Every author who’s ever quoted a few lines from another book in a review, used a song lyric as a chapter epigraph, or written a parody has relied on fair use. Most of them didn’t give it a second thought. It’s the part of copyright law that says the rules aren’t absolute, that sometimes, using someone else’s creative work without permission is perfectly legal.
For most of publishing history, fair use was a quiet legal backwater. Authors, critics, and educators invoked it routinely, courts sorted out the occasional dispute, and the doctrine hummed along in the background. Then AI companies started feeding millions of copyrighted books into their models, and fair use became the most contested legal question in a generation.
What Fair Use Means
Fair use is a legal defense built into U.S. copyright law that permits limited use of copyrighted material without the copyright holder’s permission. It exists because copyright was never meant to lock creative works away completely. Society benefits when people can quote books in reviews, teachers can photocopy articles for class, and comedians can parody pop songs. Fair use is the mechanism that makes those uses legal.
The key word is “defense.” Fair use doesn’t prevent you from being sued. It’s what you argue after someone accuses you of infringement. There’s no bright-line rule that says “you can use X number of words” or “quoting a paragraph is always fine.” Every case gets weighed on its own facts, which makes fair use both flexible and genuinely uncertain. Two lawyers can look at the same situation and reach opposite conclusions, and both can be right until a judge says otherwise.
A Judge Named Story
The American doctrine of fair use has an origin story worth knowing, partly because it involves a judge whose name reads like a novelist’s invention.
In 1841, Justice Joseph Story (yes, really) was riding circuit in Massachusetts when he heard a copyright case called Folsom v. Marsh. The dispute was prosaic: a man named Charles Upham had copied 353 pages of letters from a twelve-volume biography of George Washington to use in his own two-volume work. The question was whether this borrowing crossed the line from acceptable use into infringement.
Story ruled against Upham. But in the process of explaining his decision, he laid out a framework for evaluating when borrowing is permissible. He told courts to consider the nature of the original work, how much was taken, why it was taken, and whether the borrowing would hurt the market for the original. That framework, articulated in a case about a dead president’s mail, would quietly govern American copyright for the next 135 years.
The doctrine remained unwritten law, passed from judge to judge through precedent, until Congress codified it in Section 107 of the Copyright Act of 1976. The statutory language borrowed directly from Story’s reasoning, establishing four factors that courts must weigh in every fair use case. Congress deliberately left the standard flexible, noting that fair use needed room to adapt to technologies and circumstances that hadn’t been invented yet. They couldn’t have imagined how prescient that decision would turn out to be.
The Four Factors
When a court evaluates a fair use claim, it weighs four factors. No single factor is decisive, and the analysis is always case-by-case, which is why fair use disputes can be so maddeningly unpredictable.
The purpose and character of the use. Why did you use the copyrighted material, and did you transform it into something new? A book review that quotes a passage to critique it looks very different from a website that republishes entire chapters to drive ad revenue. Courts look for “transformative” use, meaning the borrowed material serves a new purpose rather than substituting for the original. Parody, criticism, commentary, education, and research all tend to fare well here. Straight copying for commercial gain does not.
The nature of the copyrighted work. Creative works (novels, poems, songs) get stronger protection than factual ones (news articles, technical manuals). An unpublished work also gets more protection than a published one, because the author hasn’t yet exercised their right to share it with the world.
The amount and substantiality of the portion used. This isn’t just about word count. Taking a single sentence can weigh against fair use if that sentence is the “heart” of the work, its most distinctive or memorable element. Conversely, copying an entire work can sometimes qualify as fair use if the purpose demands it (as happened in the Google Books case, discussed below).
The effect on the market. Many courts have called this the most important factor. If your use serves as a substitute for the original, pulling readers or revenue away from the copyright holder, that weighs heavily against fair use. Courts also consider whether the use, if it became widespread, would harm potential licensing markets the copyright holder might reasonably develop.
For authors, these factors come into play more often than you might expect. Quoting other authors in your blog posts, using real brand names in your fiction, creating parodies or commentary, even including song lyrics as epigraphs (though that last one is riskier than most authors realize) all involve fair use calculations.
The “Transformative” Twist
For decades after the Copyright Act passed, the first factor turned largely on one question: is the new use “transformative”? The concept came from a 1994 Supreme Court case, Campbell v. Acuff-Rose Music, in which 2 Live Crew’s raunchy parody of Roy Orbison’s “Oh, Pretty Woman” was found to be fair use. The Court held that the more transformative a use, the less other factors (like commercialism) matter. Lower courts took this and ran with it, broadly interpreting “transformative” to mean anything that conveyed a new meaning or message.
Then, in 2023, the Supreme Court pumped the brakes. In Andy Warhol Foundation v. Goldsmith, the Court ruled 7-2 that Andy Warhol’s silkscreen prints of the musician Prince, based on a photograph by Lynn Goldsmith, were not fair use when licensed to a magazine for the same purpose as the original photograph: illustrating an article about Prince. Justice Sotomayor’s opinion made a crucial distinction: adding new artistic meaning isn’t enough. The use has to serve a fundamentally different purpose. If the original and the new work compete in the same market for the same function, “transformative” won’t save you.
This ruling matters far beyond the art world. It reshapes how courts evaluate every fair use claim, including the ones now piling up around AI.
Fair Use Meets Artificial Intelligence
This is where things get complicated, consequential, and genuinely unresolved.
The large language models behind tools like ChatGPT, Claude, and Sudowrite were trained on enormous datasets that included copyrighted books, articles, and web pages, often without the authors’ knowledge or consent. AI companies argue this training qualifies as fair use because it’s transformative: the model doesn’t store or reproduce the books, it extracts statistical patterns and converts them into mathematical weights. The books go in as literature. What comes out is a prediction engine. Different purpose, different product.
Authors and publishers see it differently. The training data argument, they say, is just a fancy way of saying “we copied millions of books to build a product that competes with the people who wrote them.” The scale is unprecedented, the purpose is commercial, and the output directly threatens authors’ livelihoods by generating text that competes in the same markets as the originals.
Courts are only beginning to weigh in. In February 2025, a Delaware court ruled in Thomson Reuters v. Ross Intelligence that an AI legal research tool trained on Westlaw’s headnotes was not fair use, because it directly competed with the product it had copied from. But the judge made a point of noting that generative AI might have a stronger case, because the outputs are further removed from the inputs.
Four months later, two California rulings went the other way. In Bartz v. Anthropic, Judge William Alsup called AI training “exceedingly transformative,” comparing it to teaching a child to read. A similar ruling in Kadrey v. Meta reached the same conclusion. But both cases came with a caveat that should matter to every author paying attention: the judges found the training process could be fair use, but downloading books from pirate sites was a separate problem entirely. Anthropic settled for roughly $1.5 billion, paying about $3,000 per book for nearly half a million titles sourced from shadow libraries.
The biggest case is still pending. The New York Times v. OpenAI, filed in late 2023, alleges that ChatGPT can reproduce near-verbatim passages of Times articles, effectively serving as a substitute for the newspaper’s paywalled content. That case, along with a dozen others consolidated before a single federal judge, could produce the definitive ruling on whether AI training is fair use. No decision is expected before summer 2026.
In May 2025, the U.S. Copyright Office weighed in with a 108-page report that took a carefully middle position: some uses of copyrighted works for AI training will qualify as fair use, and some won’t. Where AI systems compete with or diminish licensing opportunities for original creators, the market-harm factor is likely to weigh against fair use. Where the use is genuinely analytical and non-expressive, fair use may apply. The only certainty, for now, is that there is no certainty.
Why This Matters for Your Writing Life
Fair use sits at the center of several questions that will shape the future of writing and publishing.
Your books may be part of the debate. If you’ve published a book, there’s a real chance it was included in training data for one or more AI models. The legal battles being fought right now are essentially about whether that use was permissible, and the outcomes will determine whether authors receive compensation, opt-in rights, or nothing at all. Organizations like the Authors Guild are actively representing authors’ interests in these cases, and the settlements and rulings that emerge will set precedents for years.
Fair use protects your work as an author, too. The same doctrine that AI companies invoke to justify training is the one that lets you quote other authors in your craft blog, write book reviews that include excerpts, create parodies, and reference real works in your fiction. Understanding how the four factors work helps you make smarter decisions about what you can safely include in your own writing without needing to call a lawyer every time.
It shapes the tools you use. The legal status of AI training data directly affects the tools available to you. If courts rule broadly in favor of fair use, AI companies will continue training on copyrighted works largely as they have been. If courts impose restrictions, companies will need to license training data or use only permissioned material, which could change the quality, capability, and cost of the writing tools built on those models. Tools like Sudowrite, which trains its Muse model exclusively on consented works, have already positioned themselves on one side of this line.
The law is still being written. Fair use has always been a doctrine that adapts to new technology. It adapted to photocopiers, VCRs, search engines, and now it’s adapting to AI. The principles Justice Story articulated in 1841 are being applied to situations he couldn’t have imagined, which is exactly what Congress intended when they kept the statute flexible. For authors, the practical takeaway is to stay informed. The rulings coming in the next few years will define what “fair” means in an age when a machine can read every book ever published in a matter of days.