Expert depositions in the age of AI

Opportunity, risk, and the new battlefield for plaintiff lawyers

2026 October

For most of my career, preparing for expert depositions required the same formula: read the records, master the medicine, study the literature, review prior testimony, and learn everything possible about the expert sitting across the table.

Those fundamentals have not changed. What has changed is the arrival of artificial intelligence. Whether we like it or not, AI is already reshaping litigation. Experts are using it. Lawyers are using it. Insurance companies are using it. Defense firms are using it. The question is no longer whether AI belongs in litigation. The question is how plaintiffs’ lawyers can use it intelligently while exposing its weaknesses when others rely on it improperly.

Expert depositions are rapidly becoming the place where those battles will be fought.

AI is a tool, not an expert

One of the first principles every plaintiff’s lawyer must understand is that AI does not replace expert testimony.

An expert witness is required to exercise judgment, apply their specialized knowledge and experience, evaluate facts, and explain opinions within a recognized methodology. Artificial intelligence cannot take an oath, cannot be cross-examined, and cannot independently satisfy the evidentiary requirements governing expert opinions. Yet AI is increasingly influencing the work product that experts generate.

Medical experts may use AI-assisted literature searches, prepare medical-record chronologies and summaries and prepare summaries of depositions and discovery. Additionally, they may only be given AI-generated summaries of documents. Economists may use AI to organize data. Accident reconstructionists may use AI-enhanced software to analyze photographs and measurements. Life care planners may employ AI-assisted research tools.

The important deposition question is no longer simply, “What is your opinion?” The question may also be:

“What role did artificial intelligence play in forming that opinion?” Ten years ago, no lawyer asked experts whether they used artificial intelligence. Today, I believe every expert deposition should include a series of AI-related questions.

For example:

• Did you use any AI platform in preparing your report?
• Which platform did you use?
• What information was entered into the system?
• What specific query did you use to obtain the information?
• Did you seek information from a particular viewpoint (e.g., on behalf of the defense)
• Did the system generate summaries, analyses, or recommendations?
• Did you independently verify all the AI-generated information?
• Can you specifically identify what portions of your report were influenced by AI?
• Did AI identify literature that you relied upon?
• Did you review that literature independently or get an AI summary?
• Did AI assist in drafting any part of your report?

Many experts will answer that AI played no role.

Others may reveal a level of reliance that creates significant impeachment opportunities. The objective is not to embarrass the witness; it is to understand whether the expert exercised independent professional judgment or merely adopted machine-generated conclusions.

Plaintiffs’ lawyers have long challenged opinions based upon undisclosed assumptions. AI presents the same issue on a larger scale. Many artificial intelligence systems operate as “black boxes.” They produce outputs without fully explaining how those outputs were generated, which becomes problematic in litigation.

Imagine a defense expert who testifies that an AI-assisted analysis helped determine collision dynamics or injury probabilities. In this situation the lawyer should immediately ask:

• How many AI searches did you generate?
• Make sure to ask for copies of all AI generated searches. (Code Civ. Proc., §§ 2034.415, 2034.270.)
• What specific data was used?
• What assumptions were incorporated?
• What error rates are known?
• How was accuracy validated?
• Can the methodology be independently reproduced?

If the expert cannot answer those questions, the credibility of the opinion may begin to unravel. As trial lawyers, we should never allow technology to become a substitute for methodology.

AI creates new opportunities to uncover bias

One of the traditional goals of expert deposition is exposing bias. Artificial intelligence creates entirely new avenues for inquiry. For example:

• Was the AI tool developed by an insurer?
• Was it trained using defense-oriented datasets?
• Were alternative datasets considered?
• Did the expert evaluate the limitations of the system?

Historically, we examined financial bias and litigation history. Now we must also examine technological bias.

An AI system can reflect the assumptions of its designers just as an expert can reflect the interests of a retaining party.

Jurors may find this concept surprisingly easy to understand because many have already encountered stories about AI producing inaccurate or skewed outputs.

AI as a powerful tool for deposition preparation 

The greatest immediate value of AI may not be in expert testimony itself, but in helping lawyers prepare for expert depositions. Before you use AI to prepare for an expert deposition you must have a firm grasp on the evidence related to that expert’s particular expertise. Used responsibly, AI can help:

• Organize medical records.
• Identify inconsistent opinions.
• Summarize scientific literature.
• Compare prior testimony.
• Generate deposition topics.
• Highlight gaps in reports.
• Create chronologies.

What once required dozens of hours can sometimes be accomplished in a fraction of the time. But there is a critical warning: Never assume AI-generated summaries are accurate! I have seen AI misstate medical records, fabricate citations, and incorrectly summarize prior deposition and trial testimony. 

Ninety percent wrong: Expert depositions in the age of artificial intelligence

A while back I asked an AI system to do something I thought was simple. I had a defense expert I was convinced had given a similar opinion in numerous other cases with a similar fact pattern and damages. I had a stack of his prior depositions and trial transcripts sitting on my desk. So, I fed them in and asked the system to find every place he had given a similar opinion or said something close to it.

The answer came back fast, and it came back promising. Several passages, each one apparently on point, each one with a citation to a transcript and a page number.

Then I went to verify. Roughly 90% of what the system told me was in that sworn testimony was not there. Not paraphrased loosely. Not taken slightly out of context. Not there at all. So, it is important to trust, but verify. Actually, verify first.

The danger of AI hallucinations

Experienced lawyers already understand the danger of unsupported opinions. AI introduces a new version of that problem. Generative AI systems sometimes produce entirely fabricated information while presenting it with complete confidence. These so-called “hallucinations” create obvious litigation risks.

Imagine questioning an expert who relied upon a medical article that does not exist or cites data that cannot be verified. The resulting cross-examination can be devastating. That is why an expert should never be permitted to hide behind technology. The expert remains responsible for every opinion offered under oath, regardless of whether AI assisted in generating it. A machine does not bear responsibility for testimony. The witness does.

AI is hearsay 

People v. Sanchez (2016) 63 Cal.4th 665 bars an expert from relating as true case-specific hearsay facts he has no independent knowledge of, and it applies in civil cases. An expert whose factual basis for his opinions came from an AI summary is relying on hearsay with no declarant at all, nothing to cross-examine, and no exception available. 

Now think about an expert whose factual predicate came out of an AI summary of the medical records, or an AI summary of a deposition he never actually read. That is hearsay, and the declarant does not exist. There is no witness to produce, no exception to argue, and nothing to cross-examine. It is the cleanest foundational objection available in this entire area.

Then there is Sargon Enterprises, Inc. v. USC (2012) 55 Cal.4th 747. Under Evidence Code sections 801(b) and 802, the trial court excludes opinion based on matter that cannot reasonably be relied upon, based on reasons unsupported by the material relied on, or that is speculative. The court may find too great an analytical gap between the data and the opinion offered. Lockheed Litigation Cases (2004) 115 Cal.App.4th 558 says it plainly: the material relied on has to furnish a reasonable basis for the particular opinion.

So, ask the question that matters. If the analytical step was performed by a tool and the witness cannot explain, whose opinion is it?

Preparing your own experts

Plaintiffs’ lawyers should also prepare their own experts for AI-related examination. Assume defense counsel will ask:

• Did you use AI?
• How did you use it?
• What did you verify?
• What role did it play in your conclusions?

The strongest answer is usually one that demonstrates professional independence. Jurors want to hear that the expert reviewed the evidence, exercised judgment, applied his/her education and experience, and reached conclusions through established methods. AI may assist the process, but it should not replace the process.

The more transparent your expert is about the role of technology, the more credible that expert is likely to appear. As you prepare your expert for their deposition it is important to keep these things in mind:

Establish clear boundaries before the deposition

Meet with the expert and determine precisely:

• Whether AI was used.
• Which platform was used.
• What tasks AI performed.
• What information was entered.
• What portions of the work product were affected.
• Was the AI information verified, and how so?
• Did he/she keep copies of all AI-generated queries and did any of the information get cut and pasted into any document he/she is producing? 

If so, how does it affect their overall opinion and basis for the opinion? (Evid. Code, § 802.)

The expert should be able to explain these facts clearly and comfortably.

It should be noted that no California decision has yet held that prompts/queries are discoverable, so know that this is only an argument and not the law. Here is the argument to be made: Evidence Code section 802 permits inquiry into the reasons for an opinion, and Sargon makes those reasons the admissibility test. Code of Civil Procedure sections 2034.270 and 2034.415 reach the expert’s writings and the materials he consulted. If a tool shaped what the expert looked at, or performed a step of the reasoning, then the prompts and the outputs are methodology, and methodology is not privileged.

It is critical to emphasize independent judgment

The safest and most persuasive testimony is: “I used AI as a research and organizational tool, but I independently reviewed all relevant records, verified the information, and reached my own opinions.” That answer immediately frames AI as an assistant rather than a decision-maker.

Prepare for questions about verification

Defense counsel will likely ask:

• How did you verify the AI output?
• What steps did you take to ensure accuracy?
• Did you review the cited literature personally?
• Did you rely on anything generated solely by AI?

Your expert should be prepared to explain a verification process. The more rigorous that process sounds, the more credible the expert becomes.

Anticipate “hallucination” questions

By now, most lawyers and many jurors have heard stories about AI generating false information. Your expert should be ready to explain: “I never rely on AI-generated information without reviewing the original source materials.” That single answer neutralizes much of the attack.

Teach experts to avoid overstating AI’s capabilities

Experts should never suggest:

• AI is always accurate.
• AI is more reliable than professional judgment.
• AI replaced traditional analysis.

Jurors tend to become skeptical when technology is portrayed as infallible.

Develop an AI theme consistent with your case

For plaintiffs’ lawyers, the most effective theme is usually: “Technology assisted the process. Human expertise drove the conclusion.” That is a theme virtually every juror can accept.

The jury perspective

Perhaps the most important issue is how jurors view artificial intelligence. Most jurors are neither technologists nor AI skeptics. They use AI tools in their personal and professional lives. They recognize the benefits, but also understand the risks. As a result, jurors are unlikely to reject expert testimony simply because AI was involved. They are far more likely to focus on a different question: “Did the expert think for himself or herself?” The lawyer who demonstrates that an expert blindly accepted computer-generated conclusions will have a powerful cross-examination theme. The lawyer who demonstrates that an expert thoughtfully employed technology as one tool among many will likely preserve credibility.

Artificial intelligence is not replacing expert witnesses anytime soon; but it is changing the way experts work, the way lawyers prepare, and the way depositions should be conducted. For plaintiffs’ lawyers, this evolution presents both opportunity and danger. The opportunity lies in using AI to prepare more efficiently and to uncover weaknesses that previously remained hidden. The danger lies in allowing technology to escape scrutiny.

At the end of the day, the fundamental principles of expert depositions remain unchanged. Credibility matters. Methodology matters. Transparency matters. Independent judgment matters.

Artificial intelligence may be the newest participant in the deposition room, but the ultimate question is the same one trial lawyers have asked for decades: Can this opinion be trusted? That question, not the technology itself, will determine who wins the battle of the experts.

If I were trying a case tomorrow involving AI-assisted expert work, my overarching theme would be: “Artificial intelligence can help experts gather information. It cannot replace human judgment, experience, responsibility, or common sense.” The expert who demonstrates independent reasoning will likely do well with a jury. The expert who appears to be merely reading the output of a machine will not.

Donna M. Maryanski is senior trial counsel at Arias Sanguinetti Trial Lawyers with more than 35 years of experience. She is a Diplomate Member of the American Board of Trial Advocates (ABOTA).  Her practice focuses on personal injury, wrongful death, catastrophic injury, sexual-abuse litigation, and employment law.