Simon and Garfunkel gave us a line in 1969 that has outlived most of the songs written since: a man hears what he wants to hear, and sets the rest aside. It was written about self-deception, not evidence. But it is the most accurate description I know of what goes wrong when an expert witness points a generative AI tool at his own assignment — because these systems are built to be agreeable, and an expert who keeps asking until he likes the answer will eventually get one.
A recent post in this series looked at how to choose the right expert when a case turns on technology. This one is about the other chair. Your opponent’s expert has almost certainly used AI somewhere in his work, and the question is no longer whether that is happening but whether you are asking about it.
What He Wanted to Hear
The case that should be required reading for every trial lawyer came out of litigation over a 3M manufacturing plant in Houston. An engineer retained to opine on the standard of care used ChatGPT to prepare his report. In the middle of his deposition, more than three hundred pages of his AI prompts were produced. He acknowledged the report was something like eighty-five to ninety percent machine-generated, for a fee reported north of ninety thousand dollars.
That is embarrassing. What came next was fatal. The prompts themselves showed him asking the tool to produce an analysis assigning zero percent fault to the party that hired him. He was not using the machine to test a hypothesis. He was using it to manufacture a conclusion he had already chosen, and the record of him doing so was sitting in a chat log he apparently never imagined anyone would read. By the end, the opposing side found him useful enough to call in their own case.
Notice what actually destroyed that witness. It was not the use of AI. It was that the prompts preserved, verbatim and timestamped, the one thing cross-examination usually has to prove by inference — that the expert decided the answer first and built the method around it.
Lies and Jest
The cruder failure is showing up too. In a 2026 case in the Western District of Washington, an expert submitted a report containing citations the AI had invented. The court struck the report in its entirety under Rule 702, reasoning that an expert who had not verified his own sources had not based his opinions on sufficient facts or reliable methods. Summary judgment followed for the other side.
That outcome is worth sitting with, because it is more severe than what has been happening to lawyers. A sanctioned lawyer pays a fine and keeps litigating. An excluded expert can take the case with him. And the amendments to Rule 702 that took effect at the end of 2023 make this easier, not harder: the proponent must now show, by a preponderance, that the opinion actually reflects a reliable application of reliable methods. An expert who cannot explain how the analysis was produced has a problem with that burden before anyone reaches the merits.
You may have read that the rule makers are working on a new Rule 707 for machine-generated evidence. They are, slowly — the Advisory Committee held it for further study in May 2026. Do not wait for it. Rule 702 already does this work.
Fair Ground for Discovery
None of this helps if you do not ask, and the asking has to be specific. A magistrate judge in the District of Connecticut this year required disclosure of an expert’s AI prompts, reasoning that an expert’s methodology is fair ground for discovery. That framing is the one to borrow, because it sidesteps the argument you will otherwise get.
There is a real line here and you should know which side you are on. Prompts that counsel writes while developing theories are opinion work product and close to undiscoverable — a federal court in the Northern District of California said as much in 2024. But when the expert does the prompting, the prompts are not lawyer thinking. They are the method. They show what he asked, what he was given, what he discarded, and in what order.
So draft the request to reach it. Ask which tools were used and which versions. Ask for complete prompt and response histories, exports included, not a summary. Ask what documents or data were uploaded to the tool, which also tells you whether confidential material went into a public platform. Ask what was verified and how. Then depose him on whether he can explain and defend the analysis without the tool in front of him — because if he cannot, the jury is being asked to accept the opinion of a witness who is himself relying on something he does not understand.
The Fighter Who Remains
The same lens points back at your own expert, and the conversation is better had at retention than at deposition. Our engagement letters with experts now say plainly that any AI use must be disclosed to us, that no confidential or protected material goes into a public tool, that every authority and data source must be independently verified, and that the work has to be defensible without reference to the machine. None of that prohibits AI. It prohibits the version of AI use that gets a report struck.
The practical test is simple. If your expert cannot sit in a chair, without notes, and explain why the analysis is right, you do not have an expert. You have a witness reading output.
What We Still Want to Hear
The song’s insight is that the hearing problem is human and permanent. Experts have always been susceptible to telling the party who hired them what it wanted to hear; that is why we have Rule 702 and why we cross-examine. What has changed is that the tool now produces a complete, timestamped transcript of the process — every question asked, every reframing, every attempt until the answer came out right.
For a careless expert, that is a catastrophe waiting in a chat log. For the lawyer across the table who thinks to ask, it is the best discovery opportunity to appear in expert practice in a generation. Ask for the prompts.
This post is part of an ongoing series on how AI is reshaping litigation and law-firm practice. An earlier installment took up how to choose co-counsel and experts when a case turns on AI; this one takes up the expert on the other side.
A Word About Silver Cain
Silver Cain PLC represents businesses in complex commercial and real estate litigation in Arizona and beyond. When Rebecca Cain and I founded the firm, we built it around direct partner involvement and senior trial-level judgment — including the expert work, on both sides of the table, where complex cases are usually decided. If the questions in this post are relevant to your business, or to the firms you retain, we are glad to have that conversation. Leon Silver is an AV-rated trial lawyer at Silver Cain PLC, focused on commercial and real property disputes since 1989. Reach him at lsilver@silvercain.com.
