If you are wondering can ai generated content be copyrighted, the short answer is: not by itself. Under current us copyright office ai guidance, content that an AI system generates on its own is generally not copyrightable, but content you make with AI can be protected when you contribute enough original expression of your own, through writing, creative editing, or a genuinely creative selection and arrangement of the material. The question that matters is not whether you used an AI tool. It is whether a human controlled and contributed the actual expression that ended up on the page.
That distinction runs through every part of ai content copyright law issued so far, and it is worth understanding in some detail if your team publishes AI-assisted work regularly.
What does US copyright law actually protect?
US copyright law protects original works of authorship created by human authors. That is not a new rule written for generative AI. It has been the standard for a long time, and the Copyright Office has simply been applying it to a new kind of tool.
This is the foundation the rest of us copyright office ai guidance builds on. The Office's practical test asks a fairly plain question: is this work basically one of human authorship, with the computer acting as an assisting instrument, or did the machine make the expressive choices that give the work its character? When you type a note into a document and spell check fixes a typo, nobody doubts you wrote the sentence. The question gets harder once the tool is doing more of the actual expressive work, choosing the words, the sentence structure, the images, or the layout.
The Office looks at a handful of factors when it works through that question. Who made the expressive choices, the wording, the composition, the visual details. What did the human actually contribute, whether that is original writing, creative editing, selection, arrangement, or modification. How much control did the human exercise over the result, since the Office does not treat time, effort, or detailed instructions as a stand-in for creative control. Whether the human contribution clears the ordinary bar for originality. And what exactly is being claimed, because an applicant should point to the human-authored material rather than try to claim the AI-generated part too.
There is no fixed minimum here. The guidance does not set a required percentage of human editing, a word count, or a number of revisions that automatically makes a piece copyrightable. Every case gets looked at on its own facts.
Is AI-generated text copyrightable on its own?
This is really the question behind how to copyright ai generated text at all. If you type a prompt and the AI system writes the words, the Copyright Office generally treats the system as the one determining how your instructions get carried out. Your prompt communicates an idea, a subject, a tone, or a desired result, but the AI decides the actual wording and structure. For the generally available AI writing tools most teams use today, the Office says a prompt by itself usually does not give you enough control to be the author of the output.
That holds even for a long, detailed prompt. A detailed prompt can describe exactly what you want, but describing what you want is different from controlling how the system turns that description into specific sentences.
Revising the prompt several times, asking for a different version, or generating a batch of drafts and picking one does not change this by itself. The Copyright Office treats that process as generating more options to choose from, not as directly controlling the expression the system produces. The time and effort you put into prompting is real, but effort is not the same thing as authorship, and copyright protects original expression rather than effort.
This is where the Thaler v. Perlmutter case fits in. The US Court of Appeals for the District of Columbia Circuit affirmed the denial of a copyright application that named an AI system as the sole author of an autonomously generated image. The court held that a copyrightable work has to be authored, in the first instance, by a human being. It is a narrow decision built around an application that openly claimed an AI-only, autonomous author, so it should not be read as deciding every question about AI-assisted work. But it does confirm the baseline: a machine cannot be the author, full stop.
If you write the article yourself and use AI only to brainstorm, outline, or organize your thinking without pulling that output into the finished piece, the final human-authored text can be protected in the normal way. That is the clearest case there is, and it is also how a lot of writers already use these tools.
What human contribution actually earns protection?
A few kinds of human work can hold up under this standard, and it helps to know what they look like in practice.
Original writing you produce yourself, whether that is a full article, a paragraph inserted into an AI draft, or original commentary layered on top of generated material, can be protected the same way any writing is protected. The protection attaches to what you actually wrote, not to the fact that you were the one operating the tool.
Creative editing and modification can also earn protection, but only when the editing itself is original. Rewriting an AI-generated sentence in your own voice, adding original passages, restructuring a draft around a genuinely different logic, or adapting material into a new form can all count. Fixing typos, adjusting formatting, or correcting a date does not. The Office does not draw a bright line here, so the safest description is that substantial, original rewriting may be protectable, while mechanical cleanup should not be assumed to convert an AI passage into something you authored.
Selection and arrangement is the third path, and it is easy to overlook. If you take a set of AI-generated passages and organize them into a structure that reflects real creative judgment, that structure and arrangement can be protected even though the individual passages are not. The bar here is the same ordinary originality standard as everywhere else. Sorting content into an obvious or dictated order does not meet it.
None of these paths change the basic answer on how to copyright ai generated text on its own, because none of them make the underlying AI-generated wording itself copyrightable. A film can include AI-generated background art, a comic can include human-authored text next to AI-generated images, and an article can include some AI-generated passages inside a larger human-authored piece, and in each case the copyright covers the human-authored elements and any sufficiently creative arrangement, not the AI-generated material on its own.
Does writing the prompt make you the copyright owner?
This is probably the question behind most of the confusion about who owns ai written content, so it is worth answering directly: usually not, on its own. Anyone searching who owns ai written content is usually really asking whether the prompt itself counts as authorship, and it mostly does not.
The Copyright Office's position is that prompting current generative AI tools generally does not give you enough control over the expressive output to make you the author of that output. You can absolutely own or claim copyright in other parts of the finished piece, your original writing, your creative edits, or a genuinely creative selection and arrangement of the material. What you cannot do is claim the AI-generated wording itself just because you wrote the prompt, revised it a few times, or paid for the tool.
It is also worth separating this from a completely different question: what an AI provider's terms of service say you can do commercially with the output. That is a contract question, and the Copyright Office's guidance does not resolve it one way or the other. A tool's terms might let you use, sell, or publish the output freely even where copyright protection is thin or absent. Whether you can copyright it and what you are contractually allowed to do with it are two separate questions, and it is easy to mix them up.
The Office's January 2025 report, the second part of its ongoing ai content copyright law study, works through this in detail. It also looked at tools that give users more granular control than a plain text prompt, things like letting you edit specific passages, focus on a particular region, request inline suggestions, or adjust reading level. Those tools can provide stronger evidence of human authorship than a bare prompt, because they hand you more direct control over specific expressive choices. They do not create an automatic rule either way. The outcome still depends on how original and how human-controlled the actual contribution is.
What should you disclose if you register your work?
If you plan to register a copyright for a piece that includes AI-generated material, the us copyright office ai guidance from March 2023 sets out what it expects from you.
You can claim copyright in your own contributions to a work that also contains AI-generated material. When you apply, you identify the human authors and describe the human-authored material in the application's "Author Created" field. You should not list an AI system, or the company behind it, as an author or co-author just because the tool was part of your process.
You also have a duty to disclose that a work contains AI-generated content and to briefly explain what you, the human author, actually contributed. Any AI-generated content that is more than a trivial amount should be excluded from your claim, named specifically in the application's limitation section, and described there as material generated by artificial intelligence. The guidance does not give a numerical definition of what counts as trivial, so do not assume a specific word count or percentage puts you in the clear.
Not every use of AI triggers this disclosure. Where the AI was genuinely assistive, brainstorming or outlining that never made it into the final text, the Office may find that no disclaimer is needed at all, because there is no human-authorship question to raise in the first place.
If you already submitted an application and are not sure it disclosed AI involvement properly, the guidance says to contact the Copyright Office's Public Information Office and flag the omission. If a registration has already issued for a work that contains undisclosed AI-generated material, you can correct the record with a supplementary registration that identifies the human-authored material and disclaims the rest. Leaving a material omission uncorrected is a real risk: the Office can take steps to cancel a registration where information essential to registrability was left out or is in question, and a court can disregard a registration in an infringement case if it finds the applicant knowingly supplied inaccurate information that would have led to a refusal.
What this means for content teams publishing at volume
Most marketing teams do not fall neatly into "fully AI-written" or "fully human-written." Real workflows land somewhere in a range, and it helps to know roughly where yours sits.
If a person writes the article and uses AI only for brainstorming or a preliminary outline that never gets pulled into the final draft, the finished piece is on solid ground. If AI produces a first draft and a person substantially rewrites it, adds original explanations and examples, and reworks the structure, the human-authored additions and genuinely creative revisions are the protectable part, even though the leftover AI-generated sentences are not automatically swept in. If AI produces nearly all of the wording and a person does light proofreading, spelling fixes, or a few factual corrections, that light pass generally does not turn the underlying AI text into something copyrightable. And if a piece is generated from a prompt, checked over, and published with little or no original rewriting, the wording itself is unlikely to be protected no matter how many prompt revisions it took to get there.
Keeping track of which category a piece falls into matters more than it might seem, especially once you are publishing at real volume with a mix of writers, freelancers, and AI tools in the loop. It is worth keeping a rough sense of what a human actually wrote, edited, or arranged in a given piece, not because the Copyright Office requires a formal log, but because that is the information you would need if authorship ever became a real question.
Publishing at volume does not have to mean guessing about who wrote what. DeepSmith's Content Studio drafts full articles grounded in your own brand, product, and persona context, but every piece still goes through Produced Content for your review and edit before it publishes, which is exactly the kind of human contribution and control that this whole analysis turns on.
