MorseSign inStart freeOpen Morse

Can expert witnesses use ChatGPT? What courts have done with an expert's prompts

The Morse teamPublished 9 min read

Yes, expert witnesses can use ChatGPT, and as of November 2026 no federal rule forbids it, but everything an expert types into it may have to be explained, produced, or defended. In May 2026 a federal magistrate judge in Connecticut ordered a party to disclose "any artificial intelligence prompts and/or queries used by" its expert "or her team in the course of producing her expert witness report" (ECF 970, now stayed), and courts in Minnesota and New York have already thrown out or discounted expert work that leaned on AI.

On November 19, 2026, we read the text of Federal Rule of Civil Procedure 26 and its 2010 committee notes, Federal Rule of Evidence 702 and its 2023 note, the docket and filings in Conservation Law Foundation v. Shell Oil Co., the opinions in Kohls v. Ellison and Matter of Weber, a District of Kansas protective order ruling, a Court of International Trade standing order, and OpenAI's help pages. Courts have mostly given experts room: the Minnesota judge wrote that the court "does not fault" the expert "for using AI for research purposes" (Kohls, ECF 46). The trouble starts with what the expert can't account for, and with who else holds a copy. This isn't legal advice.

Are an expert witness's AI prompts discoverable?

They can be. One of the first federal orders to say so came on May 18, 2026, in Conservation Law Foundation v. Shell Oil Co., No. 3:21-cv-00933 (D. Conn.). Shell asked for the prompts that historian Naomi Oreskes and her research assistant used to sort Shell's document production. Magistrate Judge Thomas O. Farrish held that "an expert witness's methodology is fair ground for discovery," and that "the process by which Dr. Oreskes culled down the defendants' document production into a subset to be worked with is an aspect of that methodology" (ECF 970).

The parties had agreed not to seek each other's "expert notes, drafts, or communications," and the plaintiff argued the prompts were notes. The judge found the agreement wasn't "quite clear" enough to cover them. He also doubted the plaintiff's answer that none existed, because the assistant's own declaration mentioned "prompt[s]." The order required a revised, signed response, and warned that "Rule 37(b) sanctions may then become available" if a claim that nothing exists "is later revealed to be untrue" (ECF 970).

The order is on hold. The district judge stayed it on June 3 (ECF 977), kept the stay on September 30 (ECF 1014), and denied Shell's motion to exclude Oreskes without prejudice because of "the open question regarding Dr. Oreskes's expert disclosure obligations" (ECF 1012). He heard argument on November 3, 2026 (ECF 1024), and on November 12 the plaintiff filed a supplement about the assistant's AI use (ECF 1027). As of November 19, the docket shows no ruling.

The team had tried to keep no record at all. The plaintiff's emergency motion to stay the order says "no prompt or output logs were preserved or exist in any form" (ECF 975). Its November 12 filing says the service was chosen because its security "essentially eliminated any risk of creating additional records or 3rd party access," and that prompts there "are not logged or visible to the service provider" (ECF 1027). On November 5, the district judge allowed that supplement on the assistant's AI use (ECF 1026), and it answers a question the court asked at the hearing: "What came back to [Dr. Kaurov] as a result of each search or prompt?" (ECF 1027). By the team's account, the tool kept the documents from outsiders, and it also left the team with no log to show the court.

Does Rule 26 protect AI drafts of an expert report?

Drafts, yes; the rest of the AI work, much less clearly. Rule 26(b)(4)(B) protects "drafts of any report or disclosure... regardless of the form in which the draft is recorded," and Rule 26(b)(4)(C) protects communications "between the party's attorney and any witness required to provide a report" (Rule 26). A chatbot isn't the party's attorney. The 2010 committee note says the rule leaves untouched any "inquiry about communications the expert had with anyone other than the party's counsel about the opinions expressed," and that "the expert's testing of material involved in litigation, and notes of any such testing" stay discoverable.

The report itself has to list "the facts or data considered by the witness in forming" the opinions, under Rule 26(a)(2)(B)(ii) (Rule 26). The same note says "facts or data" should be "interpreted broadly" to reach "any material considered by the expert, from whatever source, that contains factual ingredients." Whether an AI prompt counts is in dispute in Shell: the plaintiff's emergency motion to stay says prompts used only to cull documents were never "considered" in forming the opinions (ECF 975). No court has decided that yet.

Can an expert report written with AI be excluded?

Yes. Since December 1, 2023, Rule 702 requires the party offering an expert to show "it is more likely than not" that the testimony "is based on sufficient facts or data" and "is the product of reliable principles and methods" (Rule 702). The amendment's note says courts that treated those questions as going only to weight had applied the rule incorrectly.

  • Kohls v. Ellison (D. Minn., January 10, 2025). A Stanford professor's declaration cited two articles that don't exist, which he said came from GPT-4o. Judge Laura M. Provinzino wrote that citing fake sources "shatters his credibility with this Court" and excluded the declaration. She added that a reasonable inquiry "may now require attorneys to ask their witnesses whether they have used AI in drafting their declarations and what they have done to verify any AI-generated content" (order, ECF 46).
  • Matter of Weber (N.Y. Sur. Ct., October 10, 2024). A damages expert used Microsoft Copilot to cross-check his numbers but "could not recall what input or prompt he used." The court ran a similar question on three court computers and got three different answers, from $948,209.63 to a little over $951,000. It held that counsel "has an affirmative duty to disclose the use of artificial intelligence" before offering evidence it generated (opinion).

Discovery can also turn up the prompts themselves. In a Houston explosion case, 404 Media reported on August 17, 2026 that discovery records show a 3M expert asked ChatGPT to help "create an exceptional expert witness report defending the standard of care at 3M" (404 Media).

Can you put confidential case documents into ChatGPT?

Read the protective order and ask retaining counsel first, because many orders now answer this directly. On March 25, 2026, a federal court in Kansas amended a protective order so that every document produced in discovery, confidential or not, stays out of "open loop" AI tools. The original order already required notice to the other side, a chance to object, no training on confidential material, and deletion at the end of the case (Jeffries v. Harcros Chemicals, ECF 152). Some judges set their own rules. Judge Stephen Vaden of the Court of International Trade requires any submission in his cases with AI-drafted text to name "the program used and the specific portions of text that have been so drafted," and to certify that the use "has not resulted in the disclosure of any confidential or business proprietary information to any unauthorized party" (order of June 8, 2023).

A certification like that depends on what the app keeps. As of November 19, 2026, OpenAI says that on ChatGPT Free, Plus, and Pro, "data sharing is enabled by default" for model training (OpenAI), and that deleted chats go "within 30 days" unless OpenAI "must retain it longer for security or legal obligations" (OpenAI). We cover the training switch in does ChatGPT train on your data. Lawyers face a separate set of ethics rules on client information, covered in can lawyers use ChatGPT.

Which rule reaches each thing you put into AI?

Built from the rule text and orders above, read November 19, 2026. Whether a given use is allowed in your case is for retaining counsel and the court.

What you put into AIWhat reaches itWhat to do
Prompts used to search, sort, or cull case documentsDiscovery into methodology under Rule 26(b), as in Shell (stayed, not yet decided)Keep a dated record of each prompt and what came back, as you would notes of a test
Data or documents that shaped an opinionRule 26(a)(2)(B)(ii), "facts or data considered," read "broadly" (Rule 26)List them in the report like any other source
Your chats with an AI about your opinionsOutside Rule 26(b)(4)(C), which covers communications with the party's attorneyAssume you can be asked about them at deposition
Text an AI drafted for the reportRule 26(b)(4)(B) protects drafts; some judges, like Judge Vaden, require a notice of AI-drafted textAsk counsel what the court requires, and mark which passages started as AI text
Citations, figures, or calculations from an AIRule 702's reliability test; Kohls and WeberCheck each citation against the source and rerun each figure in your own tools
Documents produced under a protective orderThe order's own AI terms, as in JeffriesRead the order and ask retaining counsel before using any AI tool
Anything typed into a consumer appThe app's training and retention terms, and a deletion window of "within 30 days" or longer (OpenAI)Keep case material out of consumer chat apps unless counsel has approved one

Courts can also reach the history on your own laptop, as can AI chats be subpoenaed explains, and a party can be ordered to authorize providers to release account data, as in the Miami discovery order.

How Morse keeps an expert's own AI work private

Most rows in that table involve two records: the one you keep to show your method, and the one an AI company keeps under its own terms. An expert can't vouch for the second record, and Morse is built so that its servers don't keep one. Our servers don't keep messages or log message content, and Morse doesn't train AI on your chats, sell your data, or show ads. Your chat history is stored in your browser, encrypted, not on Morse's servers. Because we don't store chats, we usually have very little to give authorities (Morse privacy).

The first record stays your job. The history in your browser is part of your file, and your duty to preserve applies to it like any other notes. Clearing the browser or losing the device deletes those chats, and Morse can't recover them, so copy the prompts that shaped your method into your working papers while the case is live. None of this changes a protective order: if the order or retaining counsel keeps material out of AI tools, that includes Morse.

  • E2EE, for working through your method on anything counsel has cleared for AI use. Your browser checks the secure enclave's hardware attestation, then encrypts the message so only that enclave can read it, and Morse's servers relay text they can't read. The enclaves are run by third-party providers, not by Morse, and the guarantee rests on the enclave hardware and attestation working as designed. Every E2EE reply has a lock you can open to see each check that passed, and a full report you can download. E2EE runs open-weight models like GLM, Kimi, and DeepSeek, with no web search or image generation.
  • Private, for literature and methods questions with no case facts in them. New chats start here, on open-weight models like DeepSeek, Kimi, and GLM, plus xAI's Grok. Morse has zero data retention agreements with its inference, search, and other providers, so they've agreed to keep nothing, and where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't (what zero data retention means).
  • Anonymous, to ask GPT, Claude, or Gemini a general question about a statistical method or a standard without the provider knowing it's you. Our proxy removes your account, email, and IP address and sends the message under Morse's own account. The provider sees the text and may keep it for a limited time to check for abuse, under its own policy, so leave the parties, the matter, and the documents out.

Try a chat from the Morse home page without an account, or start free with 15 messages a day and no card. E2EE chats come with Pro, at $19 a month.

Common questions

Is an expert's chat with ChatGPT protected like a conversation with retaining counsel?

Rule 26(b)(4)(C) protects communications between the party's attorney and the expert. The 2010 committee note says questions about communications the expert had with anyone other than counsel are unaffected, so don't count on that protection for a chatbot.

If my AI tool keeps no logs, is the question of my prompts over?

No. In the Shell case, the expert's team says its tool kept no prompt logs, and the May 18, 2026 order, now stayed while the district judge reviews it, still required a signed response saying whether any prompts exist, with Rule 37(b) sanctions possible if that statement later proves untrue.

Do I have to tell the court I used AI on my report?

It depends on the court and the judge. Judge Vaden's order at the Court of International Trade requires a notice naming the program and the AI-drafted portions, and a New York Surrogate's Court held that counsel must disclose AI use before offering AI-generated evidence. Retaining counsel knows which rules apply in your case.

Can a judge throw out an expert's testimony because of AI errors?

Yes. In Kohls v. Ellison, a federal judge in Minnesota excluded an expert declaration that cited two articles that don't exist, after the expert said GPT-4o had supplied them.

Ask what you actually want to ask.

80+ models, characters, and images, with every chat encrypted and kept on your device. Free to start.

Start a private chat

Read next