Summary. Fair use is an affirmative defense you raise after being sued, which means the time to analyze it is before you copy. This checklist walks the analysis in the order courts actually apply it, starting with the threshold questions most people skip: whether the material is protected at all, and whether a specific statutory exemption applies before you reach § 107. It then works through each of the four factors with the questions that decide them, incorporating the reframing of the first factor in Warhol v. Goldsmith, which asks about the specific use rather than the new work as a whole. Later phases cover design mitigations that improve every factor, documentation and privilege, the decision to license anyway, and what to do if a demand letter arrives. Each phase is paired with the reason it matters and the primary authority behind it.

What this checklist is for. Running a defensible fair use analysis before you publish, ship, or train — and creating the written record that makes the position credible if it is ever challenged. It is not a scoring system. Fair use is decided holistically, and no checklist substitutes for judgment about a specific use. For the doctrine in depth, see Fair Use After Warhol.

Phase 1 — Threshold: do you even need fair use?

  • Confirm the material is protected by copyright at all. Facts, ideas, procedures, processes, systems, and methods of operation are outside copyright under 17 U.S.C. § 102(b).
  • Confirm it is not too short or unoriginal to protect — titles, names, short phrases, and blank forms are not protectable.
  • Check whether it is in the public domain by expiration, by failure of pre-1989 formalities, or as a U.S. Government work under § 105.
  • Check whether it is a government edict (statutes, regulations, judicial opinions, and works produced by legislators in their legislative capacity).
  • Confirm the copyright owner and whether you already have a license — including an existing enterprise license, a Creative Commons license, or an open source license that permits the use.
  • Check for a specific statutory exemption before reaching § 107: § 108 (libraries and archives), § 109 (first sale), § 110 (classroom and religious performances), § 117 (computer program copies by owners), § 120 (architectural works), § 121 (accessible formats).
  • Identify whether a contract governs the material (terms of service, database license, subscription agreement). Fair use is not a defense to breach of contract.

Why this matters. Fair use is the last resort, not the first analysis. If the material is unprotected or a specific exemption applies, you have a cleaner defense and you avoid a four-factor argument entirely. Saying "fair use" when you did not need it hands the other side a concession that copying occurred.

Phase 2 — Define the use with precision

  • Write the one-sentence use statement: "We will copy X, in quantity Y, in order to Z, and distribute it to audience A through channel B."
  • Identify the specific act of copying at issue — not the project, the act. Warhol holds that § 107 asks about "the purpose and character of the use," and that "the same copying may be fair when used for one purpose but not another."
  • Separate acquisition (how you obtained the copy) from use (what you do with it). They are analyzed independently, and acquisition from an unlawful source is a separate problem.
  • Identify every downstream use: internal only, customer-facing, marketing, licensed to third parties, embedded in a product.
  • Note who is doing the copying — nonprofit educational, news organization, commercial entity — and whether the actual user is you or your customers.

Why this matters. Every factor is applied to a defined use. An analysis of "our AI product" produces no answer; an analysis of "copying 4,000 lawfully purchased books to train a model whose outputs are short summaries" produces one. Vagueness here is the most common defect in fair use memos.

Phase 3 — Factor one: purpose and character

  • State the purpose of the original and the purpose of your use, side by side. Are they the same?
  • Ask the Warhol question: would your use substitute for the original for the same audience in the same market?
  • Identify what is added: new expression, meaning, message, function, or context — and whether it changes what the work does, not merely how it looks.
  • Ask why you need this particular work. A use that comments on the work (parody) justifies borrowing; a use that borrows to comment on something else (satire) must justify itself.
  • Assess commerciality honestly, and recognize it is not dispositive. Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569 (1994).
  • Check whether the use falls within a preamble purpose: criticism, comment, news reporting, teaching, scholarship, or research. Helpful, not decisive.
  • For software: is the copying intermediate, necessary to reach unprotected functional elements, with none of the copied expression in the shipped product? Sega Enterprises Ltd. v. Accolade, Inc., 977 F.2d 1510 (9th Cir. 1992).
  • For interfaces: is the copied material far from the core of copyright because its value derives from third parties who learned it? Google LLC v. Oracle America, Inc., 593 U.S. 1 (2021).

Why this matters and the trap. After Warhol, "we made it different" is not enough. The Court balanced the degree of difference against the commercial nature of the use and asked whether the two competed in the same licensing market. A style change that leaves the work serving the same function for the same audience is a weak first factor, however creative.

Phase 4 — Factor two: nature of the copyrighted work

  • Classify the work: highly creative (novel, song, photograph, film) or primarily factual/functional (manual, compilation, data, news report, declaring code).
  • Confirm whether the work was published. Copying from an unpublished work weighs against fair use. Harper & Row Publishers, Inc. v. Nation Enterprises, 471 U.S. 539 (1985).
  • Separate the protected expression from the unprotected facts and ideas within the work, and confirm you are only relying on fair use for the protected part.

Why this matters. Factor two rarely decides a case but is cheap to win. Where the material is functional, conventional, or interoperability-driven, say so with specifics; that framing did real work for Google in Oracle and it transfers to many technical contexts.

Phase 5 — Factor three: amount and substantiality

  • Quantify what you took, both absolutely and as a proportion of the whole work.
  • Ask whether you took the "heart of the work" — the most valuable, most recognizable, or most newsworthy part. Quality beats quantity. Harper & Row (roughly 300 words of a 200,000-word memoir).
  • Justify the amount by reference to the purpose: what is the minimum needed to accomplish it?
  • If you took the whole work, document why that was necessary (search indexing, reverse engineering, time-shifting, accessibility) and how the output limits substitution.
  • Test whether a smaller take would serve the purpose. If it would, take less.

Why this matters. Whole-work copying is not fatal — Sony, Kelly, Perfect 10, and Authors Guild all involved it — but it requires an articulated necessity. Courts respond to engineering and editorial necessity; they do not respond to convenience.

Phase 6 — Factor four: market effect

  • Identify the actual market for the original: how does the owner make money from it?
  • Identify the owner's licensing markets, including derivative works. Does the owner license uses like yours? Is there evidence of an established market?
  • Ask the substitution question: would a reasonable consumer use your version instead of buying or licensing the original?
  • Apply the widespread-use test: if everyone did this, what happens to the market for the original? Harper & Row.
  • Distinguish cognizable harm from harm caused by criticism. The market for disparaging reviews and parody is not a market the owner is entitled to control. Campbell.
  • For AI and large-scale uses, assess market dilution — whether outputs compete with the class of works in the training corpus even without reproducing any of them.
  • Check whether your use removes or alters copyright management information, which is a separate claim under 17 U.S.C. § 1202 regardless of fair use.

Why this matters. Most close cases are decided here. The licensing-market inquiry is circular unless bounded, and courts bound it to "traditional, reasonable, or likely to be developed markets." American Geophysical Union v. Texaco Inc., 60 F.3d 913 (2d Cir. 1994). The strongest defense evidence is that the owner does not license uses like yours and that your output points at the original rather than replacing it.

Phase 7 — Design mitigations

  • Reduce the take: shorter excerpts, lower resolution, thumbnails, snippets with length caps.
  • Add the commentary that makes the transformative purpose visible on the face of the work.
  • Point at the source: links, citations, and click-through to the original improve factor four measurably.
  • Cap retrieval: blacklisting, per-user limits, and prohibitions on reconstructing the whole work — the design features Judge Leval highlighted in Authors Guild v. Google, Inc., 804 F.3d 202 (2d Cir. 2015).
  • Filter outputs against verbatim reproduction where a generative system is involved.
  • Preserve provenance records for every source, especially for datasets.
  • Confirm lawful acquisition. Retaining material from a pirate source is a separate act with its own analysis.

Why this matters. Every one of these is evidence. Design choices that limit substitution are the most persuasive factor-four showing available, and they are far cheaper than litigation.

Phase 8 — Document, decide, and preserve

Why this matters. Fair use is decided on a record. The party that can produce a dated, specific, reasoned analysis is in a materially better position than the party reconstructing its thinking two years later, and courts weigh objective reasonableness heavily when awarding fees under 17 U.S.C. § 505.

Common mistakes

  • Treating fair use as a percentage rule. There is no safe word count, no 30-second rule, and no 10 percent threshold.
  • Analyzing the project instead of the use. Warhol requires the specific use to be identified.
  • Relying on attribution. Credit is not one of the four factors and does not make an unfair use fair.
  • Assuming nonprofit or educational status settles it. It helps on factor one and loses cases on factor four.
  • Confusing style change with transformative purpose. A different aesthetic serving the same function in the same market is weak after Warhol.
  • Ignoring acquisition. How you got the copy is a separate act with separate consequences.
  • Forgetting the contract. Terms of service and database licenses survive fair use entirely.
  • Deleting material after a demand arrives, which converts a defensible position into a spoliation problem.
  • Skipping the written analysis, which is the cheapest evidence available.

Primary authority

Related

This checklist is educational and not legal advice. Fair use is intensely fact-specific and the law governing large-scale and AI uses is developing rapidly. Consult qualified copyright counsel before relying on a fair use position.