Expert Witnesses and Generative AI: Case Law, Deposition Questions, and Resources

Generative AI and the Expert Witness: What the Courts Are Telling You

On our last monthly Experts.com community call, one question pushed itself to the top of the conversation and stayed there. An expert asked the group: have you been asked about your AI usage when being deposed?

It turned out to be the question everyone wanted to talk about. The inquiry came up several ways in deposition.

• How do you use AI?
• Do you use it when drafting your expert report?
• Are you doing research with AI?
• Are you using it to check your sources?

The room had a lot to say, and the more people talked, the clearer it became that this is something experts are actively running into. Honestly, in my legal technology echo chamber, it looked like the lawyers were further behind the curve, so I wasn’t prepared for the number of experts responding that it had come up in deposition.

Approximately 40% of the experts on the call have already been asked about AI usage during a deposition. And almost none of them had gotten any guidance from their retaining attorneys ahead of time. No prep on the topic. No conversation about it. They walked into the deposition and got the question cold.

After the community call, I ran into an attorney from the neighboring office suite. He does medical malpractice defense. I asked him if he’s asking about AI usage in depositions and he said he indicated he is.

Then I took to LinkedIn to see what type of feedback I could get from litigators. Here’s the LinkedIn thread. Make sure to take a look at the comments as there were some other LinkedIn posts shared of value. Please participate or add resources.

All of this told me two things. Experts need to get smart on this fast. But so do retaining attorneys, because if the lawyer is not preparing the expert for this line of questioning, the lawyer has a gap too. Here is what I found and what I think it means for expert witnesses and litigators.

The One Distinction That Runs Through All Of It

If you take nothing else from this article, take this: The courts are not punishing experts for using AI. They are punishing experts who let AI replace their own judgment, knowledge, or work product, and who put fabricated, unverified material in front of the court. Same tools, different outcomes, depending entirely on how you use it.

Two cases show you both ends of that. I did not include the cases themselves. Instead, we have linked to articles by law firms who either analyze or summarize the cases.

Let’s start with a good outcome. In this piece from Greenberg Traurig shareholder Kevin J. Quilty, Esq., citing Ferlito v. Harbor Freight Tools USA, Inc. (E.D.N.Y. 2025), the court refused to throw out an expert who used artificial intelligence to double-check his findings. The detail that saved the expert? He wrote his report first, using his decades of experience, and then used AI only to confirm what he had already concluded on his own. AI was his second opinion, his editor or double checker. It was not his first draft.

Now the bad outcome and the one we discussed in our monthly call. Again, see the brief summary from the Quilty article. Here is the order on the motion to exclude. Kohls v. Ellison (D. Minn. Jan. 10, 2025) is the case that scared a lot of attorneys and experts and for good reason. A Stanford professor submitted a declaration he had drafted with help from an LLM. It cited two academic articles that did not exist and a third article was misattributed. The most embarrassing part of all? His area of expertise was AI and deepfakes. He was the AI expert and he never reviewed his own AI-assisted report before signing or submitting it.

The court excluded the expert declaration and said the error “shatters his credibility.” The court went further and reminded the retaining counsel that they have a responsibility to ensure the accuracy of all court filings. This is important because the court said lawyers need to ask their experts if they use AI and how they verify AI generated material. I expect retaining counsel will begin asking experts how they use AI. Sadly, that did not happen for our members’ who were questioned about AI use in deposition.

The difference between Ferlito and Kohls, is how the experts used the tools. One used it to double-check his work and in the other, the expert handed over his own judgment to the machine and did not review the output for accuracy. Two experts, two tools, two opposite results.

Your prompts are discoverable, so write them that way

This is the development I most want the experts on our calls to hear because it changes how you should work starting today.

In this article from Arnold & Porter lawyers Melissa Weberman and L. Michel Marchand, the authors cover the case of Conservation Law Foundation v. Shell Oil Co. (D. Conn. May 2026). In this matter, a federal magistrate judge ordered the plaintiff to hand over the AI prompts its expert had used to analyze the other side's document production. The expert had used AI to review the production and narrow it down to a smaller set of documents worth a closer look. When the defendants asked for the prompts behind that process, the court held that the prompts were part of the expert's methodology, and that methodology is fair game for discovery under Rule 26.

The plaintiff fought it on three grounds and lost on all three. The one most instructive and most relevant is following. The parties had a discovery agreement protecting "expert notes, drafts, or communications" from the report-drafting process, and the plaintiff tried to call the prompts "notes" to fit them inside that protection offered by the agreement. The court refused to stretch the agreement that far. It said an agreement “must be quite clear" before it overrides otherwise-relevant discovery, and calling prompts "notes" did not get there. I do find this discovery agreement hopeful for expert witnesses AI usage. I would expect more attorneys to further outline that AI prompts, edits, reviews, AI output be included and protected by a discovery agreement. It feels like a safe away to avoid having the prompts exposed.

The order was later stayed while the district judge reviews the plaintiff's objection, so this is not the final word. But the reasoning is out there now, and other courts are reading it. In addition, you may wish to look at this anecdote about prompts from attorney William Moye’s LinkedIn post. He got some troublesome prompts revealed from the opposing expert.

So here is my advice. Assume your prompts are going to be read out loud at your deposition. Write every one of them like opposing counsel is looking over your shoulder, because one day they might be. Clean, well-structured prompts that show a thoughtful approach make your work look more credible. Sloppy or leading prompts do the opposite and hand the other side a map to attack your methodology.

Verification Is On You, And It Is On Your Lawyer Too

The professional-responsibility side of this has moved quickly. Maybe quicker for lawyers as they’re the ones who have been caught more than a thousand times submitting AI hallucinated content and primarily “fake citations.” The Louisiana Law Review article Egregious Errors in Expert Evidence: Ethical Oversight for Experts Who Use Generative AI lays out the new duties, and they land on both sides of the engagement.

For the retaining attorney, Kohls sent a signal that a "reasonable inquiry" may now mean the lawyer has to ask the expert directly whether they used AI and what they did to verify any AI-generated content. The article argues that lawyers should, at the very least, ask about AI use and independently confirm that every cited source exists and says what it is claimed to say before anything gets filed. Under the attorney duty to verify expert testimony, the author contends attorneys may be required to conduct a proactive investigation of the proffered expert testimony.

For us, the takeaway is simpler and a lot more professionally fraught. You are the one who is going to sit in the deposition and explain how you reached your opinion. If a citation in your report turns out to be fake, "the AI wrote it" is not a defense. It is a self-created credibility problem. You cannot hand the job of checking your work to the same tool that produced it.

The article also points to a fix I expect we will all start seeing more of, and it ties right back to the gap from our community call. Engagement contracts should address AI use directly. When the lawyer or expert puts AI expectations in the retention agreement, it forces the conversation up front about what is allowed, what is off limits, and how much risk the client is willing to carry. Your engagement letter needs to include some coverage of AI use whether you use it or not, it needs to be covered. It protects you as much as it protects them. And given that almost nobody on our call had been prepped on any of this, the engagement letter is exactly where that prep should start. Do not wait for the attorney to do the job. Be proactive

Where the rules are going

The ground under all of this is still moving. The most-cited scholarship in the area, the Grossman and Grimm work on acknowledged and unacknowledged AI-generated evidence, proposed reforms that fed straight into a proposed Federal Rule of Evidence 707. That rule, if adopted, would apply the same expert-reliability standards you already work under, Rule 702 and Daubert criteria, to machine-generated evidence offered without a supporting expert. Essentially, AI generated evidence would have to meet the Daubert standard. The current rules are just being stretched to cover the machine outputs that now feed into it. The proposed changes to the Federal Rules of Evidence, however, will take time to change. Focus on what you have control over: 1) Your Use of AI; 2) Your Retention Agreement.

The playbook

Here is everything above boiled down to a list you can keep next to your desk.

1. Write your real analysis first, on your own. Let your expertise drive the conclusion. That is why Ferlito survived.
2. Use AI to verify, not to create. A second opinion that confirms your reasoning isndefensible. A first draft that becomes your reasoning is not.
3. Check every citation yourself. Confirm each source exists, says what you say it says, and is attributed correctly. Kohls turned on fake citations the expert never checked.
4. Save your prompts and outputs. Treat them as discoverable from day one, because Conservation Law Foundation v. Shell says they are. Keep them clean and organized.
5. Disclose your AI use or non-use in the report. Being upfront lets the court evaluate your work under the normal standards. Hiding it invites the opposite.
6. Get AI into your engagement letter. Talk to your business attorney about editing your engagement letter and retainer agreement. Have the conversation with retaining counsel about what is permitted, what is not, and where the client's risk tolerance sits.
7. Be ready to explain it in deposition or on the stand. You formed the opinion, so you defend it. Know your tools well enough to testify about how they got to their outputs and what you did to confirm them.

As I am not a practicing lawyer, I reached out to colleagues on LinkedIn for their input on this subject. As always, our friend and Experts.com member Mitch Jackson, Esq. of Jackson & Wilson in Orange County, provided some outstanding contributions.

A trial attorney with more than 30 years of experience and more than 75 trials. He provided the following list to help experts in their preparation for questions they may face in expert witness.

Depositions Issues/Questions:

Per Mitch Jackson, if the expert admits using AI, move methodically from identification to scope of use, to reliability, to preservation of evidence. Use this deposition outline:

1. You testified that you used artificial intelligence in connection with your work on this case, correct?
2. Please identify every artificial intelligence, generative AI, large language model, chatbot, or AI-assisted software tool you used.
3. When did you first use AI in connection with this matter?
4. How many times did you use AI in connection with this matter?
5. Who made the decision to use AI?
6. Did anyone suggest, encourage, or direct you to use AI?
7. Describe every task for which you used AI.
8. Did you use AI to conduct research?
9. Did you use AI to identify sources, authorities, literature, studies, articles, regulations, or other materials?
10. Did you use AI to summarize materials?
11. Did you use AI to analyze data?
12. Did you use AI to perform calculations?
13. Did you use AI to identify trends, patterns, or conclusions?
14. Did you use AI to formulate, refine, or test opinions?
15. Did you use AI to generate hypotheses or possible explanations?
16. Did you use AI to prepare outlines, notes, or work papers?
17. Did you use AI to draft any portion of your expert report?
18. Did you use AI to edit, revise, proofread, reorganize, or improve any portion of your expert report?
19. Did you use AI to draft any opinions contained in your report?
20. Did you use AI to prepare exhibits, charts, graphs, demonstratives, timelines, or presentations?
21. Did you use AI to prepare for today’s deposition?
22. Did you use AI to prepare for anticipated trial testimony?
23. Please identify every opinion that was influenced in any way by AI.
24. Which portions of your report were generated, drafted, revised, or edited with the assistance of AI?
25. What specific prompts, instructions, or questions did you provide to the AI?
26. Did you create those prompts yourself?
27. Did anyone else provide prompts, instructions, or information to the AI?
28. What information did you input into the AI system?
29. Did you provide the AI with case-specific facts?
30. Did you provide the AI with deposition transcripts?
31. Did you provide the AI with documents produced in discovery?
32. Did you provide the AI with attorney work product?
33. Did you provide the AI with confidential or protected information?
34. What output did the AI generate?
35. Did you save the outputs generated by the AI?
36. Are all AI-generated outputs still available?
37. Did you review every AI-generated output before relying on it?
38. How did you verify the accuracy of the AI-generated information?
39. What steps did you take to determine whether the AI-generated information was accurate?
40. Did you independently confirm the authorities, citations, references, or studies identified by AI?
41. Were any AI-generated statements found to be inaccurate?
42. Were any AI-generated citations found to be incorrect or nonexistent?
43. Were any AI-generated analyses rejected by you?
44. What criteria did you use to decide whether an AI-generated output was reliable?
45. Did you disclose your use of AI in your expert report?
46. If not, why not?
47. Are there any opinions in your report that cannot be fully explained without reference to AI-generated material?
48. Can you identify every sentence, paragraph, conclusion, opinion, citation, chart, graph, or demonstrative that was influenced by AI?
49. Have you produced all documents reflecting your use of AI?
50. Have you produced all prompts used with AI?
51. Have you produced all AI-generated outputs?
52. Have you produced all drafts generated with AI assistance?
53. Have you retained all records reflecting your use of AI?
54. Have any prompts, outputs, drafts, logs, or related materials been deleted, destroyed, modified, or lost?
55. Are there any AI-related materials that have not been produced in this litigation?
56. Sitting here today, is there anything AI contributed to your work on this case that we have not yet discussed?

Questions 25–56 are often where the value lies. Many experts will readily admit they used AI for “editing” or “research,” but have not preserved prompts, outputs, drafts, or verification records. Those questions can expose the extent of reliance, the adequacy of the expert’s methodology, and whether discoverable materials exist that were not disclosed. In California state court, I would also consider a follow-up series focused on whether the AI-generated material formed part of the basis for any opinion, because that goes directly to expert foundation and discoverability arguments.

Bottom Line

The courts are forming some foundations we can work with. We cannot say they will all follow the rules outlined above, but this is the best guide we have at present. AI is a legitimate tool for expert analysis, and the cases admit it. What ends careers is giving up your independent judgment, signing your name to output you never checked, building conclusions you cannot defend, and treating a confident machine answer as a stand-in for your own expertise.

Forty percent of the experts on our last monthly members only call were asked about AI use under oath, all with no warning or guidance from retaining counsel. Some of them more than once. I would rather they walk into this line of questioning informed and ready.

Read this article. Read the resources I’ve linked to get a firm understanding of how courts and academia are analyzing the issue. Start building your own defensible process for AI use in your expert witness practice.

Finally, it should go without saying, if you do not feel you can reasonably use AI in the process of writing and defending your expert witness reports and testimony, it is better not to use it in those processes until you are more comfortable.

This article is for general information and is not legal advice. Experts and counsel should consult the applicable rules of professional conduct and the standing orders of the specific court where they appear.