If you are signing a book deal, journal article agreement, or freelance publishing contract in 2026, the single most important new term to negotiate is the AI clause: the language that determines whether the publisher can feed your manuscript into large language models for training, whether they can license your work to third-party AI companies, and what you get paid if they do. This clause did not exist in standard contracts five years ago. Today it appears in nearly every major trade and academic publishing agreement, and authors who sign without negotiating it are giving away rights that may be worth more than their advance.
What an AI Clause Actually Is
Also worth reading: What is a publishing contract reversion rights checklist and how do authors get their rights back? · How do I effectively manage publishing contract negotiation redlines without losing the deal? · What should AI royalty clause template authors include in 2026 publishing contracts?
An AI clause is a contract provision governing how artificial intelligence may be used in connection with your work. It typically covers four distinct permissions. First, internal use: whether the publisher's staff or contractors can upload your manuscript to tools like ChatGPT, Claude, or Gemini for editing, marketing copy, translation, or proofreading. Second, training use: whether the publisher can include your text in datasets used to train machine learning models. Third, licensing use: whether the publisher can sell or license your content to AI developers such as Microsoft, OpenAI, or Anthropic, as Taylor & Francis reportedly did when it sold access to academic authors' research to Microsoft, leaving many of those authors, in the words of reporting by The Bookseller, 'shocked' because they had never been asked. Fourth, warranty and indemnity language: representations about whether your own manuscript was written with AI assistance, which increasingly trigger disclosure requirements and can void warranties if undisclosed.
These four permissions are separable, and that is the core negotiation insight most authors miss. You can grant internal-use rights while flatly prohibiting training and third-party licensing. A publisher does not need to train models on your novel to run a spell-check tool, and conflating these uses is how publishers quietly acquire far more than they need.
Why Publishers Are Pushing These Clauses Now
The economics explain the pressure. Since 2023, licensing deals between publishers and AI companies have moved from hypothetical to routine. The Associated Press signed one of the earliest first-mover agreements with OpenAI, and as The Wall Street Journal reported, other publishers began seeking similar payments while trying to build safeguards into their deals. Johns Hopkins University Press announced plans to license its books to train AI systems. Academic publishers discovered that back catalogs of peer-reviewed research are valuable training corpora, and subscription-fee disputes like Elsevier's confidentiality-clause defenses show how opaque these arrangements can become even to institutional customers.
Publishers argue that AI clauses let them participate in this revenue stream on your behalf, and some genuinely intend to share proceeds. But the default position in most boilerplate is that the publisher holds subsidiary rights broadly enough to cover AI licensing without separate consent or payment terms specific to AI. If your contract grants 'subsidiary rights' or 'digital rights' without carving out AI, you may already have signed away the ability to object. That is why the Authors Guild has issued statements urging members, including through international chapters addressing African authors and others, to review AI provisions before signature rather than after.
The Four Deal Structures Compared
When you sit down at the negotiating table, you will generally encounter one of four structures. Understanding them lets you name what you want instead of reacting to boilerplate.
| Feature | Blanket Prohibition | Consent-Based | Revenue Share | Silent Contract |
|---|---|---|---|---|
| Publisher may train AI on your work | No | Only with written consent per use | Yes, bundled | Ambiguous, likely yes |
| Third-party AI licensing | No | With consent | Yes, bundled | Ambiguous |
| Internal AI tool use (editing, marketing) | Often also banned | Usually permitted | Permitted | Permitted |
| Your compensation | None needed | Negotiated per deal | Percentage of licensing revenue, often 25–50% | Zero |
| Who bears risk | Author keeps control but may lose deal | Administrative burden | Depends on audit rights | Entirely on author |
| Common in | Literary fiction, agented trade deals | Hybrid and midlist | Academic and reference publishing | Legacy boilerplate, small presses |
Practical Steps Before You Sign
Start by requesting the full contract early, not at acceptance. Ask your agent or, if unagented, the acquisitions editor three direct questions in writing: Does this agreement permit the publisher to use my manuscript to train AI models? Does it permit licensing my work to third parties for AI development? What internal AI tools will be used on my manuscript, and will my text be uploaded to third-party services?
That third question matters more than authors realize. Reporting from The Bookseller cited literary agents describing editors uploading confidential manuscripts to ChatGPT to read quickly. Your unpublished manuscript is confidential material, and uploading it to a consumer AI service may violate both your reasonable expectations and any non-disclosure understanding, since an NDA is precisely the legal instrument defining how confidential material between two parties must be handled. Insist on a clause stating that neither party will submit the manuscript or any portion of it to third-party generative AI services without prior written consent.
Next, define the money. If you agree to any AI licensing permission, demand a specific royalty structure rather than folding AI into generic subsidiary-rights accounting. Reasonable asks based on emerging market practice range from 25% to 50% of net licensing receipts for AI-specific deals, with some agented authors securing flat fees per licensed title. Also require annual reporting of AI licensing activity and audit rights mirroring those in your royalty clause. Without reporting obligations, a revenue-share promise is unverifiable, and confidentiality clauses on the publisher side, as the Elsevier dispute illustrated, can make it impossible to confirm what happened to your work.
Finally, address the warranty side. Increasingly contracts require authors to warrant that the work was not generated by AI, or to disclose AI-assisted portions. Disclose accurately. A false warranty can void your indemnity protections and expose you to contract termination. If you used AI for research assistance or line editing, say so in a disclosure letter rather than hoping nobody asks.
Common Mistakes Authors Make
The first mistake is treating AI clauses as non-negotiable boilerplate. Editors expect pushback; agents report that AI language is now among the most negotiated terms alongside audio rights. A blanket refusal to discuss it marks you as uninformed, not principled.
The second mistake is banning all AI use including internal tools. A publisher whose copyeditors cannot use grammar software, or whose marketing team cannot draft alt-text with assistance, may simply route around your restriction informally, which is worse than a clear rule. Distinguish human-in-the-loop productivity uses, which are low-risk, from corpus ingestion and third-party licensing, which are where value leaks.
The third mistake is ignoring reversion triggers. If your publisher licenses your book to an AI company and the resulting model outputs passages resembling your prose, what recourse do you have? Ask for a clause requiring the publisher to flow down attribution and opt-out requests to AI licensees where technically feasible, and to notify you of licensing deals within thirty days. Perfect enforcement is unrealistic given how opaque model training is, but contractual notice rights create leverage and evidence.
The fourth mistake is assuming copyright law will save you. In the United States, the Copyright Clause in Article I, Section 8 gives Congress power over intellectual property, and ongoing litigation will shape fair-use doctrine for training data, but litigation takes years and outcomes remain unsettled. A contract term you agreed to will defeat almost any statutory argument you might otherwise raise. Contracts govern where statutes are silent or ambiguous, so the negotiation table is the venue that matters.
When to Act and How Fast
Act at offer stage, not signature stage. Once you have a deal memo, send your AI-term questions within a week; delays signal indifference and compress your negotiating window against publication deadlines. If you have already signed a contract with broad digital or subsidiary rights, you still have options: request an amendment or side letter specifically addressing AI, especially if the contract predates 2023 and could not have contemplated these uses. Publishers under reputational scrutiny, particularly university presses facing faculty criticism over deals like Johns Hopkins' licensing plan, sometimes agree to amendments to avoid controversy.
For previously published works, monitor the market. Watch for announcements that your publisher has entered AI licensing arrangements, and join collective efforts. The Authors Guild and equivalent bodies in other countries have pushed for opt-in frameworks and shared bargaining power; individual authors acting alone rarely move a major publisher, but coordinated objections documented in public statements have changed contract templates.
Costs, Fees, and What Representation Runs
Negotiating costs vary widely. If you are agented, AI-clause negotiation is included in your standard 15% commission on the deal; there is no separate fee. Unagented authors hiring a publishing attorney for a contract review typically pay between $500 and $2,500 for a flat-fee review of a trade publishing agreement as of 2026, with AI-language redlining adding a few hundred dollars. An AI publishing consultant, offering specialized review of AI-specific terms across multiple contracts, generally charges $300 to $800 per contract or monthly retainers for prolific authors. Weigh this against the potential value: a single AI licensing deal for a substantial backlist can be worth five figures per title, and Taylor & Francis-style transactions show publishers will transact at scale when permitted.
There is also a cost to over-negotiating. Demanding a total AI ban on a commercial nonfiction project where the publisher plans an AI product line may cost you the deal entirely. Know your leverage: bestselling authors, authors with competitive auctions, and owners of high-value backlists have real power; debut novelists usually secure prohibitions only on training and third-party licensing while conceding internal-use rights.
The Bottom Line
An AI clause determines who profits when machines learn from your writing. The strongest position combines three elements: an express prohibition on training and third-party AI licensing unless separately negotiated, a carve-out permitting low-risk internal tool use with confidentiality protections, and, where you do permit licensing, defined revenue shares of 25% to 50% with reporting and audit rights. Read every contract as if the AI clause were the whole deal, because economically it may soon be. Silence benefits whoever wrote the document, and in publishing, that person is not you.