The first mistake people make about jury simulation is assuming it's a demographics problem. Get the age right, the ZIP code right, the education level right, and you've got a juror. You do not. You've got a census row. A census row doesn't deliberate. It doesn't get stubborn at hour three. There are no emotions, biases. It doesn't decide, against the evidence, that it simply doesn't like your client.
The gap between a census row and a person is where MockJury does its work. It's also the part that's hardest to see, because when it's done well the result is authentic: a juror persona who answers a question the way an actual human being from that county would answer it, with all the texture and contradiction that implies. This post is about what sits underneath that, the 100+ attributes given to each persona and how they work together to react like a human.
Demographics tell you who, not how
Every synthetic juror starts with a demographic foundation, and that foundation has to be right. Our personas are drawn to match the actual composition of each US county not a national average, not a stereotype, but the real distribution of age, race, education, income, household structure, and the dozen other axes the US Census measures. If you're trying a case in Middlesex County, Massachusetts, your panel should look like Middlesex County, because Middlesex County is who decides your case.
But demographics only tell you who someone is on paper. They are weakly predictive of how someone actually reasons. Two 54-year-old college- educated women from the same suburb can sit on the same jury and reach opposite conclusions about the same evidence — and if your model treats them as interchangeable because their census rows match, your model is useless precisely when it matters most. The demographic layer is the floor. The building goes up from there.
The psychology is where human decision-making starts to show structure
What makes a person's reasoning predictable isn't their demographics. It's their psychology, the stable structures that govern how they take in information, what they trust, what they find threatening, and how they decide what's fair. The research on this is well known.
A few of the frameworks we build on, named plainly because they're published science and anyone serious in this space already knows them:
Personality structure. The HEXACO model decomposes personality into dimensions that predict, among other things, how much weight a person gives to fairness, how readily they defer to authority, and how they respond to emotional appeals. A juror high in one dimension and low in another hears your opening differently than the juror beside them.
Moral foundations. People don't share a single definition of "right." Some reason primarily from harm and care; others from loyalty, authority, or sanctity. Moral Foundations Theory maps this, and it matters enormously in a courtroom, because a plaintiff's story that lands as a harm narrative for one juror lands as a violation of personal responsibility for another. Same facts. Opposite verdicts.
Belief in a just world. Some people carry a deep conviction that the world is fundamentally fair, that people generally get what they deserve. That single attitude quietly shapes how a juror hears an injury claim. A strong just-world believer is predisposed to wonder what the plaintiff did to end up hurt. Knowing where a juror sits on that axis tells you where your contributory-negligence exposure really lives.
How people process numbers. When you ask a jury for two million dollars, they don't hear two million dollars. They translate it into a rough sense of "a lot" or "reasonable" or "outrageous," and that translation depends on the juror's numeracy and their relationship to large sums. The research on how jurors actually handle damages figures is some of the most counterintuitive in the field, and it's central to how our personas respond to a number.
Similarity to the parties. Forty years of research on defensive attribution shows that a juror who sees themselves in a party ( situationally, not just demographically) assigns responsibility differently than one who doesn't. A juror who has been the person behind the wheel reasons differently about the driver than a juror who has only ever been the pedestrian.
None of these is a secret. You can read all the research and publications this weekend. What you can't read this weekend, and what took us a great deal longer than a weekend, is how they fit together.
There's no exact formula to being human
A juror isn't a list of independent scores. A real person's traits are correlated, and they're correlated in ways that aren't obvious and aren't uniform across populations. The just-world believer who is also high in a particular moral foundation is a recognizable kind of person; the same just-world score paired with a different moral profile is a completely different person. Get the correlations wrong and you produce jurors who are internally incoherent — a collection of traits that no actual human would hold simultaneously. They answer questions in ways that feel subtly, uncannily wrong in a way you sense before you can explain.
Getting the correlations right so that a generated persona hangs together as a person, with a worldview whose parts reinforce each other the way a real worldview does is the actual work. It's where the research stops being a reference list and starts being engineering. It's the difference between a juror who recites attitudes and a juror who has them.
Why a persona has a past
The last layer is the one I'm most convinced by, and the one that is hardest to do well: a person is not just a set of traits. A person is a set of traits plus a history that explains them.
The research on narrative identity holds that people understand themselves through the stories they tell about their own lives — and critically, that the shape of those stories predicts behavior. Someone who narrates a hard chapter of their life as a redemption — "it was terrible, but I came through it stronger" — reasons differently from someone who narrates the same kind of event as a contamination — "things were fine until that ruined everything." Two people with identical personality scores and identical demographics can carry opposite narrative arcs, and those arcs show up in how they judge a stranger's misfortune from a jury box.
So our personas have pasts. Not decorative backstory — load-bearing history, structured so that a juror's life experience informs the way they hear your case. When a synthetic juror tells you why a plaintiff's story does or doesn't move them, the answer is grounded in a life, because that's how it works in a real deliberation room.
What this buys you
The demographic floor, the psychological structure, the correlation work, the narrative history, exists for one reason: so that when you ask a MockJury panel a question, the answers behave the way real people behave. Differentiated. Sometimes frustrating. Occasionally surprising in the specific way that makes a trial lawyer sit up, because the surprise is the signal — it's the panel telling you something about your case you hadn't seen.
That's the bar. Not "a chatbot that role-plays a juror." A panel built from the best available science about how people make decisions, designed to react to your case in ways that resemble the patterns and tensions you would expect from that venue. The methodology is in service of that, and we'll continue to improve it as we go.
We're in private beta with a small group of plaintiff PI firms. If you want to see it work on a real case, the form on the homepage is the way.