by Jeffrey T. Donner, Esq.

Recent rulings suggesting that AI search prompts are not protected by the work product doctrine have generated enthusiastic approval in some circles. That enthusiasm is misplaced.

The issue is not whether AI is “privileged.” It is not. The issue is whether litigation-focused research inputs—typed into a modern interface—constitute protected work product.

Properly analyzed under Hickman v. Taylor and Federal Rule of Civil Procedure 26(b)(3), the answer should often be yes.

The Doctrine Protects Mental Impressions—Not Just Paper Files

Hickman v. Taylor, 329 U.S. 495 (1947), established that the adversarial system depends on protecting an attorney’s mental impressions, conclusions, opinions, and legal theories from discovery.

Rule 26(b)(3) codifies that principle. It protects documents and tangible things prepared in anticipation of litigation by or for a party or its representative. It also provides heightened protection for opinion work product.

Research queries—whether typed into Westlaw, Google, or an AI platform—are not random keystrokes. They reveal issue framing, burden analysis, anticipated defenses, and strategic direction.

A carefully crafted research prompt can expose the theory of the case more clearly than a memorandum.

That is classic opinion work product.

Work Product Is Not Limited to Attorneys

Another persistent misconception is that only attorney-generated material qualifies.

Rule 26(b)(3) protects materials prepared “by or for a party or its representative.” Courts have consistently recognized that this includes consultants, investigators, insurers, and the client.

If a client, anticipating litigation, conducts structured legal research—whether in a law library or through an AI interface—those materials may qualify as fact or opinion work product depending on content and context.

The doctrine does not belong exclusively to lawyers.

The Critical Question: Anticipation of Litigation

Work product protection is not automatic.

The threshold requirement is that the material be prepared because of anticipated litigation. If a prompt is entered for business strategy, regulatory compliance, or general curiosity, protection may not attach.

But where the prompt is litigation-driven—probing burdens, defenses, evidentiary standards, or appellate angles—the “because of” test strongly supports protection.

The medium does not alter the analysis.

The Third-Party Server Argument Is Weak

Some argue that AI prompts are different because they are transmitted to a third-party platform.

That distinction collapses under scrutiny.

Modern legal research has been conducted on third-party servers for decades. Westlaw, Lexis, Bloomberg, and countless cloud-based platforms store and process user queries.

Courts have not required attorneys to disclose their Westlaw search histories. Nor have they treated the use of research vendors as a waiver of work product protection.

The relevant question is whether confidentiality was reasonably maintained, not whether a server is owned by the law firm.

Disclosure of “AI Use” Is a Separate Regulatory Question

Courts may regulate filings. They may require accuracy certifications. They may sanction fabricated citations.

Those are appropriate exercises of judicial authority.

Compelling disclosure of litigation research inputs is something different. It risks intruding directly into protected strategic thinking.

That is precisely what Hickman sought to prevent.

The Current Landscape

It would be incorrect to say that appellate law has definitively resolved the treatment of AI prompts. The jurisprudence is still developing.

But the foundational doctrine is clear. The work product rule protects litigation-focused mental impressions prepared by or for a party.

A prompt that reveals case theory, burden analysis, or strategic testing fits squarely within that framework.

Technology evolves. The underlying doctrine does not.

If courts depart from that principle, they should do so openly and doctrinally—not by treating modern research interfaces as categorically different from every other research tool lawyers have used for decades.