Two New York lawyers filed a brief in 2023 citing six court decisions that supported their case perfectly. The decisions did not exist. ChatGPT had invented every one, complete with case names, judges, and quotations. Judge Kevin Castel fined the attorneys $5,000 and made them write to every real judge whose name got stapled to a fake opinion. That case, Mata v. Avianca, is the story most personal injury lawyers know about AI. It is also the wrong lesson to take from it.
You are a personal injury firm owner deciding how much of the AI noise to believe. Vendors promise to run your whole practice. Skeptics point at the sanctions and say stay away. Both are selling you something. Here is what AI does for a PI firm right now, what it cannot touch, and the one habit that turns a useful tool into a bar complaint.
PI firms adopt AI faster than the rest of the bar
Personal injury lawyers are out ahead. In 2025, 37% said they use generative AI in their work, against 31% of lawyers overall. The reason is structural to your business: you run on volume, thin margins on early-stage files, and a paralegal team drowning in medical records. AI touches all three. The catch is that only 19% of PI firms have rolled it out firm-wide, so most of that 37% is individual lawyers experimenting on their own, without a policy, which is where the trouble starts.
Four places AI earns its keep today
Intake and speed-to-lead. This is the clearest win, because the problem is measured and brutal. A prospect contacted within one minute converts at 391% higher rates than one reached a minute later. Wait five minutes and your odds drop 8x. After an hour, the lead is cold and your recovery rate falls under 5%. Yet a Hennessey Digital study of nearly 1,400 firms found 27% never responded to web leads at all, and the ones that did took a median of 13 minutes. AI chat and intake tools close that gap: they answer at 2 a.m., in Spanish, ask the qualifying questions, and hand you a warm file in the morning. Burt is one option here among several. The value is speed, not the vendor.
Medical-record review and chronologies. This is the paralegal-hour black hole, and it is where AI saves the most raw time. Tools that summarize records and build treatment timelines cut real work. One firm reported summary turnaround dropping to 3.2 days and paralegal time on summarization falling from 60% to 28% of hours after switching to AI-assisted review. You still read the summary. The machine does the sorting.
Demand letters and document drafting. Drafting correspondence is the single most common AI task among lawyers who use it, at 52%. AI turns a record set into a timeline, a plain-language client summary, and a demand-letter draft from the same source. It writes the first version. You bring the argument, the number, and the judgment about what to leave out.
Marketing and content. Blog posts, ad copy, review responses, first drafts of the FAQ page nobody wants to write. Low stakes, high volume, easy to check. Start your AI experiments here, not on a court filing.
Where AI gets oversold
Case valuation is the one to watch. Vendors sell "predictive case value" tools that score a file and forecast a settlement range. Insurers already run the same play, with Colossus, ClaimIQ, and Guidewire spitting out numbers that adjusters treat as gospel. The models miss what actually moves a settlement: the client's credibility, how a jury reads their pain, the intangible damages no algorithm quantifies. Use the number as a sanity check. Never let it set your demand, and never let the adjuster's number set it either.
Legal research is the other overhyped lane, and it is the one that ends careers. The pitch is that AI reads the case law for you. The evidence says otherwise. A Stanford study tested the paid, lawyer-grade tools built for exactly this, and found hallucination rates of 17% for Lexis+ AI and 33% for Westlaw's AI research product. These are the good tools, marketed as hallucination-free. One in six answers was still wrong, either a fabricated case or a real case twisted to say something it never said.
The mistake that gets lawyers sanctioned
Mata v. Avianca was not a freak accident. It is a pattern, and it accelerated. A data analyst tracking the problem logged 120 incidents; researchers counted roughly 712 court decisions worldwide addressing hallucinated content, with about 90% of them written in 2025. The fines keep landing. A federal judge hit one attorney with a $6,000 sanction for citing nonexistent cases. Two lawyers for MyPillow's CEO filed a brief with nearly 30 defective citations and drew $3,000 each in Colorado. A Denver attorney who denied using AI took a 90-day suspension.
The common thread in every one of these is not AI. It is a lawyer who did not check the citation before signing the filing. The tool wrote something plausible and false, and a human put their bar number under it without opening the case. Rule 11 does not care that a machine drafted it. Your name on the signature line means you verified it.
Run AI on the work where a wrong answer costs you an hour, not your license. Draft with it, summarize with it, answer leads with it. Then read everything it produces about the law as if a first-year associate with a talent for confident nonsense wrote it, because that is what you have.
Frequently asked questions
What can AI actually do for a personal injury law firm today? Four things reliably: answer and qualify leads 24/7 so you stop losing the 391% conversion edge that comes with a sub-one-minute response, summarize medical records and build treatment chronologies, draft demand letters and correspondence, and produce marketing content. It handles volume and first drafts. It does not replace your judgment.
Is it safe for lawyers to use AI for legal research? Only with verification. A 2025 Stanford study found the leading paid legal-research tools hallucinated on 17% to 33% of queries, including fabricated cases. Every citation an AI produces must be pulled and read before it goes in a filing. Lawyers who skipped that step have been fined and suspended.
Why do lawyers keep getting sanctioned for AI citations? Because they sign filings without checking the cited cases exist. Mata v. Avianca drew a $5,000 fine in 2023, and courts issued roughly 712 decisions on hallucinated content by 2025. The sanction is never for using AI. It is for filing fake law under your name.
We break down one AI tool, tactic, or cautionary tale for PI firms every week, the stuff that grows your caseload and the stuff that gets firms fined. Subscribe and get the next one in your inbox.