Straight answers about whether you can copyright AI-assisted work, what the US Copyright Office actually requires, and how Copyrightable documents the human authorship behind your creations. Plain language, accurate to the law, no legalese.
Copyrightable is an authorship-evidence tool for AI-assisted creative work. As you write, design, or code with AI, it captures the human creative process behind the work: your prompts, your edits, your selections, the rounds of revision. It ties that record to a filing and produces a tamper-evident evidence file you can show if anyone questions whether a human authored the result. It does not grant you a copyright. It documents the human authorship that copyright law actually cares about.
No. Copyright is created automatically the moment a human fixes original expression in a tangible form. No company can hand you one. What Copyrightable gives you is evidence of human authorship: a documented, time-stamped, tamper-evident record of the creative decisions you made. That evidence is what helps you register a work, defend it, or rebut a claim that "a machine made it." Think of us as the witness, not the law.
Three ways, depending on your setup. A Claude Code hook captures decisions locally as you work in your editor. A hosted connector (MCP) plugs into Claude Desktop, Cursor, and other connector-aware tools: paste the connector URL, log in, and register the sessions you want on the record. And a paste-transcript path lets you drop in the text of a ChatGPT, Claude, or Gemini conversation after the fact. All three feed the same filing record.
The hook is a small Claude Code integration that runs on your own machine. When you prompt, accept, edit, or redirect the AI, the hook records that decision locally and attaches it to the filing you are building. It runs in your environment, so the capture happens at the moment of creation rather than being reconstructed later.
The hosted connector is a Model Context Protocol endpoint you add inside Claude Desktop, Cursor, or any connector-aware tool. Paste the connector URL into the tool, log in once, and you can register the sessions you want evidenced. There is nothing to install or run on your own machine, which makes it a good fit when you work across several connector-aware tools.
If you already made something in ChatGPT, Claude, or Gemini, you can paste the text of that conversation into Copyrightable and we build a filing from it. It is the retroactive path: useful when you did the creating before you set up capture, or when the work lived entirely inside a chat UI.
A filing is the core object in the product: the container for one creative project, a logo, an article, a song, a feature. All the captured decisions, prompts, edits, and iterations attach to it. The authorship profile and the evidence file are computed for the filing as a whole. One filing, one piece of creative output, one evidence record.
Every captured decision is hashed. Those hashes are combined into a Merkle tree, and the single value at the top, the Merkle root, is a fingerprint of the entire record. If anyone changes, adds, or removes a single decision after the fact, the root no longer matches. That is what makes the evidence tamper-evident: you can prove the record is exactly what it was when captured, without exposing every detail.
No. We do not crawl the web looking for copies of your work, and we are not a content-monitoring service. Copyrightable documents that you authored something. Monitoring for theft and enforcement is handled through partners, not by us. Keep the two ideas separate: we prove authorship, others watch for infringement.
No. Copyrightable produces evidence and explains the public legal standard in plain language, but it is not a law firm and nothing here is legal advice. For decisions about registering, licensing, or litigating a specific work, talk to a qualified copyright attorney.
It depends entirely on the human contribution. Purely AI-generated output, with no meaningful human control over the expression, is not protected by copyright. But a work made with AI assistance can be protected to the extent a human controlled the expressive elements through selection, arrangement, or modification. The question is never "was AI used?" It is "did a human author the expression?"
No. Using AI does not poison or erase copyright. It just means the protectable part of the work is the part a human authored. If you meaningfully selected, arranged, or modified the material, that human contribution can be protected. The raw, unmodified AI output on its own is not.
You can protect the parts you authored. If you took ChatGPT output and substantially rewrote it, selected and arranged passages into a larger work, or used it as raw material that you shaped, those human contributions can be protectable. Text you pasted verbatim with no meaningful change is not. The more your expressive control shows, the stronger the claim.
Same standard as any other AI tool. The model that produced the raw output does not change the test. What matters is whether you controlled the expression: did you select among alternatives, arrange the material, and modify it into your own work? If yes, those contributions are protectable.
Code is treated like any other work of authorship. AI-suggested code you accept verbatim is on the same footing as any unmodified AI output. But the architecture you chose, the parts you edited and refined, the way you selected and arranged components, and the iterative direction you gave are human authorship. Most real software built with an assistant carries substantial human creative control, and that is what is protectable.
The raw image that comes out of a text-to-image generator is the hard case. The Copyright Office has consistently found that prompting alone, even with very detailed prompts, does not make the output human-authored, because you do not control how the model renders the expression. If you meaningfully modified the image afterward, or selected and arranged generated elements into a larger composition, those human contributions can be protected. The unedited generation, on its own, generally is not.
The same human-authorship test applies to music. Lyrics you wrote are yours. Arrangement, selection, and substantial editing of generated audio can be protectable human contributions. Audio the model generated from a prompt, taken as-is, faces the same problem as raw image output: the prompt conveys an idea, but you did not control the expression.
There is no line, no percentage, and no word count. The law deliberately refuses a quantitative threshold. A work is not "51 percent human" or "20 percent AI." What matters is the kind of control, not the amount: did a human exercise creative control over the expressive elements? The Copyright Office decides this case by case, looking at what the human actually did.
Not in the way people hope. A longer prompt is still a prompt: it conveys your idea, but it does not give you control over how the model turns that idea into expression. The Copyright Office has been explicit that prompts alone are insufficient no matter how detailed. Effort and volume are not the test. Control over the expression is.
These are the three recognized ways a human can author expression in AI-assisted work. Selection: choosing this output over others, or choosing what to keep and discard. Arrangement: ordering, structuring, or composing material, deciding what goes where. Modification: altering AI-generated material beyond the trivial, with rewrites and edits that change the expression. Do these meaningfully and you are authoring, not just prompting.
Effectively, yes, for the purely AI-generated parts. If a work has no human authorship, copyright does not attach to it, which means there is nothing stopping others from copying it. It is not "in the public domain" in the sense of an expired copyright, but the practical result is similar: no one owns the purely machine-made expression, so anyone can use it.
If the image is purely AI-generated with no human authorship, then yes, there is no copyright to stop them. This is exactly why documenting your human contribution matters. If you selected, arranged, and modified the material into something you authored, that human contribution is protected, and copying it can infringe.
Protection is automatic: it exists the moment a human fixes original expression in a tangible medium. You do not have to file anything to own a copyright. Registration is the formal record with the US Copyright Office. In the United States, registration is generally required before you can sue for infringement, and timely registration unlocks remedies like statutory damages. Protection is the right; registration is the proof and the gateway to enforcement.
Yes. This is the foundation. The Office protects works that are the product of human authorship. It protects AI-assisted works only to the extent a human controlled the expressive elements through selection, arrangement, or modification. A work with no human authorship is not registrable.
You file as usual, but you must disclose that the work contains AI-generated material and disclaim the parts you did not author. You register the human-authored contributions: your text, your selection and arrangement, your modifications. Copyrightable helps here by giving you a clear, documented record of what you actually authored, which makes describing your human contribution far easier.
You must identify AI-generated content in the application and disclaim it, while describing the human authorship you are claiming. You do not register the machine-made parts. Misrepresenting authorship, or failing to disclose significant AI-generated material, can jeopardize the registration. Honesty about what the AI did and what you did is the safe path.
No. There is no quantitative threshold, no percentage, and no word-count line. The Office has been clear that it evaluates the nature of the human contribution case by case. Anyone quoting you a magic number is making it up.
You can register the human-authored elements of it, however small or large, as long as those elements meet the standard for authorship. You disclaim the AI-generated portions. So a work can lean heavily on AI material and still carry a valid, narrower registration covering your selection, arrangement, and modifications.
Because an unregistered copyright is hard to enforce. Protection exists the moment you create the work, but for a US work you cannot file an infringement lawsuit until the Copyright Office has acted on your application. And only timely registration unlocks statutory damages and attorney fees, which are usually what make a lawsuit economically worth bringing. Without them you are limited to proving actual damages, which is expensive and often impractical. See our article "Why Register Your Copyright (When You Already Own It)" on the blog.
Three things. The right to sue in federal court (a prerequisite for US works). Eligibility for statutory damages of $750 to $30,000 per work, up to $150,000 for willful infringement, plus attorney fees, but only if you register before the infringement or within three months of first publication. And a legal presumption that you own a valid copyright, if you register within five years of publication. The blog article walks through each with current figures.
As of mid-2026, a standard online application through the Copyright Office eCO system is $65, and a single author registering one solely-owned work that is not made for hire can use the $45 single application. A clean online filing currently averages about 3.6 months. Always confirm current rates at copyright.gov, since a fee increase was proposed in 2026. Importantly, your protection dates from the day the Office receives your complete application, not the day the certificate arrives.
No. We are not a law firm and we do not submit applications on your behalf. We capture the human-authorship evidence the Office cares about and produce an authorship statement shaped the way the application asks for it, so that when you file at the Copyright Office yourself, describing your contribution and disclaiming the AI material is straightforward. You can also file on behalf of a client: the creator or owner is the claimant and you are the applicant.
Stephen Thaler tried to register a work and list an AI system as the sole author, with no human creator. The Copyright Office refused, and in 2025 the courts affirmed: copyright requires human authorship, so a work authored entirely by a machine with no human creator is not registrable. The decision is narrow. It is about wholly machine-authored works, and it confirms the human-authorship requirement rather than banning AI-assisted creation.
Zarya of the Dawn is a comic book where the author wrote the text and selected and arranged Midjourney-generated images. The Copyright Office protected her human contributions, the text and the creative selection and arrangement, but disclaimed the raw Midjourney images themselves as not human-authored. It is the clearest worked example of the rule: the human-authored layer is protected, the raw AI generations are not.
The EU AI Act is European regulation of AI systems. Article 50 sets transparency obligations: providers and deployers must mark synthetic content so people know it was AI-generated or manipulated, with the relevant provisions taking effect in August 2026. That is a labeling and provenance rule. It is not an authorship test, and it does not decide who owns a copyright. Disclosure of AI use and ownership of expression are different questions.
C2PA is an open standard for content provenance, and Content Credentials is its consumer-facing implementation: signed metadata that travels with a file describing how it was made and edited. It answers "where did this file come from and was it AI-touched?" That is provenance, not authorship strength. It does not assess whether a human controlled the expressive elements, which is the copyright question. The two complement each other but are not the same thing.
The core principle is settled: copyright requires human authorship, and AI-assisted works are protected to the extent a human controlled the expression. The hard part, drawing the line in specific cases, is still developing through Copyright Office decisions and litigation. The direction is consistent, but the edges keep getting sharper. That is exactly why keeping good evidence of your human contribution is worth doing now.
Generally yes. Using AI to generate ideas, outlines, or rough material that you then write yourself is normal authorship. Ideas are not protected by copyright anyway; expression is. If the expression in the final work is yours, the work is yours. Documenting that the final expression came from you only strengthens the position.
The human-authorship rule applies the same way: someone has to have authored the expression for there to be a copyright to assign or to vest in an employer. Work made for hire changes who owns the copyright, not whether one exists. If a deliverable is purely AI-generated with no human authorship, there may be no copyright to transfer, which is a real risk in client contracts. Documenting human authorship protects both sides of that deal.
It can count for a lot. Substantial modification of AI material, rewrites and edits that change the expression, is one of the three recognized authorial acts. The more your editing shapes the final expression, the stronger your authorship claim. Trivial tweaks do not move the needle, but real, expressive editing does.
Selection is a recognized authorial act, but selecting a single output is usually a thin contribution on its own. Choosing among alternatives matters more when it is part of a larger creative process: selecting and then arranging multiple elements, or selecting and then modifying. One pick, with nothing else, is the weakest version of the claim. Combine it with arrangement and modification and it gets much stronger.
For registration, disclosure is required. Outside registration, disclosure is increasingly a transparency and trust matter, and in some jurisdictions a regulatory one, as with the EU AI Act labeling rules. Being upfront about AI use rarely hurts and often helps. Documenting your human contribution alongside that disclosure is the best of both: honest about the tool, clear about the authorship.
Because automatic protection is only as good as your ability to prove it. If someone challenges your work as machine-made, or you need to register, or a client questions ownership, "trust me, I authored it" is weak. A documented, tamper-evident record of your creative decisions, captured as you worked, is strong. The right exists automatically. The record does not.
Your content is stored so that you can see it, review it, and certify it: that is the point of having an evidence record. We are not in the business of reading or mining your creative work, and we do not sell it. Client-side encryption, where content is encrypted before it ever reaches us, is on the roadmap and something we will build with Enterprise customers who need it. If zero-knowledge handling is a requirement for you, tell us and we will scope it as part of an Enterprise engagement.
Captured decisions and the content tied to your filings are stored so the evidence is available to you and verifiable. The cryptographic Merkle root means you can prove the record has not been altered without us having to expose every underlying detail. You stay in control of what happens to it.
Yes. There is a one-click hard delete. When you delete a filing or your data, it is removed, not soft-hidden. You own your creative record, and that includes the right to erase it.
No. Your captured content is there to serve as your authorship evidence, not as training data. We do not use your filings to train models.
It is tamper-evident, which is the honest and useful claim. Every decision is hashed into a Merkle tree, and the root acts as a fingerprint of the whole record. Any later change to any part breaks the match, so alteration is detectable. That is what makes the record credible as evidence: not that it cannot be touched, but that touching it cannot go unnoticed.
Yes, and there is no clock on it. Free is capped by quantity, not by time, so there is no 7-day trial winding down in the corner. No card either. You capture your process on every surface, watch your authorship profile build live, and certify one filing, all without paying. We charge only when you want unlimited filings and the clean, submittable file.
Free is zero, forever. Pro is 99 dollars per seat per month and unlocks unlimited filings, full certification with Merkle proofs, the filing-ready export, team seats, and privacy controls. Enterprise is a custom engagement: tell us what your org needs, whether that is self-hosting, client-side encryption, SSO with audit controls, or an SLA, and we scope it with you. See the pricing page for the full breakdown.
The filing-ready export. On every plan you can see your real certified artifact at a private link: the authorship profile, the decision breakdown, the Merkle proof. On Free that page is watermarked "preview, not for filing" and the Material Excluded declaration the Copyright Office requires is withheld. Pro lifts the watermark and lets you download the clean file with that declaration included, ready to attach to a copyright.gov application. You pay for the submittable asset, not for access to the product.
Same evidence, two states. The preview is the live page anyone with the link can open. It proves the record exists and shows what is in it, watermarked so nobody mistakes it for a finished filing. The export is the downloadable document with the Copyright Office disclaimer language built in and the watermark removed. The preview tells the story. The export is the thing you file.
Because the parts that scale are cheap. We retain hashes rather than your full transcript, which cuts roughly 97 percent of the bytes, so storage is not the bottleneck. The real variable cost is the AI scoring that reads your human-authored work, which is why Free analyzes once at capture and Pro re-analyzes across the full window on demand. Composing a certificate costs about a tenth of a cent. So the gate is not about our costs. It is about the value of a file you can genuinely file, and keeping junk filings out of the Copyright Office.
You can capture your creative process, see the resulting authorship evidence, certify one filing, and view that real certified artifact at a shareable private link. The only thing held back is the filing-ready export: on Free the artifact is watermarked as a preview. It is a real, usable tier, enough to document a project and see exactly what your evidence looks like before you pay.
Free is the trial, and it never expires. If you run a team or an AI writing platform and want to feel the full flow with real exports before committing, talk to us. We hand out comped, time-limited beta access so you can take a finished filing-ready file all the way through. Email copyrightable@thedevelopers.dev.
Free lets you capture, profile, and certify one filing, with the artifact viewable as a watermarked preview. Pro (99 dollars per seat per month) gives you unlimited filings, full certification and Merkle proofs, the filing-ready export, team seats, and privacy controls. Enterprise is a custom engagement for larger organizations: tell us what you need, whether that is self-hosting, client-side encryption, SSO with audit controls, or an SLA, and we scope and build it with you.
No. Registration happens with the US Copyright Office, not with us, and the human-authorship rules apply regardless of which Copyrightable plan you are on. A paid plan gives you more filings and capture options, which makes documenting and describing your authorship easier, but the legal right and the registration process are independent of your subscription.
The monthly plans are month to month. Because there is a one-click hard delete and you own your record, you are never locked in by your data. For Enterprise terms, that is part of the custom conversation.
The vocabulary of copyright and AI authorship, defined plainly. These are the terms, cases, and standards anyone working in this space should know.
Still have a question? Read the authorship methodology, see pricing, or try it free and watch your own evidence build as you create. Copyrightable is evidence of authorship, not legal advice.