Who Owns AI-Written Content? Copyright in the Age of ChatGPT

A copywriter named Maya built a freelance practice around marketing copy assisted by AI tools. She used ChatGPT for first drafts, Claude for editorial passes, and her own writing for the final polish. Her clients were happy with the output. The legal question of who owned what stayed in the background for months.

Then a client used one of her pieces in a major campaign and a competitor copied the campaign nearly word for word. The client wanted to enforce. Maya’s question became urgent: what could anyone actually copyright in pieces that had passed through three different AI models and her own editing?

The answer matters more in 2026 than it did in 2022. Most professional writers now use AI somewhere in their workflow. Most don’t have a clear picture of what’s protected and what isn’t. Here’s the actual state of the law and what to do about it.

The base rule

Content generated entirely by AI without meaningful human creative input is not copyrightable in the US. This applies to text the same as it applies to images.

What this means concretely:

• A blog post written entirely by ChatGPT from a single prompt is not copyrightable
• An email drafted by Claude with minimal user input is not copyrightable
• A product description generated by AI is not copyrightable
• Marketing copy where the AI did all the actual writing is not copyrightable

These outputs are in the public domain by default. Anyone can use them. Your prompt that generated them doesn’t give you ownership of what came out.

This is well-established as of 2026, supported by multiple US Copyright Office decisions and the Thaler v. Perlmutter case we covered in our AI art piece. The same logic applies to text.

What you can claim

The category of AI-assisted content that IS copyrightable involves substantial human creative contribution. The line:

Substantial human creative work = copyrightable. Your edits, additions, restructuring, and creative refinement of AI output are your copyright. If you take an AI draft and meaningfully rewrite it, you have copyright in the rewriting.

Selection and arrangement = copyrightable. Choosing among multiple AI-generated options and arranging them into a larger work involves human creative judgment. The selection/arrangement is copyrightable; the underlying selected elements may not be.

Original human writing alongside AI assistance = copyrightable. Your introduction, your transitions, your conclusion, your specific examples, your distinctive voice — all copyrightable when substantively yours.

Prompts alone = not copyrightable. Writing the prompt doesn’t make you the author of the output. The US Copyright Office has been explicit on this.

What ChatGPT, Claude, and Gemini’s terms actually say

Each AI service has terms about ownership of output:

OpenAI (ChatGPT): Their terms grant users “all right, title and interest in and to Output.” This sounds like ownership transfer, but it’s limited by the fact that ChatGPT can’t grant copyright it doesn’t have. If the output isn’t copyrightable to begin with (because no human authored it), there’s no copyright to transfer.

Anthropic (Claude): Similar language. Users get rights to use output, but the underlying copyrightability question is unaffected by the terms.

Google (Gemini): Similar approach.

The pattern: AI companies grant users broad usage rights to outputs but cannot transfer copyright in works that aren’t copyrightable. The terms protect you from the AI company claiming ownership; they don’t create copyright where none exists.

What this means in practice:

• You can use AI-generated text commercially without permission from the AI company
• You can publish it, sell it, modify it
• You cannot prevent others from also using AI-generated text that’s identical or similar to yours
• Your enforcement options against copying are limited to whatever human creative contribution you added

Maya’s specific situation

Going back to Maya from the opening. Her marketing copy passed through:

1. ChatGPT first draft based on a brief she wrote. The first draft is AI-generated.
2. Claude editorial pass that restructured and refined. Still AI-generated.
3. Her own final polish with specific brand voice, examples, and creative additions.

What’s copyrightable in the final piece:

• Her brief (her original written prompt) — copyrightable as her writing
• The AI-generated first and second drafts — not copyrightable
• Her specific edits, additions, and creative refinements in the final polish — copyrightable
• The overall structure if she made substantial structural choices — possibly copyrightable as her selection/arrangement

The enforceable copyright is in her contributions. The competing campaign’s exact copying might still be infringement if it specifically copied Maya’s distinctive elements. Pure copying of AI-generated portions wouldn’t be.

For her case to be strong:

• She needs documentation of what she specifically wrote
• The competing version needs to copy her specific contributions, not just the AI-generated baseline
• The percentage of copying that overlaps her human work matters significantly

How to make AI-assisted work clearly copyrightable

For writers who use AI in their workflow, here’s how to ensure substantial copyright protection:

1. Write the structural framework yourself

Decide what the piece is about, what sections it has, what the argument or arc is. Write the outline. Use AI to generate first-pass content for sections you’ve defined.

The structural framework is creative work. Your outline, your decision-making about what goes where, your editorial judgment about what to include are all copyrightable.

2. Rewrite substantially

Take the AI output and rewrite it. Not “fix typos” rewriting; substantive rewriting. Restructure sentences. Replace phrases with your distinctive voice. Add your specific examples and references. Remove what doesn’t fit.

After substantial rewriting, the output is meaningfully your creative work. The original AI draft becomes a starting reference, not the final product.

3. Add original content

Augment AI-generated drafts with original content you write directly. Specific examples from your experience. Quotes from people you interviewed. Data you collected. Analysis based on your expertise.

Original additions are clearly yours and clearly copyrightable. The more substantial the additions, the more substantial your copyright in the combined work.

4. Document the process

Keep records of:

• Your prompts and editorial direction
• The original AI outputs
• Your iterative editing across versions
• Final approved version

This documentation supports your authorship claim if disputes arise.

5. Disclose appropriately

For commercial works, especially journalism and books, disclose AI use. The US Copyright Office requires disclosure when registering. Beyond that, transparency builds reader trust.

The disclosure typically describes:

• That AI tools were used in the process
• How they were used (drafting, editing, research)
• That substantial human authorship was performed

What to do if you’re publishing AI-assisted work

For a working writer using AI tools:

For commercial writing

• Use AI for first drafts, brainstorming, or research
• Substantially rewrite or compose the published piece
• Document the creative process
• Register the work with appropriate disclosure when copyright matters

For client work

• Be transparent with clients about AI use in your process
• Make sure your contract addresses ownership of deliverables
• Note in deliverable language whether AI was used
• Consider whether your billing/positioning is appropriate for AI-assisted work

For your own publishing

• Maintain editorial control of voice and substance
• Use AI as one tool in your kit
• Don’t represent purely AI-generated content as your own creative work
• Disclose AI involvement to your audience when relevant

For how to register a substantially human-authored book, see our book copyright guide. The same principles apply to AI-assisted manuscripts: register the human-authored work with disclosure of AI involvement.

The journalism and book contexts

Two professional contexts where this gets specifically thorny:

Journalism

Major news outlets have varied policies on AI use. Some prohibit it entirely. Some allow it for specific tasks (research, summarization) but not for published content. Some have disclosure policies.

For freelance journalists, AI use can affect:

• Editorial guidelines compliance
• Copyright in your work product
• Your standing if disputes arise

The reasonable practice is to use AI for research and brainstorming, to write the published piece substantially yourself, and to be transparent if asked about your process.

Books

For book authors, the AI question affects:

• Publisher contracts (some publishers now require AI disclosure)
• Amazon KDP submission (KDP requires AI disclosure)
• US Copyright Office registration (disclosure required)

Major publishers have varied policies. The conservative approach: write the substantial work yourself, use AI for research, brainstorming, and editing assistance, disclose use when required.

The training question (again)

Like AI image generation, AI text generation is built on training data that includes copyrighted works. The training question is being actively litigated.

If you’re a published writer whose work has been used in training data:

• You may have potential claims as part of class actions
• Opt-out registries exist for some platforms
• Your registration discipline (for traditional copyright purposes) also positions you for AI training claims

When AI-generated content is fine

A specific point: not every use of AI-generated content needs to be copyrightable to be valuable. Many use cases work perfectly well with public-domain AI output:

• Internal communications
• Drafts to share with collaborators
• Templates and boilerplate
• Research summaries
• Brainstorming and ideation

The copyright question only matters when you specifically need exclusive ownership. For internal or routine use, AI-generated content works fine without copyright.

Where the law is heading

Three trends to watch in 2026 and beyond:

More specific Copyright Office guidance. The Office has been issuing increasingly detailed guidance. Expect more specificity about what counts as substantial human contribution.

Pending court decisions. The NYT v. OpenAI and related cases will produce significant decisions on training, and potentially on output question by implication.

Industry standards. Major publishers, journalism outlets, and AI platforms are developing more sophisticated disclosure and attribution practices. Standards are emerging.

For working writers, the practical implication is: stay informed, document carefully, and continue producing substantially human-authored work.

The honest summary

The current rule (mid-2026): pure AI text generation is not copyrightable; substantial human creative contribution is.

What this means for working writers:

• Use AI as a tool, not as the author
• Make substantial human creative contributions to anything you’ll claim copyright on
• Document your creative process
• Disclose AI use when required (which is increasingly often)
• Don’t try to claim copyright on output that’s purely AI-generated

For the parallel discussion of AI-generated art, see our companion piece.

Maya, from the opening, ended up with a workable claim. Her client’s lawyer was able to pursue the case based on the specific creative elements Maya had added in the final polish: a distinctive opening, specific brand metaphors, particular sentence rhythms. The competing campaign had copied those elements. The case settled before trial.

The lesson: AI tools change the workflow, but the copyright question still comes down to “what specifically did a human creatively contribute?” Make that contribution substantial. Document it. Register it when it matters. The protection follows the human work, just as it always has.