_v2_2026-11-02.txt. If an AI tool suggested a line and you kept it, mark it, for example [AI suggestion, kept]. Your own words, kept in order with dates, are the clearest human contribution most AI-assisted songs have.
Step 3 — Log every session in one line or two. For each session, note the date, the tool and the model or version shown in the app, and what you did. For example: 2026-11-02 · Suno (model shown in app) · 12 generations from lyrics_v2 + style prompt "warm acoustic folk, female vocal, fingerpicked guitar" · kept generation 7 for verse, 11 for chorus. Copy the exact style prompt into the log. Prompts won't make you the author, but they tie each file to a session.
Step 4 — Keep the generations you chose, and say why. Download the generations you actually used into 02-generations and give them names you'll understand later. Add a short note on why you picked one over another.
Step 5 — Record each edit with a before and after. This is where most of the claimable human work usually happens. If you split stems, rewrote a line and regenerated one section, cut a verse, moved the bridge, re-recorded a vocal or played a new guitar part, write it down and keep both versions. For example: 2026-11-04 · moved chorus 2 after bridge, cut 8 bars of intro, replaced AI guitar with my own recorded part (guitar_take3.wav). Save your DAW project files with the date in the name, so the arrangement itself is preserved.
Step 6 — Keep anything you made outside the AI tool. Voice memos of a melody you hummed, chord charts, a recorded guitar or vocal part: these are your own expressive input, and if they stay recognisable in the finished song they're the strongest evidence of all. Keep the original files with their dates.
Step 7 — Write a contribution summary when the song is finished. In 04-final, write a short plain statement of who did what. For example: Lyrics written by [name]. Melody of the chorus written by [name]; verse melody, vocal and instrumental performance generated with [tool]. Arrangement, edits and added guitar part by [name]. Be honest about the AI parts. The point is accuracy, not making your share look bigger.
Step 8 — Back it up and leave it alone. Copy the whole folder somewhere safe, such as a second drive or cloud storage, once the song is released. Don't rename files afterwards: the original dates are part of the evidence.
Turning the Record Into a Registration
The Copyright Office's 2023 guidance says to use the Standard Application, describe the human contribution in the "Author Created" field, and exclude AI-generated material that is more than minimal under "Limitation of the Claim", in the "Material Excluded" area. It gives example wording along the lines of "selection, coordination and arrangement of [human content] created by the author and [AI content] generated by artificial intelligence". It also says not to list an AI tool or the company behind it as an author. Your contribution summary from Step 7 is the draft for these fields.
If you've already registered a song without mentioning AI, the guidance says to correct it with a supplementary registration. Leaving AI out risks losing the benefits of the registration: the Office can cancel it, and a court can disregard it if the applicant knowingly gave inaccurate information. A National Law Review article from May 2026 adds a practical point: if a registration is ever challenged, prompt logs and workflow records can come out in the legal process, so the record should match what you claimed.
Common Mistakes
Counting prompts as authorship: a long prompt history shows effort, not authorship. Log it, but don't build your claim on it.
Rebuilding the log later: a record written months afterwards is weaker and easy to get wrong. Record it on the day.
Overwriting drafts: one lyrics file edited in place loses the history. Save versions.
Claiming the whole song: claim the human parts and exclude the AI parts. Over-claiming can cost you the registration.
Forgetting the outside material: voice memos and played parts are often your best evidence. Keep them.
The same folder also helps with platform AI labels. Our guide to documenting an AI-assisted track for platform disclosure shows how to map it to Spotify, Apple Music, YouTube Music and TIDAL.
Where That Leaves Us
The law here is still settling, but the practical answer won't change much: the human parts of an AI-assisted song are the parts you can show, and a dated folder of drafts, logs and edits is how you show them. If you'd like to hear the songs we make together, a human songwriter and Petunia, our AI collaborator, they're on the homepage, and our Suno Guide has the prompts, style tags and how-to guides we keep coming back to.
Sources
US Copyright Office — Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence (March 2023; disclosure duty, "Author Created" and "Material Excluded" fields, supplementary registration, risks of not disclosing)
Copyright Alliance — Summary of the USCO Copyright and AI Report, Part 2: Copyrightability (February 2025; prompts, expressive inputs, modifications, selection and arrangement)
Copyright Lately — Thaler Is Dead. Now for the AI Copyright Questions That Actually Matter (March 2026; Supreme Court cert denial in Thaler, Allen v. Perlmutter pending, documentation advice)
TCAM Today (Faegre Drinker) — Can AI Prompting Satisfy the Copyright Office's Human Authorship Requirement? (May 2026; Allen facts, Review Board decision, case status)
Mondaq — When 600 Prompts Still Aren't Enough: What Allen v. Perlmutter Means (January 2026; limited registration with AI parts disclaimed)
National Law Review — Who Owns AI-Generated Content? Human Authorship Still Controls, and Documenting the Creation Process Is Critical (May 2026; documentation recommendations, prompt logs in disputes)
