Skip to content
Search resources Open menu

Jurors and Jury Decision-Making

Scientific Jury Selection, Trial Consultants, and AI Tools

Published

Snap Shot

  • The question: Do scientific jury selection, trial consultants, online juror research, and AI tools predict juror decisions well enough to justify their cost and risks?
  • What the research shows: Results are modest and mixed, and the value of scientific selection likely varies with the case. Demographics are weak and inconsistent predictors; case-specific attitudes predict somewhat better, though still modestly. In two recent studies, general-purpose language models did not reproduce human judgments: they predicted individual mock jurors’ ratings less accurately than a human benchmark, and, acting as jurors, they convicted far less often than people did.
  • What it does not show: No study shows that scientific selection reliably changes verdicts in real trials, and none shows it never helps. Whether online juror research improves prediction has not been tested, no published independent validation of commercial AI tools was located, and no method has been shown to predict reliably how a particular juror will evaluate the evidence.
  • The law: Batson and J.E.B. apply to every party, and a strike recommended by a consultant or software is the party’s strike; Foster shows selection notes can become evidence. ABA Formal Opinions 466 and 517 address online juror research and discriminatory strikes. In Arizona, a juror may be removed for cause, by good-faith stipulation, or for statutory disqualification, so no profile by itself supports a challenge, and A.R.S. § 21-312 restricts juror information.
  • Issues for each side: Prosecutors should keep juror research within ethical limits and base cause challenges on what jurors actually said; defense counsel should seek voir dire reaching case-specific attitudes, recognizing that Batson and the ABA opinion apply equally to defense reliance on consultants and software.

Lawyers have always tried to predict how prospective jurors will respond to a case. For most of the history of jury trials those predictions rested on experience, intuition, and folklore about occupations, appearance, and background. Since the early 1970s a professional field has offered a different approach: the use of surveys, statistical modeling, focus groups, and other social science methods to identify the attitudes and experiences associated with favorable or unfavorable reactions to a case. The practice is usually called “scientific jury selection,” and the professionals who provide it are trial consultants or jury consultants. More recently, vendors and researchers have begun applying online research tools and artificial intelligence (AI), including large language models, to the same tasks.

These methods reach court in several ways. They shape the questions counsel ask in voir dire and in written questionnaires, the challenges counsel make, and the arguments counsel prepare. They raise questions of professional ethics when counsel research jurors online, and questions of constitutional law when a recommended strike corresponds to a juror’s race or sex. And they raise a recurring empirical question: do they work well enough to justify what they cost and what they risk?

This paper describes the origins and methods of scientific jury selection, summarizes the research on its effectiveness, explains why case-specific attitudes are better predictors than demographic characteristics, and reviews the emerging research on AI tools. It then addresses the ethical and legal limits that apply in federal, military, and Arizona courts, where the abolition of peremptory challenges has changed what jury selection tools can be used for.

Origins

Scientific jury selection is usually traced to the 1972 federal conspiracy trial in Harrisburg, Pennsylvania, of antiwar activists known as the Harrisburg Seven. A team of social scientists led by sociologist Jay Schulman, working with the defense, surveyed and interviewed residents of the trial district to learn which characteristics and attitudes were associated with sympathy toward the defendants, and used the results to develop a profile of jurors expected to be more and less receptive to the defense. The jury deadlocked on the central conspiracy charge and returned convictions only on lesser counts. The team described its methods in a widely read article (Schulman et al., 1973). The contribution of the survey work to that outcome cannot be isolated, because the evidence, the conduct of the trial, and the jurors’ own deliberations also bore on the result; the same limitation applies to every reported consulting success in a single case.

The Harrisburg work drew early criticism on both empirical and ethical grounds. Critics questioned whether survey-based predictions were accurate enough to matter and whether well-funded parties’ access to social science expertise would create an imbalance in the adversary system (Saks, 1976). Trial consulting nonetheless grew into an established industry serving civil litigants, criminal defendants, and, less often, prosecutors (Strier, 1999).

Methods

Community attitude surveys remain the core method. A consultant presents a summary of the case to a sample of jury-eligible residents of the trial venue, asks how they would evaluate it, and measures their demographic characteristics, general attitudes (for example, toward police, corporations, or the criminal justice system), and case-specific attitudes and experiences. Statistical analysis then identifies the variables that best predict reactions to the case. The results may be used to draft voir dire questions and juror questionnaires, to rate prospective jurors, or to support a motion for change of venue based on pretrial publicity. Reviewers caution that surveys are better at identifying which kinds of people tend to react in certain ways than at predicting the vote of any individual juror (Kovera et al., 2003).

Focus groups and mock trials present a condensed version of the case to community participants who discuss or deliberate while counsel or consultants observe. They are used less to select jurors than to test themes, identify confusing evidence, and learn how laypeople talk about the issues. Shadow juries are panels of people matched to the seated jury who observe the actual trial and report their reactions as it proceeds. Both methods use small, nonrandom samples, and their participants know that their decisions have no consequences. They can reveal how some people reason about a case; they cannot estimate how often the seated jury will reason the same way.

In court, consultants may observe voir dire and rate jurors against a profile; those ratings depend on what jurors disclose and on how much questioning the court allows.

Research on Effectiveness

Evaluating scientific jury selection is difficult. In a real trial there is no way to know what verdict a different jury would have reached, and reported successes cannot separate the effect of selection from the strength of the evidence and the skill of counsel. Researchers have therefore relied on indirect approaches: comparing jurors chosen by different methods, testing whether the variables consultants measure predict verdicts in mock trials, and surveying jurors after trial.

The results have been modest and mixed. In an experiment comparing conventional attorney selection with a systematic social science method across four simulated criminal cases, the social science method predicted jurors’ responses better when the relationships between juror characteristics and verdicts were strong, and the conventional method did better when those relationships were weak or absent; overall, neither method was superior (Horowitz, 1980). Reviews of attorney jury selection folklore have concluded that many of the traditional rules of thumb about which kinds of jurors favor which side have little empirical support (Fulero & Penrod, 1990; Kovera et al., 2003). A review of 27 telephone surveys, 9 focus groups, and 2 post-trial juror studies conducted for actual cases concluded that the usefulness of scientific selection depends in part on the type of case (Seltzer, 2006).

The most comprehensive book-length review devoted a chapter to the overall effectiveness of scientific jury selection and described juror personality traits and general legal attitudes as weak and inconsistent predictors of verdicts (Lieberman & Sales, 2007). A later summary by the same lead author reached a similar conclusion: the method may have some utility, but flaws in the existing research prevent clear conclusions, and high-quality evaluation studies are still needed (Lieberman, 2011).

No study shows that scientific jury selection reliably changes verdicts in real trials, and none shows that it never helps. The fairest summary is that its value is uncertain, likely varies with the case, and is greatest where a case-specific attitude is both strongly related to how people view the evidence and measurable in the questioning the court permits.

Demographics Versus Case-Specific Attitudes

The most consistent finding in this literature concerns what predicts juror judgments. Demographic characteristics such as age, sex, race, education, income, and occupation are weakly and inconsistently related to verdicts. Associations that appear in one case or sample often fail to appear in another, and their direction can depend on the type of case (Kovera et al., 2003). A meta-analysis of studies relating juror characteristics to guilt judgments concluded that most individual characteristics, including demographic characteristics, had small or negligible associations with guilt judgments, and that a limited number of attitudinal and personality measures showed somewhat stronger, though still modest, associations (Devine & Caughlin, 2014).

Attitudes that bear directly on the issues in the case tend to predict better than general characteristics. Examples include attitudes toward the death penalty in capital cases, attitudes toward civil litigation in tort cases, and beliefs about sexual assault in sexual offense trials (Kovera et al., 2003; Willmott et al., 2026). Even these associations are modest and vary from study to study, but they explain why the field has shifted from demographic profiling toward measuring attitudes and experiences related to the case.

The research on sexual offense trials illustrates both the promise and the limits of attitude measurement. Scores on scales measuring acceptance of what researchers call rape myths have predicted mock jurors’ verdicts before and after deliberation, while gender predicted verdicts only before deliberation in one of two studies (Willmott et al., 2026). Other researchers argue that screening jurors by such scales would be impractical, easy for jurors to defeat by giving socially desirable answers, and likely to make juries less representative, and the debate is unresolved. The Arizona Forensic Services resource on counterintuitive testimony reviews that dispute in detail in its paper “Correcting Juror Misconceptions: Directions, Expert Evidence, and Juror Screening.”

The legal and scientific implications point in the same direction. Group-level associations between demographics and verdicts are small, inconsistent across studies, and cannot predict how a particular juror will vote; the law independently forbids using race, ethnicity, or sex as a basis for removing jurors. Questioning that asks about attitudes and experiences related to the case is both more informative and legally appropriate.

Online Research on Jurors

Lawyers and consultants now often review jurors’ public online presence to supplement voir dire. The American Bar Association’s ethics committee has concluded that, unless limited by law or court order, a lawyer may review a juror’s or prospective juror’s public internet presence, but may not, personally or through another person, send an access request (such as a request to connect or “friend”) to view information that is not public, because such a request is a prohibited ex parte communication. The committee also concluded that an automatic notification sent by a platform telling a juror that the lawyer viewed the juror’s profile is not a communication by the lawyer, and that a lawyer who discovers criminal or fraudulent juror conduct related to the proceeding must take reasonable remedial measures, including, if necessary, disclosure to the tribunal (ABA Standing Committee on Ethics and Professional Responsibility, 2014). Courts and state ethics authorities may impose additional limits.

Whether online research improves the prediction of verdicts has not been tested. Public posts are a selective sample of a person’s views, names can be matched to the wrong person, and information about political, religious, or other affiliations invites the group-based inference the research shows to be unreliable and the law in some instances forbids. Online research may be most useful for identifying undisclosed connections to the parties, prior experiences directly relevant to the case, or answers on voir dire that appear inaccurate, which can then be raised with the court.

Artificial Intelligence and Algorithmic Tools

Commercial services now market AI tools for jury work, including software that ranks prospective jurors and programs that use language models to simulate juror reactions (Harrington & Stillwell, 2026; ABA Standing Committee on Ethics and Professional Responsibility, 2025). Their accuracy has received little independent study. The published study most directly relevant to juror selection compared three general-purpose language models with about 1,200 jury-eligible U.S. adults, recruited online, who had rated on a 0–10 scale how much various versions of a codefendant’s redacted confession increased the likelihood of guilt (Harrington & Stillwell, 2026; Stillwell, 2026). Each model was given every participant’s demographic profile and asked to predict that person’s rating. As a benchmark, the authors used the human “noise floor”: a typical participant’s rating differed from the group average by about 2.0 points. The models’ predictions missed participants’ actual ratings by about 2.5 to 3.4 points on average, and each model’s error was significantly larger than the human benchmark. One model showed no significant directional bias; the other two erred on average in opposite directions, one predicting ratings lower than the humans gave and one higher. The authors reported no consistent ideological skew in any model. In earlier exploratory work, the models often declined the task, produced implausible jurors, and tended to script jurors’ reasoning according to their demographic identities. In a supplementary analysis, a smaller open model fine-tuned on the human data reportedly performed better, though figures for that analysis are not reported here. The authors concluded that general-purpose models cannot yet simulate individual jurors, cautioned that commercial jury-selection tools have not been validated (no published independent validation of commercial tools was located for this paper), and suggested that models trained on real human data offer a more credible path.

The study has limits its authors acknowledge. It used an online sample whose participants selected themselves and was not representative on income or education; some subgroups were small; it addressed a single judgment task, about a vignette without other evidence, instructions, or deliberation; and model versions tested are no longer available, so the work cannot be replicated exactly. Its human data come from a companion study by one of its authors (Stillwell, 2025), and the comparison has not been independently replicated. It nonetheless shows a specific risk. A model that predicts juror reactions from demographic inputs may reproduce demographic stereotypes, which the broader literature shows to be weak predictors and which the law forbids as a basis for removing jurors.

A peer-reviewed study approached the question differently. Researchers had three general-purpose language models act as mock jurors in a replication of a published sexual assault experiment that varied the defendant’s race and social status, the number of allegations, and the delay in reporting, and collected 2,304 responses (Sun et al., 2025). Compared with the original human participants, the models gave a slightly higher average probability of guilt (about 64 vs. 59 on a 0–100 scale) but returned guilty verdicts far less often (21% vs. 49%). Like the humans, they judged White defendants more harshly and assigned more guilt when there were multiple allegations; unlike the humans, who were little affected by reporting delay, they assigned more guilt when the delay was shorter. The study tested models as decision makers, not as predictors of particular jurors, and used one written case without deliberation. It shows that a model’s responses to case facts can differ from people’s in ways that are not obvious in advance.

A related question is how much people defer to algorithmic output. In two experiments, mock jurors’ judgments shifted toward an algorithm’s recommendation even when it conflicted with the testimony, and a warning about AI errors and biases reduced that influence (Álvarez & Matute, 2026). In two preregistered online experiments with about 1,700 Japanese adults each, participants deciding sentences in murder vignettes deferred to AI judgments as much as to judges’ judgments (Watamura et al., 2025). These studies concern advice given to decision makers, not jury selection, and whether lawyers give juror-ranking software similar weight has not been studied.

Professional guidance has begun to address these tools. The ABA ethics committee has advised that lawyers using generative AI must have a reasonable understanding of the capabilities and limitations of the particular tool, acquired personally or by drawing on others with the relevant expertise (ABA Standing Committee on Ethics and Professional Responsibility, 2024). In a 2025 opinion on discrimination in jury selection, the committee concluded that a lawyer who knows or reasonably should know that a peremptory challenge constitutes unlawful discrimination violates Model Rule 8.4(g); that a lawyer may not rely on a client’s direction, a consultant’s advice, or AI software’s recommendation to excuse jurors in a way the lawyer knows or should know is unlawful; and that a lawyer using an AI juror-ranking program should exercise due diligence to acquire a general understanding of its methodology (ABA Standing Committee on Ethics and Professional Responsibility, 2025). The opinion ties a violation to strikes that other law, such as Batson, forbids, and Model Rule 8.4(g) has not been adopted in the same form in every jurisdiction, so counsel should consult the rule in force where they practice.

Limits of the Evidence

The research on scientific jury selection is old relative to current practice; much of it dates from the 1970s through the 1990s and used methods, venues, and case types that may not generalize. Most studies are mock-jury experiments or analyses of a small number of cases, often without deliberation. Consultants’ own case work is rarely published, so successes and failures in practice cannot be counted. Research on online juror investigation has not tested whether it improves selection, and research on AI tools consists of a small number of recent studies of general-purpose models rather than the commercial products marketed to lawyers. Most fundamentally, every method in this field estimates group tendencies. None can determine how a particular prospective juror will evaluate the evidence, and none can show that a particular jury was or was not fair.

Questions to Ask About This Research

When the Harrisburg Seven outcome (Schulman et al., 1973) or another single-case success is cited to show that scientific selection works, ask whether the conclusion follows from the design, since the evidence, the conduct of the trial, and the deliberations also bore on that result. When the finding that rape-myth acceptance scores predicted mock jurors’ verdicts (Willmott et al., 2026) is offered to support excusing a prospective juror in a sexual offense case, ask who was studied and whether the finding applies to this person: the studies were mock trials, the associations are modest, and group-level scores cannot show how a particular juror will vote. When the language-model comparison (Harrington & Stillwell, 2026) is cited to reject or to endorse an AI tool, ask how much evidence it rests on and whether it has held up: one study of individual-juror prediction by general-purpose models, errors of about 2.5 to 3.4 points against a human benchmark of about 2.0, no independent replication, and no test of commercial products; the separate study of models acting as jurors (Sun et al., 2025) tested a different task. When a consultant’s profile rests on demographics, ask how strong those associations are in absolute terms (Devine & Caughlin, 2014). These questions are explained in “Reading Research Claims: A Guide for Attorneys” on this site.

Jury Selection Tools and the Law

Federal and Military Standards

The Sixth Amendment guarantees an impartial jury in criminal cases, and the Jury Selection and Service Act requires that federal juries be selected at random from a fair cross section of the community (28 U.S.C. §§ 1861–1878; see Taylor v. Louisiana, 1975; Duren v. Missouri, 1979). Federal Rule of Criminal Procedure 24(a) allows the court to examine prospective jurors itself or to permit the attorneys to do so, and the scope of questioning is largely within the trial court’s discretion (Rosales-Lopez v. United States, 1981). Because the information consultants and AI tools can use depends heavily on what the court allows counsel to ask, these tools are constrained by the voir dire the court permits.

The Equal Protection Clause forbids peremptory strikes based on race (Batson v. Kentucky, 1986) or sex (J.E.B. v. Alabama ex rel. T.B., 1994). The prohibition applies to the party exercising the strike regardless of the source of the recommendation; a strike suggested by a consultant or a ranking program is the party’s strike. Foster v. Chatman (2016) relied on the prosecution’s jury-selection notes, including markings of jurors’ race, as evidence of discriminatory purpose, and Flowers v. Mississippi (2019) emphasized that courts consider the full history and pattern of a party’s strikes. Consultant work product and tool outputs that rely on race or sex may therefore become evidence in a Batson inquiry. The ABA’s 2025 opinion addresses the corresponding ethical duty.

In courts-martial, members (who serve the jury’s function) are detailed by the convening authority under the criteria of Article 25, UCMJ, rather than drawn at random from the community; since December 23, 2024, the convening authority details a larger pool that the military judge then randomizes. Each side initially has one peremptory challenge, with one more when new members are detailed (R.C.M. 912(g)), and military judges must liberally grant defense challenges for cause (United States v. Clay, 2007). Batson applies to peremptory challenges in courts-martial (United States v. Moore, 1989). Published information on consultant use in courts-martial was not located, and the selection rules that shape their usefulness are described in this hub’s paper on court-martial panel selection.

Arizona

Arizona eliminated peremptory challenges in criminal and civil trials effective January 1, 2022 (Ariz. Sup. Ct. Order R-21-0020). Removal of a prospective juror now depends on a challenge for cause, which under Arizona Rule of Criminal Procedure 18.4(b) requires the court to excuse a juror when there are reasonable grounds to believe the juror cannot render a fair and impartial verdict. Under A.R.S. § 21-211, bias or prejudice for or against a party is a statutory ground for disqualification. Arizona voir dire is governed by Rule 18.5.

This change alters what jury selection tools are for. A juror’s rating on a consultant’s scale is not a ground for excusal, and no statistical profile, however accurate, can by itself support a challenge for cause. The tools that matter in Arizona are those that help counsel and the court identify, through questioning and questionnaires, specific attitudes or experiences that bear on a juror’s ability to be fair in the particular case. Survey and focus-group research can help identify which topics are worth asking about and how to phrase questions so that jurors answer candidly; the challenge itself must rest on what the juror says or what the record shows.

Arizona law also protects juror information. Under A.R.S. § 21-312, records containing jurors’ biographical information are closed to the public, must be returned after jury selection, and may not be further disclosed by the parties or their attorneys. Counsel considering the use of an outside consultant or software service to analyze questionnaire responses should consider whether that use is consistent with the statute and any order of the court. Arizona’s rules of professional conduct, including ER 3.5 on contact with jurors, apply to online juror research.

Special Issues in Legal Practice

The principles described above apply with equal force to every party in a legal proceeding. The science operates the same way regardless of who is asking the questions, and the research summarized here does not by itself favor either side. What differs between the parties is not the science but the vantage point — the particular places where these issues tend to surface, and the particular errors each party is best positioned to guard against. The two summaries below identify those recurring issues. They are offered for orientation, not as litigation strategy; in every case the controlling question is what the evidence in that case will, and will not, support, and the expert’s role is to inform the trier of fact rather than to advocate for either side.

Special Issues for Prosecutors

Research indicating that case-specific attitudes predict judgments somewhat better than demographics is consistent with using questionnaires and voir dire to identify attitudes and experiences that may affect a juror’s ability to evaluate the evidence fairly, such as strong views about police, forensic evidence, or the type of offense charged; surveys and focus groups may help identify topics worth asking about, although their effect on verdicts has not been established. In Arizona, since peremptory challenges were abolished, a juror may be removed for cause, by good-faith stipulation, or under statutory disqualification, so attitudes the juror discloses are the principal basis for a contested removal. Prosecutors are often best positioned to ensure that juror research by their offices and investigators stays within the limits described in ABA Formal Opinion 466 and any court order, and that the record of a cause challenge reflects what the juror actually said rather than a profile or rating.

The research also counsels caution. A tool or consultant that ranks jurors by demographic characteristics relies on associations that are weak and inconsistent, and strikes that track race or sex are unlawful regardless of who recommended them. Foster shows that selection notes and work product can become evidence of purpose. In military practice, where defense challenges are liberally granted, the research on attitudes supports thorough, neutral questioning rather than reliance on members’ general assurances of fairness.

Special Issues for Defense

The research supports seeking voir dire and questionnaires that reach case-specific attitudes and experiences, since those are the variables most likely to bear on a juror’s evaluation of the evidence and the information consultants and tools most need. Community survey research may also support motions concerning pretrial publicity or venue, subject to the same limits of method and sampling described above. Defense counsel are often best positioned to make the record that supports a cause challenge, and in courts-martial to invoke the liberal grant mandate with specific facts rather than general assertions.

The research also counsels caution. No study shows that scientific jury selection or AI tools reliably change verdicts, and in the one published comparison of individual-juror prediction, general-purpose language models predicted mock jurors’ judgments less accurately than the variation among the participants themselves and, in exploratory work, tended to script jurors’ reasoning by demographic identity; in a separate study, models acting as jurors responded to some case facts differently from people. Batson and J.E.B. apply to defense strikes as well as prosecution strikes (Georgia v. McCollum, 1992), and the ABA’s 2025 opinion applies to defense counsel’s reliance on consultants and software in the same way. Juror information shared with outside services may be subject to confidentiality restrictions, and online research that crosses into contact with a juror may violate the rules of professional conduct.

This paper is for educational purposes only and is not legal advice. The author is a psychologist, not an attorney; summaries of law are general and must be verified against current authority or with a licensed attorney.

References

Álvarez, M., & Matute, H. (2026). Susceptibility to algorithmic recommendations in simulated jury decision making. Legal and Criminological Psychology. Advance online publication. https://doi.org/10.1111/lcrp.70068

American Bar Association Standing Committee on Ethics and Professional Responsibility. (2014). Lawyer reviewing jurors’ internet presence (Formal Opinion 466). American Bar Association.

American Bar Association Standing Committee on Ethics and Professional Responsibility. (2024). Generative artificial intelligence tools (Formal Opinion 512). American Bar Association.

American Bar Association Standing Committee on Ethics and Professional Responsibility. (2025). Discrimination in the jury selection process (Formal Opinion 517). American Bar Association.

Devine, D. J., & Caughlin, D. E. (2014). Do they matter? A meta-analytic investigation of individual characteristics and guilt judgments. Psychology, Public Policy, and Law, 20(2), 109–134. https://doi.org/10.1037/law0000006

Fulero, S. M., & Penrod, S. D. (1990). The myths and realities of attorney jury selection folklore and scientific jury selection: What works? Ohio Northern University Law Review, 17, 229–253.

Harrington, S., & Stillwell, H. (2026). Michael Scott is not a juror: The limits of AI in simulating human judgment. UNT Dallas Law Review: On the Cusp, 8(2), 73–131.

Horowitz, I. A. (1980). Juror selection: A comparison of two methods in several criminal cases. Journal of Applied Social Psychology, 10(1), 86–99. https://doi.org/10.1111/j.1559-1816.1980.tb00695.x

Kovera, M. B., Dickinson, J. J., & Cutler, B. L. (2003). Voir dire and jury selection. In A. M. Goldstein (Ed.), Handbook of psychology: Vol. 11. Forensic psychology (pp. 161–175). Wiley. https://doi.org/10.1002/0471264385.wei1110

Lieberman, J. D. (2011). The utility of scientific jury selection: Still murky after 30 years. Current Directions in Psychological Science, 20(1), 48–52. https://doi.org/10.1177/0963721410396628

Lieberman, J. D., & Sales, B. D. (2007). Scientific jury selection. American Psychological Association. https://doi.org/10.1037/11498-000

Saks, M. J. (1976). The limits of scientific jury selection: Ethical and empirical. Jurimetrics Journal, 17, 3–22.

Schulman, J., Shaver, P., Colman, R., Emrich, B., & Christie, R. (1973). Recipe for a jury. Psychology Today, 6(12), 37–44, 77–84.

Seltzer, R. (2006). Scientific jury selection: Does it work? Journal of Applied Social Psychology, 36(10), 2417–2435. https://doi.org/10.1111/j.0021-9029.2006.00110.x

Stillwell, H. (2025). The reasonable juror on trial: A Bruton-inspired reality check [Preprint]. SSRN. https://doi.org/10.2139/ssrn.5371115

Strier, F. (1999). Whither trial consulting? Issues and projections. Law and Human Behavior, 23(1), 93–115. https://doi.org/10.1023/A:1022378824280

Sun, Y., Zappalà, A., Di Maso, E., Pompedda, F., Nyman, T. J., & Santtila, P. (2025). Large language models (LLMs) as jurors: Assessing the potential of LLMs in legal contexts. Law and Human Behavior, 49(5), 440–458. https://doi.org/10.1037/lhb0000620

Watamura, E., Liu, Y., & Ioku, T. (2025). Judges versus artificial intelligence in juror decision-making in criminal trials: Evidence from two pre-registered experiments. PLoS ONE, 20(1), Article e0318486. https://doi.org/10.1371/journal.pone.0318486

Willmott, D., Boduszek, D., & Booth, N. (2026). Profiling jury bias in rape trials: Examining the link between juror characteristics and trial decisions. In E. Dowds, S. Weare, & S. Leahy (Eds.), Researching rape trials: Methods, ethics and empirical perspectives (pp. 127–144). Routledge. https://doi.org/10.4324/9781003613015-10

Legal Authorities

Batson v. Kentucky, 476 U.S. 79 (1986).

Duren v. Missouri, 439 U.S. 357 (1979).

Flowers v. Mississippi, 588 U.S. 284 (2019).

Foster v. Chatman, 578 U.S. 488 (2016).

Georgia v. McCollum, 505 U.S. 42 (1992).

J.E.B. v. Alabama ex rel. T.B., 511 U.S. 127 (1994).

Rosales-Lopez v. United States, 451 U.S. 182 (1981).

Taylor v. Louisiana, 419 U.S. 522 (1975).

United States v. Clay, 64 M.J. 274 (C.A.A.F. 2007).

United States v. Moore, 28 M.J. 366 (C.M.A. 1989).

U.S. Const. amend. VI.

U.S. Const. amend. XIV, § 1.

Jury Selection and Service Act, 28 U.S.C. §§ 1861–1878.

Article 25, Uniform Code of Military Justice, 10 U.S.C. § 825.

Ariz. Rev. Stat. Ann. § 21-211.

Ariz. Rev. Stat. Ann. § 21-312.

Fed. R. Crim. P. 24.

R.C.M. 503(a)(1), 912(g), Manual for Courts-Martial, United States.

Ariz. R. Crim. P. 18.4(b).

Ariz. R. Crim. P. 18.5.

Ariz. R. Sup. Ct. 42, ER 3.5.

Ariz. Sup. Ct. Order No. R-21-0020 (Aug. 30, 2021) (eliminating peremptory challenges, effective Jan. 1, 2022).

Model Rules of Pro. Conduct r. 3.5 & r. 8.4(g) (Am. Bar Ass’n).

Related topic papers: Voir Dire and the Detection of Juror Bias · Peremptory Challenges, Batson, and Arizona’s Abolition of Peremptories · Court-Martial Panel Selection: Article 25, Random Selection, and Challenges · Juror Demographics and Verdicts: What Predicts Guilt Judgments · Mock Jury Research: Methods, Realism, and Limits

Related topics from Arizona Forensic Services: Counterintuitive Testimony · Courts-Martial