AI Best Practices for Authors

Generative AI has become a ubiquitous technology, and some writers are already using it in various ways to assist in the writing process. Many people have turned to AI to draft things like emails, slideshows, recommendations, and other work products, as well as to help with research. Professional writers have also been experimenting with AI and seeking to understand the ethical and legal boundaries of using it in their writing practice.

The Authors Guild first published best practices for AI use by writers in February 2024 to help writers understand where those boundaries lie for their own writing. Our newly updated best practices—the result of conversations with many writers and deliberations by the Authors Guild Council—seek to add some context and clarity around the best practices and are provided in response to the many inquiries we have received around AI issues. These are guidelines, not rules, and are intended to assist members of the writing community in making their own ethical and informed decisions about AI use or non-use.

Guiding Principles

• The Authors Guild believes that it behooves us all to maintain the standards of writing in the profession, to preserve human voices and the thinking that goes into writing, and to prevent quality human writing from becoming a rare luxury good representing only a minority of views.
• All of the commercially available foundational large language models (LLMs) have been trained on pirated, unlicensed books without compensating authors or publishers or giving authors and publishers any control over the use of their works in AI outputs. Anyone using these AI platforms should be aware that they are supporting companies that have engaged in widespread theft of authors’ works.That’s why the Authors Guild supports efforts for fair compensation for the theft of the copyrighted work of professional writers.
• As a writer, it is your original voice, thinking, and creativity that make you the writer that you are. As an author or journalist, you contribute your unique view and thoughts and your unique voice. AI outputs, by contrast, are generic mashups of pre-existing works ingested during training.
• When you claim authorship in a work, it means that the words, thoughts, and imaginings are yours.
• Trust among readers, writers, and publishers is vital to a healthy literary culture. Falsifying the source of a work, including by misrepresenting AI-generated text as human-written, damages that trust.
• Writers should be the ones to determine how AI technologies are used in their profession and should not allow AI companies or other commercial interests to dictate or influence those uses. AI companies have shown themselves to be agnostic, if not scornful, as to the future of literature and the arts and the ethics surrounding AI use.

Risks to Be Aware of When Using AI

1. AI-generated text is not copyrightable, because it is not original human authorship. As such, AI-generated text that is included in a work submitted for copyright registration must be disclosed and disclaimed in the copyright application. What that means is that, assuming there is sufficient human authorship for the work as a whole to be copyrightable, the AI-generated text will not be protected and will be filtered out before any infringement analysis is undertaken. Knowingly failing to disclose AI-generated text may be deemed fraud on the Copyright Office and invalidate the registration. If your publisher is registering the book on your behalf, it needs to know if any elements are AI generated so that it can complete the application correctly.
2. Book contracts contain a representation and warranty that the manuscript is original to the author. AI-generated material is not considered “original authorship.” Inclusion of AI-generated text in the final manuscript may violate the writer’s contractual warranty and be a breach of the contract, allowing the publisher to terminate the agreement in some cases.
3. Once a book is ingested by an LLM in its training, the LLM knows it and may regurgitate its words, plot, characters, etc. AI outputs may include recognizable elements from other authors’ copyright-protected works. Despite the large AI companies’ protestations to the contrary, we are learning that their LLMs do in fact know and sometimes memorize the books they are trained on.
4. AI outputs cannot be trusted. AI tools often generate incorrect or fabricated information and present it as factual. AI companies call these incorrect and misleading outputs “hallucinations” in an attempt to anthropomorphize the AI models and avoid calling their unreliability by name. AI tools also reproduce biases and views present in training data and of the people who fine-tune them. You should fact-check AI outputs and review them for bias before relying on them.
5. By default, material you enter into almost all public chatbots can be used to train those models. Uploading drafts, outlines, research, or interview transcripts may expose confidential material, may compromise your own future rights, and may conflict with a publisher’s exclusivity clause. Most consumer AI tools include a setting to opt out of training on your inputs, but in some cases you need to dig to figure out how to opt out, and it is not fully clear how those settings will be honored over time.
6. You are responsible for the manuscript that you deliver to a publisher or self-publish.

Categories of AI Use

Different uses of AI raise different concerns, and it is not useful to treat all AI use as ethically or professionally equivalent. Uses at one end of the spectrum, such as copying AI-generated outputs directly into text that is presented as professionally human-authored, create the greatest risks and ethical concerns; background uses, such as research where the writer is careful not to allow the AI to think for them or the outputs to enter their work, present the least. For instance, generating text and claiming it as your own writing without substantial revision or mimicking another writer’s voice or style are generally considered unethical and possibly plagiarism. AI-powered spelling and grammar checks, on the other hand, are ubiquitous and uniformly considered ethical and professional, as is the use of AI to research (if you are careful not to copy from AI outputs, just as you would prevent plagiarism from third-party sources).

Recommended Best Practices

1. If you are claiming authorship, don’t let AI replace your unique voice, thinking, and creativity. The words and thoughts should be your own. That is what authorship means.
2. If you use AI for research or as a sounding board, be careful to use it as background only and not copy AI-generated output into your work, literally or non-literally, just as you would take measures to prevent plagiarism from third-party sources. AI generates text based on its training inputs and as noted above, outputs may incorporate others’ copyrighted material.
3. You should thoroughly review and fact-check all AI outputs for the accuracy of facts and information provided by LLMs. Also, check for potential biases in the AI output, including gender, racial, socioeconomic, or other biases that could perpetuate stereotypes or misinformation.
4. Disclaim any AI-generated text in your work on your copyright registration application. If your publisher is filing a registration application on your behalf, be sure to inform your publisher of AI-generated uses so they can prepare an accurate application.
5. Ensure that any AI use in your manuscript does not violate your representations and warranties to your publisher in your contract. Authors warrant that their works are their original creations in the contract, and inclusion of AI-generated text may violate that warranty. Let your agent (if you have one) and publisher know up front that you have incorporated AI outputs and have a conversation with them about it. There are cases where using AI (such as a book about AI or a character using AI) makes sense and the publisher will approve it—in which case be sure your contract or an amendment reflects the approved AI use.
6. Some publishers are developing specific guidelines and rules around authors’ use of AI,apart from what is in your contract, so you should ask your editor if your publisher has any special guidance and carefully review the rules.
7. If you hire a ghostwriter, co-author, independent editor, book coach, or other person to work on your manuscript, be sure to enter into written contracts with third-party advisors that prohibit the addition of more than de minimis AI-generated text without your written approval and require that they disclose their use of AI.
8. Publicly disclose any substantial use of AI to generate text. Readers have the right to know whether the books they buy, borrow, and read were written by a human. Many readers care deeply about the human connection with authors. Buying or reading a book expecting it to be authored by a human only to find out that it was fully or partly generated by AI sows distrust between writers and readers, and harms the entire writing community. The disclosure of AI use can be made by describing the AI use in the front material of a book or at the bottom of an article.
9. Respect your fellow authors and do not use generative AI to purposely copy or mimic the unique styles, voices, or other distinctive attributes of other writers’ works in ways that harm the value of their works or attempt to profit from them. Apart from the ethical issues, mimicking a fellow writer’s unique voice or style could subject you to claims of unfair competition or copyright infringement.
10. Consider showing solidarity with professional creators in related fields, including voice actors/narrators, translators, illustrators, indexers, etc., who are also trying to protect their professions. Be aware of whether you’re displacing a human being when you use AI. When you need to use AI instead of hiring a human, look for AI programs that were trained on licensed content and not built on copyright infringement. Legally trained AI models do exist in the visual arts and for narration, for instance. We have recommended model contract clauses to prevent your publisher from using AI to generate translations, cover designs, or audiobook narration or to make editorial changes without your prior approval. We especially encourage authors to continue to work with and support human translators, who are part of our membership.
11. If you fine-tune an AI model on your own work to generate new material, understand that the fine-tuning is done on top of a foundational large language model that was likely trained and developed on mass copyright infringement, as well as the fact that uploading your work to an AI could enable your work to be included in a large dataset or could infringe your publishing contract giving your publisher exclusive rights. Fine-tuning is the process by which a smaller AI model is created on specific datasets with specific functionalities to work with a foundational LLM. Though it arguably raises fewer ethical concerns since the primary expression in the outputs is generated from the author’s own work rather than the work of others,  you should consider and disclose your use of AI to generate any text since you are still using an underlying LLM that was trained on books without authorization.