AI in Jury Selection: Voir Dire Risk in Michigan (2026)
Jury-analytics software now promises to read a panel before a single question is asked. Feed it names, addresses, and social media, and it returns scored profiles predicting how each prospective juror is likely to lean. For a trial lawyer the appeal is obvious. The danger is just as real. A tool that quietly ranks people by race or sex does not relieve the lawyer of responsibility for a discriminatory strike. It can manufacture one. This guide explains how AI is being used in voir dire, what the new ABA ethics opinion requires, and how Michigan’s own rules and the constitutional limits of Batson still control the courtroom.
What AI Jury Tools Actually Do
Modern jury-selection products gather public and semi-public data on the venire, then apply predictive models to estimate attitudes, sympathies, and likely verdict leanings. Inputs can include voter and consumer records, litigation history, and a prospective juror’s social media footprint. The output is usually a ranking or a score that a lawyer can use to decide whom to strike with a peremptory challenge and whom to keep.
Used carefully, these tools are an extension of the jury consultant’s craft, faster and cheaper than a human team combing public records by hand. The problem is that a score is only as clean as the data and the model behind it. If a model has learned, even indirectly, that race, sex, or religion correlates with a verdict tendency, its rankings can encode exactly the discrimination the law forbids. The lawyer who acts on that ranking owns the result.
The New Ethics Line: ABA Formal Opinion 517
On July 9, 2025, the ABA Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 517, addressing peremptory challenges and the use of jury consultants and AI. The opinion’s core holding is direct. A lawyer who knows or reasonably should know that exercising a peremptory challenge constitutes unlawful discrimination violates the ethics rules, and carrying out that strike is not legitimate advocacy.
For AI specifically, the opinion warns that lawyers can unknowingly discriminate if a program ranks jurors using impermissible criteria. To manage that risk, a lawyer should conduct enough due diligence to gain a general understanding of the methodology the program uses. The opinion also makes clear that a lawyer may not hide behind a client’s instruction, a consultant’s advice, or an AI tool’s output. If the lawyer knows or should know the resulting strike is unlawful discrimination, the lawyer cannot make it.
Michigan’s Framework Has Not Changed
Whatever a software vendor markets, jury selection in a Michigan civil trial is still governed by MCR 2.511. Each party may peremptorily challenge three jurors under MCR 2.511(E), and voir dire exists to expose grounds for challenges for cause and to inform the intelligent use of peremptory challenges. MCR 2.511(F) bars discrimination in voir dire on the basis of race, color, religion, national origin, or sex, and it forecloses the argument that a lawyer was merely trying to build a balanced or representative jury.
Layered on top of the court rule is the constitutional floor. Batson v. Kentucky, 476 U.S. 79 (1986), held that peremptory challenges may not be used to strike jurors because of race. Although Batson arose in a criminal case, Edmonson v. Leesville Concrete Co., 500 U.S. 614 (1991), extended the same prohibition to civil trials, and J.E.B. v. Alabama ex rel. T.B., 511 U.S. 127 (1994), extended it to strikes based on sex. All apply in a Michigan civil case. When an opponent raises a Batson or J.E.B. objection, the striking lawyer must offer a neutral, non-discriminatory reason, and the court decides whether that reason is genuine or pretext. “The algorithm told me to” is not, by itself, a neutral reason. It can be an admission that the criteria were never examined.
| Limit | Source | What it forbids |
|---|---|---|
| Race-based strikes | Batson v. Kentucky, 476 U.S. 79 (1986) | Peremptory challenges on account of race |
| Race-based strikes (civil) | Edmonson v. Leesville Concrete Co., 500 U.S. 614 (1991) | Extends Batson to civil peremptory challenges |
| Sex-based strikes | J.E.B. v. Alabama, 511 U.S. 127 (1994) | Peremptory challenges on account of sex |
| Voir dire discrimination | MCR 2.511(F) | Strikes by race, color, religion, national origin, or sex |
| Challenge limits | MCR 2.511(E) | Three peremptory challenges per party in civil cases |
| Knowing discrimination | ABA Formal Op. 517 (2025) | Acting on AI output the lawyer knows is discriminatory |
The Bias Trap Hiding in the Score
The hardest part of AI jury selection is that the discrimination can be invisible. A model rarely announces that it weighed race. Instead it weighs ZIP code, magazine subscriptions, or the pages a person follows online, any of which can act as a close proxy for a protected characteristic. The score that lands on the lawyer’s screen looks neutral. The path that produced it may not be.
That is why ABA Opinion 517 frames the duty as one of reasonable diligence rather than perfection. A lawyer cannot audit a vendor’s neural network, but a lawyer can ask what data the tool ingests, whether protected characteristics or obvious proxies are used, and whether the vendor can articulate the basis for a score. A lawyer who asks nothing, then strikes the only three jurors of one race because the tool ranked them lowest, is exposed on both the constitutional and the ethics fronts. Our discussion of how courts have sanctioned lawyers for unverified AI output in AI hallucinations and court sanctions shows how quickly an unexamined tool becomes the lawyer’s problem.
Practical Guardrails for Trial Lawyers
AI jury tools are not prohibited in Michigan, and used with care they can sharpen voir dire. The goal is to capture the benefit without inheriting a discrimination claim:
- Vet the methodology. Ask the vendor, in writing, what inputs feed the model and whether race, sex, religion, or national origin, or proxies for them, influence the score. Keep the answer.
- Keep a human reason for every strike. Document a specific, neutral basis for each peremptory challenge that is independent of any AI ranking, so you can satisfy a Batson or J.E.B. challenge on the record.
- Do not delegate the decision. The score is input, not instruction. Under ABA Opinion 517, you cannot follow an AI tool, a consultant, or a client into an unlawful strike.
- Mind juror privacy and contact rules. Reviewing a juror's public internet presence is generally permissible under ABA Formal Opinion 466 (2014), but you may not send a connection or access request or take any step that communicates with a juror. Some courts also restrict online juror research, so confirm local practice before running a panel through a tool.
- Preserve the audit trail. If you use a tool, keep records of what it recommended and why you did or did not follow it. That file protects you if a strike is later questioned.
Frequently Asked Questions
Is it legal to use AI for jury selection in Michigan?
There is no rule banning AI jury-analytics tools in Michigan. They are permitted, but the lawyer remains fully responsible for every peremptory challenge. The tool cannot be used to make a strike that would violate Batson, J.E.B., or MCR 2.511(F).
What does ABA Formal Opinion 517 require?
Issued July 9, 2025, the opinion holds that a lawyer may not exercise a peremptory challenge the lawyer knows or reasonably should know is unlawful discrimination, even if a client, consultant, or AI tool recommends it. With AI, the lawyer should perform enough due diligence to understand the program’s general methodology.
Does Opinion 517 bind Michigan lawyers?
Not directly. The opinion interprets ABA Model Rule 8.4(g), which Michigan has not adopted verbatim. It is persuasive guidance. In Michigan the enforceable limits on discriminatory strikes are the constitutional rules of Batson and J.E.B. and the anti-discrimination command of MCR 2.511(F).
How can an AI tool cause a Batson problem?
A model can rank jurors using data that acts as a proxy for race or sex, such as neighborhood or online activity, even when it never names the protected trait. If a lawyer strikes jurors based on that ranking, the pattern can look like discrimination, and “the software ranked them low” is not a neutral explanation under Batson.
How many peremptory challenges do I get in a Michigan civil trial?
Each party generally has three peremptory challenges under MCR 2.511(E). Special rules adjust the count when multiple parties with adverse interests are aligned on the same side, so check the rule and confirm with the court before trial.
What should I keep in the file if I use a jury tool?
Keep the vendor’s written description of the tool’s inputs and methodology, your own neutral reason for each strike, and a record of where you followed or rejected the tool’s recommendation. That documentation helps you answer a Batson or J.E.B. challenge and shows the diligence Opinion 517 expects.
Questions about trial strategy or a verdict in your injury case?
Attorney Manny Chahal handles Michigan personal injury and no-fault cases statewide at no charge to start. No fee unless we recover.
Call 1-844-624-2425