What Copyrightable proves, and what it doesn't.

• That a specific stream of prompts, edits, selections, and revisions existed at a specific time.
• That the stream has not been altered since it was captured, because every record is hashed into a Merkle tree and the root would no longer match.
• That the captured decisions show a particular pattern of human control over the expression, assessed against our published standard.

• It does not prove a human, rather than another tool, produced the keystrokes. We capture what happened on your machine. We do not verify biological humanity, and we do not try to.
• It does not certify that you told the truth on a copyright application. That statement is yours to make to the government. We give you the record to make it from.
• It is tamper-evident, not tamper-proof. We can show that a record was changed. We cannot make it physically unchangeable.
• A pasted transcript is weaker than a live capture. A transcript supplied after the fact could in principle be staged, so we label provenance rather than hide it.

The US Copyright Office runs no AI detector. Registration is an honor system: you disclose your AI use and describe your human authorship, and the Office takes your word. That means someone could fabricate a creative process and lie on a filing. They could do that with a notebook, a Git history, or a screen recording too.

Here is how we think about it. Copyrightable is a tool that records a process. If you use it to document real authorship, you get strong, contemporaneous evidence. If someone uses it to manufacture a false record and then lies to the government, they made that false statement, not us, the same way a spreadsheet doesn't commit tax fraud when someone cheats on their taxes with it. We build the honest tool. The claim is always yours.

And there's a quiet incentive built in. The cheapest convincing way to produce a rich record of creative decisions is to actually make them. The record shows passivity as plainly as it shows control. So the tool nudges real authorship rather than rewarding the fake kind.

A piece of evidence that claims more than it can support is the easiest kind to discredit. We'd rather hand you a record that says exactly what it is, so that when a registration is questioned or a dispute reaches discovery, the evidence holds because it never overreached. Our methodology is public and versioned for the same reason. Transparency isn't a disclaimer here. It's the product. Read the full authorship methodology.

The Copyright Office has registered AI-assisted works - on exactly the basis this record documents. "A Single Piece of American Cheese" (reg. VAu001543942) was registered for the "selection, coordination, and arrangement of material generated by artificial intelligence," with the raw AI components disclaimed. "Zarya of the Dawn" (reg. VAu001480196) was registered for the author's text and the selection and arrangement of elements, excluding the AI images. Elisa Shupe's AI-assisted novel was registered for the selection, coordination, and arrangement of AI-generated text. The throughline: the human's selection, coordination, arrangement, and substantive modification is what's registrable, and the pure-AI material is disclaimed. That is precisely the human contribution your captured record evidences.

Public records: USCO Public Records System (publicrecords.copyright.gov). General information about what the Office has registered, not legal advice.

Under Unicolors v. H&M (2022), a copyright registration is invalidated only if the applicant had actual knowledge that what they filed was false. The practical takeaway runs in your favor when you're honest: an accurate, contemporaneous record is your best protection against an accidental misstatement, and your best answer if anyone claims your work was machine-made. The same record that documents real authorship is the record that keeps a good-faith filing defensible.

This is general information about how copyright registration works, not legal advice. Copyrightable is not a law firm.

Want to see your own record? Try it free, or read how we assess authorship.