Shishene Jing argues that the fair use doctrine’s central question—whether a use is sufficiently transformative to avoid licensing requirements—breaks down when applied to AI. Transformativeness worked as a test because transformative uses rarely competed financially with the originals. AI training severs that relationship, as it is both the most transformative use of copyrighted material and the use best equipped to displace their markets.
For three decades after Judge Pierre Leval’s 1990 article “Toward a Fair Use Standard” and the Supreme Court’s 1994 decision in Campbell v. Acuff-Rose, the fair use doctrine effectively focused on one question: is the defendant’s use of another’s copyrighted material “transformative” enough that it can be used without the owner’s licensed permission? If the defendant’s work added new meaning or expression, courts would forgive substantial unlicensed copying. However, AI is making visible the weakness in this approach.
The underlying economic assumption of the fair use doctrine is that transformative works do not compete directly in the market of the works they transformed. Parodies don’t compete with the original, legal research databases like Lexis and Westlaw don’t compete with underlying legal briefs, and a snippet view of a book provided by Google is not a substitute for the book itself.
AI challenges this assumption. In the 2025 case Kadrey v. Meta, wherein a group of authors sued Meta for using their works to train its model, the presiding judge noted how AI is simultaneously one of the most dramatically transformative uses courts have ever evaluated and one of the most plausibly market-displacing. Follow up legal scholarship in the Journal of Copyright elaborated on how AI training “unsettles the traditional relationship between transformativeness under the first factor and market substitution under the fourth.” An LLM does not reproduce Marilynne Robinson’s prose “to crave and to have are as like as a thing and its shadow”; it produces statistical abstractions that generate genuinely new text. That is about as transformative as it gets. And yet, Meta’s model, trained on that corpus, can generate a novel that competes for the same reader’s attention and the same publisher’s advance. The court ultimately ruled for Meta, finding that although it believed market substitution could defeat fair use, such evidence was not provided.
AI challenges copyright law to reassess whether fair use should focus on transformativeness at all. Section 107 of the Copyright Act lists four factors without hierarchy. Transformativeness arises from the first factor, while the fourth factor concerns the impact on the market of the copyrighted work. The other two factors concern the nature of the copyrighted work—if it’s factual or creative, with factual works receiving less protection—and the proportion of the new work that borrows from the copyrighted work. This article argues that a focus on the fourth factor of market harm is more administrable, more predictable, and more consistent with recent Supreme Court precedent than a continued focus on the first factor of transformativeness, which is fundamentally a vague judgement of aesthetics and what makes a work novel.
Transformativeness was never the singular organizing principle of fair use. It functioned as a heuristic that worked only because transformative uses almost never displaced markets. AI exposes the heuristic by producing cases where transformation and market substitution increase together rather than inversely.
The two poles of fair use
Judge William Alsup called Anthropic’s book-training “exceedingly transformative” in the 2025 case Bartz v. Anthropic and moved briskly toward a fair-use finding exonerating Anthropic. Meanwhile, Judge Stephanos Bibas, applying the same four factors to similar conduct in Thomson Reuters v. Ross Intelligence, also decided in 2025, found no fair use because of direct market substitution. The two rulings echoed a long-running disagreement in copyright scholarship about what fair use is actually for.
Fair use scholarship has swung between two poles since the early 1980s that loosely map onto factors one and four of the fair use inquiry. Wendy Gordon articulated the first pole in her 1982 paper “Fair Use as Market Failure.” Gordon argued that fair use should exist only where transaction costs or other frictions prevent a licensing market from pricing the use of the copyrighted material. On her account, fair use is not a reward for creativity, but rather a stopgap for markets that cannot clear. Where no transaction would have occurred anyway, prohibition earns the owner nothing and destroys the value of the use. Examples include a teacher making last-minute photocopies of a text for unanticipated student registrants of their class. Contacting the copyright owner would be impractical and no market transaction would occur.
While the formation of a licensing market for a particular use of copyrighted material is conceptually distinct from whether that use causes market harm under factor four, the scholarly and statutory focus are closely related. The existence of an actual or reasonably developing licensing market can provide evidence that an unauthorized use causes market harm, particularly where the use substitutes for transactions that copyright owners would otherwise make. AI provides a useful example, as licensing markets for AI-related uses of copyrighted works have begun to develop alongside claims of market displacement. Conversely, personal videotaping of copyrighted television programs for private use, as in the 1984 case Sony v. Universal City Studios, caused little meaningful market harm precisely because such individualized use did not materially substitute for the market in television programming nor generate a licensing market for that use.
The second pole is Leval’s framework, which supplied the transformative-use vocabulary that has become dominant in copyright law. Again, scholarship loosely mapped onto the first statutory factor, which focuses on purpose and character of the use and asks whether the use is commercial. Leval argued that a fair use justifies itself by adding new meaning or purpose, independent of what it does to the original market. Leval’s approach became the standard for the next three decades. The first factor as well as fair use as a whole came to largely focus on the question of transformativeness. Gordon’s market-focused framework was cited respectfully and applied rarely. Leval’s framework was able to serve as the standard seemingly independently, largely because most fair-use defendants before 2024 were not actually capable of economically substituting for the markets from which they drew, rendering Gordon irrelevant.
AI makes this no longer the case and forces the choice. Judge Bibas’s opinion in Ross is, without saying so, a Gordon opinion: it asks whether a licensing market existed (Thomson Reuters had begun licensing headnotes for AI training). Judge Alsup’s opinion in Bartz, focused on transformativeness, is a Leval opinion.
How transformativeness became the organizing principle
The Supreme Court adopted Leval’s transformativesness-focused framework in Campbell, but Campbell itself is more modest than the fair use doctrine it spawned. Justice David Souter’s unanimous opinion held that hip-hop group 2 Live Crew’s parody of Roy Orbison’s “Oh, Pretty Woman” could be transformative because parody, by its nature, comments on the original. But the Court sent the case back to the lower court, because a transformative purpose did not settle the market question. Transformative purpose was one factor among four. Market harm remained an independent inquiry that did not evaporate simply because a use was creative or repurposed.
What followed was doctrinal drift, not faithful application. Circuit courts—especially the Second Circuit in cases like Cariou v. Prince (2013)—stretched “transformative” to cover minimal recontextualization: cropping, recoloring, resizing Patrick Cariou’s photographs. Market harm was treated as largely subsumed by the transformative-use finding on the theory that a transformative use, by its nature, does not serve as a substitute. That theory assumes its own conclusion. Whether a use is a substitute is an empirical fact about the market, not a logical result of the defendant having done something creative to the source material.
The Supreme Court’s 2023 decision in Andy Warhol Foundation v. Goldsmith began to pull the doctrine back. The Court rejected the idea that any new aesthetic or expressive overlay automatically renders a use transformative. Instead, it insisted that commercial uses sharing substantially the same purpose as the original are less likely to qualify for a fair use defense. In doing so, the Court signaled that transformativeness is not a free-floating aesthetic judgment capable of swallowing the rest of the fair-use analysis. The opinion did not abandon the concept, but it made clear that market realities—whether the new work competes in the same market or for the same audience—cannot be waved away by a finding of new meaning alone. That course correction arrived just as generative AI cases began forcing the issue into sharper relief.
The advantages of a market displacement standard
The fourth factor of the Copyright Act—“the effect of the use upon the potential market for or value of the copyrighted work”—is the one Congress described in market rather than aesthetic terms. It is the factor the Supreme Court in Harper & Row v. Nation Enterprises (1985) called “undoubtedly the single most important element of fair use,” a formulation Campbell narrowed but never repudiated.
The transformative-use juggernaut of the 2000s and 2010s effectively demoted Congress’s most concrete, most economically legible factor in favor of the vaguest one. Whether a use “adds something new, with a further purpose or different character” is a question courts answer through literary and aesthetic judgment, which varies enormously by panel and by circuit. Moreover, it is an aesthetic question that federal judges are not trained to answer.
Market displacement is not immune to line-drawing problems, but it is at least the kind of objective empirical question courts, economists, and licensing markets are equipped to answer with evidence. Does the transformed product compete in the same market as the original art? Does a functioning or emerging license market exist, and does the defendant’s use take for free what that market would have priced?
Reordering the test does not gut fair use—it makes fair use fair
A shift away from treating transformativeness as the principal factor of the fair use doctrine does not jeopardize the doctrine itself. Parody, criticism, scholarship, and genuine non-substitutive transformation remain as protected under a market-displacement-first framework as they were under Campbell, because none of those uses genuinely competes in the market for the original. A parody does not reduce demand for the work being parodied; if anything, it can increase it. A scholarly quotation does not substitute for the book being quoted.
What a market-displacement-first framework actually screens out is the narrower, more troubling category the transformative-use juggernaut had started shielding by accident: uses that are creatively transformative and commercially substitutive at the same time. Leval’s framework, built out of Campbell’s parody fact pattern, never could have contemplated that combination, because parody rarely substitutes for the market of the original.
Generative AI training is the first technology to occupy that combination at scale. A foundation model can be, in the same breath, the most transformative use of copyrighted text in the history of the doctrine and the most direct threat to the markets its data sources originally served. The doctrine’s job now should be to stop asking the aesthetic question first and start asking the economic one. Transformativeness was never truly capable of being the organizing principle of fair use. It was an evidentiary shortcut that happened to correlate with market effects until AI destroyed the correlation.
Author Acknowledgement: The author wishes to thank Judge Jed S. Rakoff for his comments.
Author Disclosure: The author reports no conflicts of interest. You can read our disclosure policy here.
Articles represent the opinions of their writers, not necessarily those of the University of Chicago, the Booth School of Business, or its faculty.
Subscribe here for ProMarket’s weekly newsletter, Special Interest, to stay up to date on ProMarket’s coverage of the political economy and other content from the Stigler Center.
