You are writing a novel with ChatGPT, you publish it, and it sells. Six months later, someone copies your text word for word and puts it on sale under another name. Under French law, nothing guarantees that you can assert your rights over this text: copyright protection is not automatic when artificial intelligence has participated in the writing.
French Copyright Law and AI-Generated Texts
Most articles on the subject list the uses of ChatGPT for writing a book or detail the steps of publication. Few focus on the legal foundations that frame the issue in France.
The principle is clear. An AI-generated text without significant human intervention falls into the public domain. No one owns it, neither the user who typed the prompt nor the software publisher. This is not a temporary legal void: it is the direct application of an ancient foundation of French copyright law.
The central question, as detailed on Le Site Tech, concerns what distinguishes a protectable text from one that is not when an AI is involved in the writing.
Author’s Personality Imprint: The Central Criterion in French Law
French copyright law protects works that bear the imprint of their author’s personality. This criterion has existed for a long time, well before the advent of AI. It means that a text must reflect human creative choices to be considered a work.
Applied to the use of ChatGPT to write a book, this criterion changes everything. A prompt alone is not enough to constitute a protectable act of creation. Typing “write me a thriller in 20 chapters about a detective in Lyon” and then publishing the raw result gives you no rights over the text obtained.
On the other hand, if you use ChatGPT as a tool for assistance, reworking each chapter, rewriting passages, restructuring the plot, and adding your personal style, the legal situation changes. The final text then bears your creative imprint. It is this identifiable human contribution that opens the way to copyright protection.
Where to Draw the Line Between Assistance and Generation
No precise threshold exists in the law. The doctrine relies on a case-by-case assessment.
Some concrete benchmarks allow for positioning:
- Content entirely generated by ChatGPT and then published without modification: no protection, public domain
- A text where the author uses AI to produce drafts that they deeply revise (structure, style, dialogue, tone): protection remains conceivable if the personal imprint is demonstrable
- A book where ChatGPT corrects grammar or suggests synonyms, but where the writing is human: the work remains normally protected, as with any software proofreader
The more your creative intervention on the text is documented, the stronger your legal position.
European AI Act and Transparency Obligations for Books
The AI Act, the European regulation on artificial intelligence, is often cited as imposing a labeling obligation for AI-generated content. The reality is more nuanced when it comes to a book.
Article 50 of the AI Act primarily targets the technical marking of content by AI model providers. For users (referred to as “deployers”), the obligation to indicate AI-generated content mainly concerns texts disseminated to inform the public on matters of public interest. An exception exists when the text has undergone human editorial control under the responsibility of an identified person.
A novel or essay published under your name, proofread and revised by you, does not automatically fall under this reporting obligation. The AI Act does not create a general rule such as “any book assisted by AI must be labeled.”
Training Models and Rights of Existing Authors
European law also acts upstream. Protected works used to train AI models raise a distinct question: that of the rights of authors whose texts have contributed to the learning of ChatGPT.
This issue does not directly concern the author using ChatGPT to write. It affects the rights holders whose creations have been integrated into the training data. Europe is gradually strengthening transparency and licensing obligations for providers of generative AI models on this point.
Concrete Ways to Protect Your Book Written with ChatGPT
Knowing that copyright requires a human imprint is not enough. You also need to be able to prove it if a dispute arises.
- Keep your successive versions of the manuscript: drafts, corrections, rewriting notes. These traces document your creative process and show that the final text results from personal choices
- Maintain a history of your exchanges with ChatGPT (prompts and responses) to show what the AI produced and what you modified
- Deposit your final manuscript with a timestamped organization (Soleau envelope, deposit with a bailiff, certified online service) to establish a certain creation date
The issue of public aid for literary creation also deserves attention. An audit by the Court of Auditors reported in September 2026 mentioned the possibility of an “identifiable human contribution” as a condition for eligibility for certain cultural aids. This is not yet a rule in force, but a signal to watch for authors seeking grants or residencies.
The legal framework surrounding the use of ChatGPT to write a book is based on a stable principle of French law: no identifiable human author, no protection. Those who revise, restructure, and bring their voice to the text hold rights. Those who merely transmit a prompt hold none.
