The Work Product Doctrine and AI Legal Research: What FRCP 26(b)(3) Protects—and What It Doesn't

The work product doctrine is one of the foundational protections in civil litigation — shielding the attorney's mental processes, research strategies, and analytical materials from disclosure to opposing counsel. It predates digital practice by decades. Its application to materials generated using AI tools requires working through a doctrine developed for handwritten notes and dictated memos and mapping it onto a practice where a machine generates the first draft.

The core question is straightforward: when an attorney uses an AI tool to research an issue, generate case summaries, or draft litigation materials, do those materials — including the inputs (prompts) and the outputs — qualify as work product that is protected from discovery? The answer turns on the same two-part analysis that governs all work product claims under Federal Rule of Civil Procedure 26(b)(3), with some AI-specific complications that practitioners should understand before treating AI research as automatically protected.

The Work Product Framework: FRCP 26(b)(3)

Federal Rule of Civil Procedure 26(b)(3)(A) protects "documents and tangible things that are prepared in anticipation of litigation or for trial by or for another party or its representative (including the other party's attorney, consultant, surety, indemnitor, insurer, or agent)." That protection is qualified — the materials may be discovered if the requesting party shows both that they are otherwise discoverable and that the party has substantial need for the materials to prepare its case and cannot, without undue hardship, obtain their substantial equivalent by other means.

Rule 26(b)(3)(B) provides a separate, heightened protection for "opinion work product" — materials that reflect the mental impressions, conclusions, opinions, or legal theories of a party's attorney or other representative concerning the litigation. Opinion work product receives near-absolute protection that courts rarely pierce: even if ordinary work product is ordered disclosed because of substantial need, a court ordering that disclosure must still protect against disclosure of opinion work product.

Do AI Prompts Qualify as Work Product?

An attorney's AI prompts — the queries, instructions, and context that the attorney types into an AI tool — are likely to qualify as work product when two conditions are satisfied: the prompts were prepared in anticipation of litigation, and they were prepared by or for the attorney or another representative of the party. Both conditions are typically satisfied when an attorney uses an AI tool to research legal questions or draft litigation documents. The attorney prepared the prompts; the prompts were prepared in anticipation of the matter; they are "documents" in a functional sense — recorded text prepared by a representative of a party.

Moreover, an attorney's prompts frequently reveal litigation strategy in exactly the way that classic work product does. A prompt that asks an AI to identify weaknesses in a particular legal theory, or to analyze how a specific rule has been applied in factual contexts similar to the client's situation, reflects the attorney's assessment of the issues — potentially qualifying for opinion work product protection under Rule 26(b)(3)(B). A prompt that is merely a generic research query ("summarize FRCP Rule 11") reveals less about strategy and receives correspondingly weaker protection.

Do AI Outputs Qualify as Work Product?

AI-generated outputs — case summaries, draft arguments, factual analyses — are more complicated. They are not generated solely by the attorney's effort; the AI system participates in generating them. The better analysis is to ask whether the output, as used in the representation, was prepared by the attorney's representative (including the AI tool operating under the attorney's direction) in anticipation of litigation. Under that framing, an AI-generated research memo that the attorney directed the AI to produce, reviewed, annotated, and incorporated into litigation preparation should qualify as ordinary work product.

Where the analysis becomes more uncertain is when the AI output contains factual information — not attorney analysis, but raw data or information about the world — that would be independently discoverable through other means. A rule of thumb from general work product doctrine applies: work product protection extends to the attorney's selection and compilation of facts, but not necessarily to the underlying facts themselves. An AI-generated summary that represents the attorney's selection of which cases to include and what aspects to emphasize may be protected; the underlying cases are not.

The Waiver Risk: Third-Party AI Services

One aspect of AI-assisted practice creates a specific work product risk that does not arise with traditional research methods: the transmission of attorney materials to a third-party AI service. Work product protection can be waived by voluntary disclosure to a third party who is not part of the legal team and whose interests are not aligned with those of the client. Whether submitting prompts and documents to an AI service constitutes a waiver-triggering disclosure to a "third party" is a question courts have not definitively resolved in the AI context.

Attorneys using AI services with weaker confidentiality protections — services that retain inputs, use inputs for model training, or grant employee access to submitted content — face the greatest waiver risk. The analysis intersects directly with the Rule 1.6 confidentiality analysis: a service whose data practices are incompatible with attorney confidentiality may also create work product waiver risk. Enterprise-tier AI services that operate under attorney-specific data processing agreements, with no retention for training, are much less likely to create waiver problems.

Practical Documentation Recommendations

The work product analysis argues for treating AI prompts and AI-generated research materials the same way traditional research notes are treated: maintained in the client file, marked as attorney work product, and shared only within the legal team. That practice is appropriate both because it matches the likely legal analysis and because it creates the documentation trail that would support a work product objection if one is needed later.

None of this documentation is required by FRCP 26(b)(3) itself, which focuses on the nature of the materials rather than how they were labeled. But in a dispute over work product protection, showing that the attorney treated the materials as work product from the time of creation is strong evidence that the protection applies.

This content is legal information, not legal advice. It does not create an attorney-client relationship and cannot substitute for consultation with a licensed attorney about your specific circumstances.