Feist v. Rural Telephone: Originality, Sweat of the Brow, and AI

In Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991), the Supreme Court unanimously held that copyright protection requires a minimal degree of creativity, and that a garden-variety white pages telephone directory does not have it. Justice O’Connor’s opinion killed the “sweat of the brow” theory that had let compilers claim copyright based on effort alone, and it established that facts themselves are never copyrightable, no matter how much work went into gathering them.¹

What the Dispute Was About

Rural Telephone Service Company was a certified public utility serving several communities in northwest Kansas. State regulations required every telephone company to publish an annual directory, so Rural produced a standard book of white and yellow pages as a condition of its monopoly franchise.¹

Feist Publications produced area-wide directories covering a much larger region than any single utility served. It approached all eleven phone companies in northwest Kansas and offered to license their white pages. Ten agreed. Rural refused. Feist then extracted roughly 4,000 listings from Rural’s published book without permission. When the two directories were compared, 1,309 of the 46,878 listings in Feist’s book matched Rural’s, including four fictitious listings Rural had planted to catch copiers.¹

Rural sued for copyright infringement and won at both the District Court and the Tenth Circuit, each relying on older cases treating directories as copyrightable because of the effort behind them.² The Supreme Court took the case to test whether that reasoning could survive.

The Holding: Originality Is a Constitutional Floor

Justice O’Connor, writing for eight justices with Justice Blackmun concurring in the judgment, grounded the originality requirement in the Constitution itself. Article I, Section 8, Clause 8 authorizes Congress to secure rights for “Authors” in their “Writings,” and the Court read both words to require creative human authorship.³

Originality has two parts. The work must be independently created by the author, and it must possess at least some minimal degree of creativity. The bar is low but real.¹ Facts fall below it because they are discovered, not authored. A person’s name, town, and phone number exist regardless of who writes them down. Recording a fact is not creating one.

Statute reinforces the point. Under 17 U.S.C. § 102(b), copyright never extends to ideas, procedures, processes, or discoveries, no matter how they are expressed.⁴ Authors can own only the creative expression layered on top of facts, never the facts themselves.

Rural’s white pages failed on every count that matters. The selection of who to include was dictated by state law. The coordination of name, town, and number was mechanical. The alphabetical arrangement was, in the Court’s words, “an age-old practice, firmly rooted in tradition” with no creative spark at all.¹

The End of Sweat of the Brow

Before Feist, many lower courts followed a doctrine called “sweat of the brow.” The idea was simple: if you worked hard to compile facts, you earned copyright as a reward for that labor, and copying your data was tantamount to stealing your investment.¹

The Supreme Court said this reasoning “flouted basic copyright principles.” Copyright does not exist to reward effort. Its constitutional purpose is to promote the progress of science and useful arts by giving authors an incentive to create original works. Protecting facts based on labor alone would let the first compiler of a dataset lock everyone else out of the underlying information, which runs against the design of the system.²

The practical result freed publishers, researchers, and businesses to reuse publicly available factual data. If your compilation shares only the underlying facts with someone else’s, you have not infringed. The first compiler’s investment may have been real, but it does not translate into ownership of the information.

When Compilations Still Get Protection

Feist did not hold that all compilations are uncopyrightable. The Copyright Act defines a compilation as a work formed by collecting and assembling preexisting materials or data that are selected, coordinated, or arranged in a way that makes the resulting work original.⁵ Three choices carry the creative weight:

Selection: which data to include and which to leave out.
Coordination: how to link related pieces of data together.
Arrangement: how to order and organize the final product.

If any of those choices reflects genuine creativity, the compilation as a whole can qualify. Yellow pages illustrate the contrast with Rural’s white pages. In BellSouth Advertising & Publishing Corp. v. Donnelley Information Publishing, a federal appeals court found that yellow pages involved real creative decisions: the publisher chose the geographic boundaries for each directory, created its own business classification headings, and decided how to fit each business under the appropriate category.⁶ Those choices went beyond mechanical alphabetizing.

Even where a compilation qualifies, the protection is narrow. Under 17 U.S.C. § 103(b), copyright in a compilation extends only to the material contributed by the compiler, not to the preexisting facts.⁷ Lawyers call this “thin copyright.” A competitor can use all the same underlying facts. What they cannot do is replicate the specific creative choices the compiler made in selecting and organizing them.

How Database Owners Protect Data After Feist

Feist‘s rule that facts are free to copy pushed database creators toward other legal tools. Copyright will not protect the raw data itself, so owners typically combine several other approaches.

Trade Secret Law

Under the federal Defend Trade Secrets Act, a compilation of data can qualify as a trade secret if the owner takes reasonable measures to keep it confidential and the data derives independent economic value from not being publicly known.⁸ This depends on secrecy. A proprietary customer database restricted to authorized employees can be protected; a published telephone directory cannot. The moment data becomes publicly available, trade secret protection evaporates.

Contract Law

Terms of service create another layer. When a user creates an account and clicks “I agree” to terms prohibiting copying or scraping, that agreement is enforceable, and copying data afterward can support a breach of contract claim even where the data is uncopyrightable. Clickwrap agreements requiring active consent hold up in court. Browsewrap terms buried in a footer link, without any affirmative agreement, are harder to enforce, and courts have been reluctant to bind scrapers to terms they never saw.

The Computer Fraud and Abuse Act

The federal Computer Fraud and Abuse Act prohibits accessing a computer “without authorization.” Some database owners tried to use the statute against scrapers, but recent decisions have narrowed that option. In Van Buren v. United States (2021), the Supreme Court held 6–3 that someone “exceeds authorized access” only by reaching areas of a computer system that are off-limits, like files behind a password.⁹ The Ninth Circuit then applied that logic in hiQ Labs v. LinkedIn, concluding that scraping data from public-facing web pages likely does not violate the CFAA because there is no gate to bypass.¹⁰ If a website shows information to anyone with a browser, automated collection of that same information is not “unauthorized access.”

Why Feist Matters for AI-Generated Content

Feist‘s originality requirement has taken on new weight as AI-generated material spreads. The Copyright Office has applied the same logic to draw a clear line: copyright protects only works of human authorship. Content generated by a machine without creative human input is not registrable, for the same reason a mechanical alphabetical list was not. If no human made the creative choices, there is no “author” in the constitutional sense.¹¹

Works that mix human and AI contributions get evaluated piece by piece. If a person selects and arranges AI-generated material in a sufficiently creative way, that selection and arrangement can be protected, much like a creative compilation of facts, while the AI-generated portions themselves must be disclaimed. The Copyright Office has applied this framework in several registration decisions since 2023, consistently requiring applicants to identify which parts were AI-generated.¹²

A separate question is whether training an AI model on copyrighted works infringes. Some developers have argued that extracting patterns from text is like extracting facts from a phone book: the model learns information, not expression. The Copyright Office’s 2025 report on generative AI training pushed back, noting that language models absorb not just the meaning of words but how they are “selected and arranged at the sentence, paragraph, and document level,” which is “the essence of linguistic expression.” Whether training qualifies as fair use remains unsettled, with dozens of lawsuits pending.¹³ The line Feist drew between unprotectable facts and protectable expression sits at the center of that fight, thirty-plus years after a Kansas phone book put it there.

1
Legal Information Institute. Feist Publications, Inc. v. Rural Telephone Service Co., Inc.
2
Justia. Feist Publications, Inc. v. Rural Tel. Serv. Co., 499 U.S. 340 (1991)
3
Legal Information Institute. Authorship, Writings, and Originality
4
Office of the Law Revision Counsel. 17 USC 102 – Subject Matter of Copyright: In General
5
Office of the Law Revision Counsel. 17 USC 101 – Definitions
6
Justia. Bellsouth Advertising and Publishing Corporation v. Donnelley Information Publishing, Inc.
7
Office of the Law Revision Counsel. 17 USC 103 – Subject Matter of Copyright: Compilations and Derivative Works
8
Office of the Law Revision Counsel. 18 USC Chapter 90 – Protection of Trade Secrets
9
Supreme Court of the United States. Van Buren v. United States, 593 U.S. 374 (2021)
10
United States Court of Appeals for the Ninth Circuit. hiQ Labs, Inc. v. LinkedIn Corp.
11
U.S. Copyright Office. Works Containing Material Generated by Artificial Intelligence
12
U.S. Copyright Office. Copyright and Artificial Intelligence
13
U.S. Copyright Office. Copyright and Artificial Intelligence Part 3: Generative AI Training