AI in the Courtroom: Lessons for Employers and Their Counsel from the ABA ERR Conference

As I mentioned on Monday, I had the opportunity to recently attend the ABA Section of Labor and Employment Law’s ERR conference in Nashville. One program that stood out was a panel titled “AI in Action: Discovery and Motion Practice in Employment Law.”

If you’ve been reading this blog over the years, you know I’ve been writing about AI in the employment law space for a while now — from AI face-scanning in hiring back in 2019, to the three legal areas employers need to think about when using AI in the workplace, to the more recent posts on updating litigation holds for the age of generative AI and the discoverability of GenAI data.

But this panel shifted the focus from how AI affects the workplace to how AI is transforming the practice of employment law itself. And it raised a few issues that employers and their counsel really need to be thinking about.

The panel was moderated by Sue Ann Van Dermyden of Van Dermyden Makus Law Corporation and featured an impressive lineup: the Honorable Anthony Porcelli of the United States District Court for the Middle District of Florida, Kimberly Duplechain of Littler Mendelson, and Lindsey Wagner of Wagner Legal. The discussion covered the full lifecycle of litigation — from intake and investigation all the way through motion practice and trial — and how AI tools are being deployed at every stage.

Here are a few takeaways that employers and their counsel should have on their radar.

Check Your Citations — The Judge Is Checking Too

Judge Porcelli’s message to the room was direct: whatever output you get from a generative AI tool, you must verify every case citation before you file it with the court. This is fundamental, and it’s not optional.

Judge Porcelli noted that he uses a tool called “Judicial Quick Check” to verify the accuracy of citations in briefs submitted to his court. But he also shared a cautionary tale of a hallucinated case that came from a major legal research provider. AI can be a terrific co-counsel, but it’s a co-counsel that needs supervision.

I’ll add a personal note here: our firm has started using AI tools in our practice as well. We’ve found them to be genuinely useful for research, drafting, and case analysis. But we treat them the way you’d treat a very smart, very fast junior attorney. And as I’ve told the people I work with time and again: You check the work. Every time. No matter how “confident” the AI sounds.

AI Across the Litigation Lifecycle

One of the more practical parts of the discussion focused on how AI is being used across the entire lifecycle of an employment case. The panelists walked through the stages — early case assessment, drafting complaints, e-discovery and document review, interrogatory preparation, deposition planning, expert witness analysis, opposing party document review, and motion practice — and discussed how AI tools are being deployed at each step.

On the investigation side, for example, AI can assist with initial complaint analysis, evidence collection and preservation, document review, witness interview preparation, pattern recognition and statistical analysis, and even investigation report generation.

But the panelists were quick to emphasize that the quality of your AI output depends entirely on the quality of your input. One of the speakers walked through a structured prompting framework: assign a professional persona (the “role”), define a specific task (the “objective”), provide the relevant facts and sources (the “context”), set guardrails (the “constraints”), and specify the formatting you need (the “structure”).

Then — and this is critical — review the output and ask the AI to critique its own work. What are the problems with these questions? How can I improve them? Avoid compound questions. Avoid accusatory or cross-examination-style questions. Iterate.

This kind of disciplined approach is what separates useful AI-assisted work from the kind that gets you a show cause order.

The Evidentiary Questions Are Coming

For employers and their counsel, perhaps the most forward-looking part of the panel involved the emerging questions around AI-generated evidence in the courtroom. The Advisory Committee on the Federal Rules of Evidence has proposed a new Rule 707, which would regulate the admissibility of machine-generated evidence and require that AI-produced evidence meet the same standards as expert testimony.

This is a big deal. As the panelists discussed, there are fundamental challenges with AI-generated evidence that courts are only beginning to grapple with. If you run the same query through an AI tool twice, you may get different output each time. How do you articulate the reliability of a system that doesn’t produce consistent results? How does a court evaluate something that can’t be cross-examined?

We used to rely on experts to perform various analyses — statistical modeling, pattern recognition, document review. Now, clients can push data through AI tools and potentially argue they don’t need to retain an expert at all. But as the panelists made clear, that’s not going to cut it. Courts are going to want to understand the methodology, the inputs, the reliability, and the limitations. Just because a machine produced the output doesn’t mean the output is reliable, and judges are going to demand that parties prove it is.

The Discoverability of Prompts — A New Frontier

The panel also tackled a question I talked about recently – the discoverability of prompts.

The short answer is: it depends, and the law is developing quickly. As the panelists discussed, there’s a threshold question about what type of tool you’re using and what confidentiality protections are in place. If you get past those protections, the judge on the panel suggested that prompts could be another form of work product — reflecting the attorney’s mental impressions about how to develop legal strategy.

This is consistent with what courts have been finding. In Tremblay v. OpenAI, a Northern District of California court held that AI prompts crafted by lawyers can constitute opinion work product when used for litigation-related purposes. But in United States v. Heppner, decided just this February by Judge Rakoff in the Southern District of New York, documents generated by a non-lawyer client through a consumer AI platform were held not to be protected — because the client acted on his own initiative, without counsel’s direction, and the platform’s terms of service effectively destroyed any expectation of confidentiality.

The comparison between prompts and traditional search terms came up as well. Are prompts the equivalent of the keyword searches that parties have long been required to disclose in e-discovery? Judge Porcelli suggested that requiring the production of prompts themselves is probably a bridge too far. The more appropriate question is whether you met your discovery obligations — not how you got there. But that line is going to be tested, and tested soon.

The practice of employment law is changing fast. AI is making lawyers more efficient, but it’s also creating new risks and new questions that the courts are only beginning to answer. And be sure to ask your counsel how they are using AI tools.