Attorneys using generative AI can borrow a familiar litigation skill: Evaluate AI-generated output the way they would evaluate an opposing expert witness opinion under Federal Rule of Evidence 702. An expert witness cannot simply “waltz into the courtroom and render opinions” without adequate foundation. Clark v. Takata Corp., 192 F.3d 750, 759 n.5 (7th Cir. 1999). Nor can an expert bridge the gap between data and conclusion with nothing more than ipse dixit — the expert’s own say-so. Gen. Elec. Co. v. Joiner, 522 U.S. 136, 146 (1997).
Rule 702 imposes that discipline directly. It requires expert testimony to be “based on sufficient facts or data,” be “the product of reliable principles and methods,” and reflect “a reliable application of the principles and methods to the facts of the case.” So, when an opposing expert offers an opinion, a litigator’s instinct is not to accept it but to probe its basis and reliability.
That same mindset can be extremely helpful when working with AI-generated output. AI can be useful, but there can be gaps in the record, weak reasoning, or assumptions the user never intended to make hidden behind what appears to be a polished output. Like an expert witness, AI synthesizes information, applies a methodology, and offers conclusions. So, don’t treat AI output like a final answer. Treat it like an expert opinion.
The expert-witness playbook translates into five practical habits: Check the inputs, look for what’s missing, scrutinize surprises, get help when needed, and test for consistency.
1. Check the Inputs. Rule 702(b) requires that expert testimony be “based on sufficient facts or data.” If an expert opines on the cause of a plaintiff’s injury but did not review the plaintiff’s deposition or medical records, how would the expert know? Ask a similar question when using AI: Could the tool have given a well-founded answer based on the materials I provided? Did it identify and rely on the relevant facts? If an AI tool is evaluating a statute of limitations defense, for example, it needs both the case facts and the relevant case law — and should cite the sources supporting its conclusion.
- Lesson: Give the tool the sources you would consult, ask for citations, and account for any materials it could not review.
2. Look for What’s Missing. Litigators routinely accuse experts of “cherry-picking” — selectively crediting favorable facts while ignoring contrary evidence. When evaluating an opposing expert, a good litigator identifies key evidence the expert’s report does not address and probes why. The same process helps with AI. If a tool’s summary of a plaintiff’s medical records omits a hospital where the plaintiff was treated, ask the tool why. Perhaps the hospital records were not relevant, but perhaps the tool was unable to read the file or provided what it considered a complete analysis before reaching those records.
- Lesson: When key information is conspicuously absent from AI output or seems to undermine it, ask how the tool accounted for it.
3. Scrutinize the Surprise. Courts may consider whether an expert’s methodology has gained “general acceptance” in the relevant field. Daubert v. Merrell Dow Pharms., Inc., 509 U.S. 579, 594 (1993). Before Daubert, general acceptance was the entire admissibility inquiry under Frye v. United States, 293 F. 1013 (D.C. Cir. 1923). It is now just one of several non-exclusive factors courts may consider, and methodology that is not generally accepted “may properly be viewed with skepticism.” Daubert, 509 U.S. at 594. AI users should strike a similar balance: Be open to unexpected conclusions, but treat them with the same skepticism an unconventional expert methodology would receive.
- Lesson: A surprising answer should not be rejected automatically, but it should be tested against accepted authorities.
4. Get a Second Opinion. Experienced litigators know the themes opposing experts advance and their weaknesses. But even experienced litigators rely on their own experts to spot substantive problems, and new litigators — or those who are navigating a case with unfamiliar themes — rely even more heavily on experts to help build their understanding. AI users face a similar dynamic: When you are less familiar with the subject matter, it is harder to know whether the output is reliable. In those situations, run AI-generated output past a colleague or consultant with deeper expertise — someone who can check the tool’s reasoning the way a consulting expert checks an opposing methodology.
- Lesson: If you cannot evaluate the output yourself, bring in relevant expertise or ask the tool to benchmark its answer against authoritative sources.
5. Test for Consistency. Expert witnesses who testify frequently run into trouble if they start each case with a conclusion in mind and then reverse engineer a methodology to reach that conclusion. If an expert applies different rules in different cases to reach preordained conclusions, the conflicting approaches can demonstrate that the expert fails to satisfy Rule 702’s reliability requirements. Likewise, inconsistencies in AI outputs may signal that the tool’s approach is unreliable and needs adjustment.
- Lesson: For repetitive tasks, compare outputs over time; for unique prompts, ask the question more than one way and investigate discrepancies.
AI-generated output can be strengthened using the same disciplined scrutiny a litigator brings to an opposing expert’s opinion. Ask what the tool relied on, what it may have missed, and whether its reasoning holds up under pressure. Attorneys using generative AI should neither trust AI blindly nor distrust it — instead, cross-examine it.
The material contained in this communication is informational, general in nature and does not constitute legal advice. The material contained in this communication should not be relied upon or used without consulting a lawyer to consider your specific circumstances. This communication was published on the date specified and may not include any changes in the topics, laws, rules or regulations covered. Receipt of this communication does not establish an attorney-client relationship. In some jurisdictions, this communication may be considered attorney advertising.
