In a deposition there is no judge. The witness's own attorney is in the room but can do very little: objections are noted for the record and the witness answers anyway. A deposition runs up to seven hours in a day under Rule 30(d)(1), and each side may take ten depositions without leave of court.
Everyone in that room has done this hundreds of times, except, usually, the witness.
In the federal courts' 2024 fiscal year, district courts terminated 396,635 civil cases and tried 0.83% of them: 3,295 trials, 1,162 of those before a jury (Judicial Business 2024). Marc Galanter's The Vanishing Trial put the same rate at 11.5% in 1962. State courts are lower still: 0.21% in California in 2015.
For more than 99% of civil cases, the deposition transcript and video are the only testimony under questioning the case will ever produce. Summary judgment briefs quote it. Mediators read it. The settlement number is set against it.
Everything about the day is bought except the witness
The Federal Judicial Center's 2009 survey of closed federal civil cases found a majority of attorneys on both sides reporting at least one non-expert deposition; in cases that had any, plaintiffs took a mean of 3.8 and defendants 2.6. Nobody counts depositions nationally — courts record filings, not discovery events — but the industry that transcribes them can be counted: court reporting is a $1.66 billion business in the United States this year, and a single vendor reports handling more than 300,000 depositions annually. A deposition with a reporter, a transcript, and video runs from about $1,000 to over $5,000 in hard costs.
Every one of those line items is paid without hesitation. The witness's readiness is the one input that is not bought. It comes out of the defending attorney's own hours, and those are the hours a week runs out of.
What decides a deposition is delivery, not knowledge
Cutler, Penrod and Stuve ran a videotaped-trial experiment on 321 subjects in 1988, manipulating ten features of an eyewitness identification — disguise, weapon visibility, lineup instructions, and witness confidence among them. One of the ten reliably moved mock jurors' judgments of culpability: confidence. Brewer and Burke found the same shape for consistency in 2002 — jurors say inconsistency matters, but its effect was weak, while confidence strongly influenced judgments whether or not the testimony was consistent.
Confidence is not a dial you turn to maximum. Cramer, Brodsky and DeCoster showed in 2009 that the relationship is curvilinear: moderate confidence produced the highest credibility ratings, and those ratings went on to predict sentencing outcomes. Too little reads as unreliable. Too much reads as coached.
These are mock jurors, mostly in eyewitness-identification settings, and no study has run the same design on deposition witnesses. But the mechanism they describe — that the audience judges the performance and not the underlying knowledge — is precisely what a defending attorney is worried about when the camera is on.
Preparation moves delivery. Reading about it moves it less.
Boccaccini, Gordon and Brodsky published the closest thing to a controlled trial of witness preparation in 2005. Mock criminal defendants testified twice on video with about half receiving preparation in between; eight real public-defender clients were recorded before and after. Prepared witnesses scored better on eleven delivery behaviors — posture, fidgeting, sustained eye contact, hedging — and were rated more credible and less likely to be guilty. The authors also flag an unintended effect: preparation reduced expressiveness.
Wheatcroft and Ellison cross-examined sixty mock witnesses in 2012, half of whom had read a guidance booklet first. The prepared group answered more accurately, made fewer errors on complex questions, and were more willing to ask the examiner for clarification. The complex questions that tripped the unprepared — compound questions, negatively phrased questions, jargon — are the standard deposition toolkit.
The honest complication is the 2020 follow-up by Gous and Wheatcroft, which we include because it cuts against us. Questioning style dominated everything: witnesses asked non-directive questions got 15.6 of 20 answers right, against 12.0 for directive leading questions. Preparation moved accuracy in the right direction, 13.15 unprepared to 14.40 with the booklet, but at twenty subjects per cell it landed at p = .07.
Read the three together and the finding is not "preparation works." It is that reading about hostile questioning does a little, and being asked hostile questions does most of the damage. That is an argument for rehearsal over a handout, and against overclaiming from any study where the preparation was a piece of paper. Michael Boccaccini's 2002 review found the science of courtroom credibility well understood and almost nobody testing whether preparation changes it. Twenty-four years later that is still close to true.
Two hours of preparation per hour of testimony. Most witnesses get less than one.
The ABA's own guidance in Prepping a Witness is to spend at least twice as long preparing as the client will spend testifying, starting two to four weeks out. For a seven-hour deposition, that is fourteen attorney-hours.
Almost nobody gets fourteen. In our conversations with litigators the pattern is consistent, and it tracks what the case is worth. At the volume end of personal injury and insurance defense, preparation is thirty to ninety minutes, frequently the morning of, frequently a junior associate, frequently a do's-and-don'ts handout with no mock questioning at all. In the middle — commercial, employment, medical malpractice, serious injury — it is two to six hours with the defending lawyer and some real practice questions if there is time. At the top, a key witness gets videotaped mock sessions with a partner playing opposing counsel and a trial consultant in the room. Those consultants bill $400 to $1,500 an hour, and a one-day mock exercise at a major litigation-consulting firm runs about $30,000.
The litigators we talk to say the mock questioning is the part of preparation that matters most, and the part that gets cut when the week is full. It is not a disagreement about what works. A partner's afternoon is the most expensive thing a firm owns, and a case never has enough of them.
So the ordinary witness — the plaintiff, the employee, the nurse, the small-business owner, the person who will do this once in their life — has usually been told what a deposition is like, and often walked through the questions they will get and how to answer them. What most have not had is someone across the table trying to make them slip: pressed on a hedge, asked the same question a third time, handed their own email and told to explain it.
What we built
Truss runs the practice deposition. The witness joins from a private link and is examined by a live AI examiner that works from the attorney's case background, strategy notes, documents, and plan, and presses the way opposing counsel will. The attorney gets the recording, the transcript, and a report on where the witness held and where they gave ground, and the witness can go again as many times as it takes, without a partner's afternoon.
It drills composure, never content. What the witness says is theirs and their attorney's; Truss does not coach testimony or judge the merits of an answer. The record stays with the firm, nothing a firm puts in trains a model, and the security page says where every piece of it lives and for how long.
We are three founders from UT Austin, two of us raised in a family of trial lawyers, now in San Francisco as part of Y Combinator's Fall 2026 batch. Truss is live with litigation firms now. If you have a witness to prepare, get in touch.