The Fair Use Database

Federal fair use, holding by holding, quote by quote

Search the cases. Ask the corpus.

About

The Fair Use Database codes every substantive federal fair use opinion, from the doctrine's early antecedents through July 8, 2026. Search verified, pin-cited court language directly or ask Folsom, a research assistant grounded only in the coded corpus. Every Folsom answer cites its evidence and links back to the underlying case records. The coding is validated against 453 human-coded analyses.

How to use it

1 Search the courts' words by language, case, court, factor, or outcome.
2 Ask Folsom for cited answers and corpus charts drawn only from the coded data.
3 Follow and organize the authorities through case-level deep links, My Cases sets, and copy-ready citations.

Signing in records your name and email address, and the site logs your searches and the pages you view, so we know who uses the database and how. Aggregate usage statistics are collected with Google Analytics; nothing is sold or otherwise shared.

FUD Search Ask Folsom My Cases Statistics Methods

Ask Folsom Beta

Research question Compare Warhol and Google on transformative purpose.

Grounded answer

Both decisions make the inquiry use-specific. Google examines copied code in a new computing environment; Warhol focuses on the purpose of the challenged magazine license.

Google LLC v. Oracle Am., Inc., 593 U.S. 1, 28–30 (2021). Andy Warhol Found. for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508, 526–31 (2023).

Case comparison 2 controlling opinions

Open cited cases ↗

Cited answers Charts with Bluebook citations Case-level deep links

Folsom answers only from the coded corpus. Verify linked authorities before relying on them.

Landmarks in the database

Sony Corp. v. Universal City Studios Fair use

464 U.S. 417 · U.S. · 1984

“…this record amply supports the District Court’s conclusion that home time-shifting is fair use.”

Harper & Row v. Nation Enters. Not fair use

471 U.S. 539 · U.S. · 1985

“The Nation’s use had not merely the incidental effect but the intended purpose of supplanting the copyright holder’s commercially valuable right of first publication.”

Campbell v. Acuff-Rose Music, Inc. Remanded

510 U.S. 569 · U.S. · 1994

“The more transformative the new work, the less will be the significance of other factors, like commercialism…”

Google LLC v. Oracle Am., Inc. Fair use

593 U.S. 1 · U.S. · 2021

“These and related facts convince us that the ‘purpose and character’ of Google’s copying was transformative…”

Andy Warhol Found. v. Goldsmith Not fair use

598 U.S. 508 · U.S. · 2023

“AWF’s use of Goldsmith’s photograph does not target the photograph, nor has AWF offered another compelling justification for the use.”

Authors Guild v. Google, Inc. Fair use

804 F.3d 202 · 2d Cir. · 2015

“While Google makes an unauthorized digital copy of the entire book, it does not reveal that digital copy to the public.”

Hachette Book Grp. v. Internet Archive Not fair use

115 F.4th 163 · 2d Cir. · 2024

“IA’s Free Digital Library serves as a satisfactory substitute for the original Works.”

My cases

Folders are saved to your account. Save cases from any case page or search result.

Ask Folsom beta

Folsom answers only from the coded corpus (release v0.3.8, opinions decided through July 8, 2026) and cites every claim. AI responses may include hallucinations. Verify quotes and holdings against the linked opinions before relying on them. Research findings, not legal advice.

Loading…

Data & methods

This database codes every federal court opinion that substantively analyzes copyright fair use under 17 U.S.C. § 107. Each opinion is decomposed into work/use units; each unit carries factor-by-factor coding with direction, weight, doctrinal components, and verbatim quotes tied to page numbers in the underlying opinion.

Is this dataset suitable for your study?

Population	Federal fair use opinions, early antecedents through July 8, 2026, screened from 3,446 candidates drawn from a Lexis+ boolean union reconciled against CourtListener sweeps, the Copyright Office Fair Use Index, and a merits seed set. Verified superset of the Beebe (1978–2019) and UCI FUJP (2019–2026) case lists.
Unit of analysis	The work/use pairing: one court's fair use analysis of one copyrighted work or homogeneous work family against one challenged use. 2,454 units across 1,659 substantive opinions; 252 substantive opinions carry opinion-level metadata but no codable unit.
Cohorts	decisive_merits (744): fair use decided on the merits. substantive_other (905): substantive fair use analysis without a decisive merits holding. uncertain (10): screening could not classify. excluded (1,787): non-substantive; listed in the screening ledger only.
Coding depth	Per unit: holding outcome, scope, controlling status, final merits status. Per factor: direction, directional score, stated weight, 34 component codes with polarity. Plus factor relationships, alternative grounds, procedural posture, motion outcomes, and 30,828 selected verbatim quotes.
Validation	453 human-coded gold units: 97.4% holding-outcome agreement (95% CI 95.5–98.6, κ 0.950), 94.7% factor-direction agreement (CI 93.4–95.8, κ 0.911), polar direction disagreement 0.29% (CI 0.11–0.74). Corpus-scale test-retest reliability separately measured. Every machine-human disagreement is documented in a queue; none is silently resolved.
Provisional share	1,136 of 2,454 units (46.3%) carry machine-proposed boundaries flagged provisional.
Release	v0.3.8, coverage through 2026-07-08.
Formats	Interactive site and API today. Frozen downloads (SQLite and unit-level CSV) with DOI and checksums arrive with the release layer now in progress.

About Folsom

Ask Folsom is a research assistant grounded exclusively in this coded corpus. It answers by running the same searches and counts available on this site — quotation search, case lookup, and corpus statistics — and every citation it offers is issued by the server from the records those queries actually returned. Folsom cannot cite anything outside the corpus, and citations it has not been issued are discarded before display.

Its limits: Folsom is a language model and may still misread or misconnect the evidence it retrieves ("hallucinate"), so verify quotes and holdings against the linked opinions before relying on them. Its knowledge ends at the corpus coverage date; it does not know later decisions. Statistics it reports carry their denominator, missing-value count, and release identifier, and exported charts embed a suggested citation with the date accessed. Folsom offers research findings, not legal advice.

Screening

The corpus was screened from a wider sweep of candidate opinions. The coded database covers the substantive cohorts; a screening ledger records every screened opinion with its cohort assignment, so the funnel from candidate pool to coded corpus is auditable end to end.

Coding is model-assisted with mechanical validation: every quote is verified verbatim against the source text, and a statistically meaningful sample is validated against human gold coding. Provisional codings are flagged as such.

Court opinions are works of the United States government and are not subject to copyright. Quotations shown here come from the public record.