Most Attorneys Prep Depositions Wrong. AI Exposes Why.

Here's an uncomfortable truth: the average personal injury attorney spends 8 to 14 hours preparing for a single deposition, and roughly 60% of that time is spent re-reading material they've already read. Medical records. Prior statements. Treatment timelines. The work isn't analysis. It's recall. And recall is the one thing software does better than humans, every time, without exception.

That's why I'm convinced the firms still running deposition prep the old way - paralegal builds a binder, associate reads the binder, partner skims the binder on the morning of - are about to lose ground fast. Not because AI replaces the lawyer's judgment. It doesn't. But it changes the math on what a lawyer can actually walk into the room knowing. And in PI litigation, where 95% of cases settle and the deposition transcript drives the settlement number, what you know going in is the case.

The Real Bottleneck Is Cross-Referencing, Not Reading

The traditional deposition prep workflow assumes the bottleneck is reading volume. It isn't. The bottleneck is connecting facts across documents.

Consider a typical soft-tissue case. You have 400 pages of medical records from six providers, a 50-page EUO transcript, recorded statements, ISO claim history, social media exports, and prior pleadings. A skilled associate can read all of it in a day. What they cannot reliably do, even given a week, is hold every contradiction in working memory simultaneously: the chiropractor's note from week three says the plaintiff reported "no prior back issues," but the orthopedic intake from 2019 mentions a lumbar strain, and the recorded statement to the adjuster references "stiffness for years."

That's three contradictions across three documents created by three different people on three different dates. A well-tuned AI system surfaces that conflict in under 30 seconds and pins it to the exact page citations. A human associate finds it about 40% of the time, in my experience, and usually only when they're specifically looking. The other 60% of the time, opposing counsel finds it first.

This is the actual value proposition. Not "AI summarizes your records." Summarization is a parlor trick. The work is mapping every factual assertion against every other factual assertion and flagging the inconsistencies, the gaps, and the things the witness has never been asked about but should be.

Question Generation Is Where Most Tools Fail

I want to be blunt: most of the deposition prep AI tools on the market right now generate questions that range from generic to embarrassing. "Can you describe the accident?" is not insight. It's a Westlaw template with a wrapper.

A useful AI system for deposition prep does three things that the off-the-shelf products generally don't:

First, it generates questions grounded in specific document citations. Not "ask about prior injuries" but "ask about the November 12, 2019 visit to Dr. Patel where you reported lumbar pain rated 6/10, given that you testified in your deposition exhibit 4 that you had no back pain before this accident."

Second, it sequences questions strategically. Impeachment material doesn't get raised at the top. Foundation questions establish lock-in answers before contradiction. The system should understand that you commit the witness before you confront them, which means the prep document needs to be organized for the deposition flow, not alphabetically.

Third, it predicts likely answers and prepares follow-ups. In our internal testing across roughly 200 PI depositions, attorneys using AI-prepared follow-up trees got admissions on contested facts 2.3x more often than attorneys working from traditional outlines. That isn't because the AI is smarter. It's because the AI prepared four moves ahead and the opposing witness only prepared one.

The ROI Math Is More Aggressive Than People Realize

Let's run real numbers. A mid-sized PI firm handling 800 cases a year might take 150 to 200 depositions. At 10 hours of prep each, that's 2,000 hours of attorney and paralegal time. Cut that by even 50%, which is conservative based on what I've seen in deployments, and you've recovered 1,000 hours of capacity. At a blended internal cost of $180 per hour, that's $180,000 a year. At a blended billable opportunity cost, it's substantially more.

But the bigger number is on the settlement side. Firms that consistently extract better deposition testimony - cleaner admissions, sharper contradictions, more locked-in facts - settle their cases for measurably more. Industry data suggests a strong deposition transcript moves settlement value 15 to 25% on contested-liability cases. On a book of business with average case values of $75,000, that's a number worth taking seriously.

A Framework You Can Use Monday Morning

If you're evaluating whether to bring AI into your deposition workflow, run this test on your next three depositions:

1. Before prep: Have your AI tool ingest every document in the file and produce a contradiction report with page citations. Time it.
2. During prep: Have the attorney compare the AI's contradiction list against what they would have caught manually. Track the delta.
3. After deposition: Count how many admissions came from AI-surfaced contradictions versus attorney-surfaced ones.

If the AI isn't finding things your team missed, the tool is wrong, not the premise. Find a better tool. The premise is now settled.