Will AI make lawyers richer or put them out of work?

The potential for generative artificial intelligence to transform the legal profession was obvious even before OpenAI's ChatGPT appeared on the scene in 2022. What large-language models do - plow through huge quantities of text and data and come up with convincing-sounding things to say about them - is similar to what many lawyers do. But numerous attorneys and judges have learned to their great embarrassment that ChatGPT and other LLMs do not confine themselves to existing case law when making legal arguments, often inventing references and quotes instead.

"You've got this really clear use case in legal, except that these models, these foundational models, they're built around efficiency, not accuracy," Sean Fitzpatrick, chief executive officer of LexisNexis North America, UK, and Ireland, a large provider of information and services to lawyers, told me a few weeks ago. "As a matter of fact, they don't really deal in the truth. They deal in probabilities, and then they make plausible arguments."

LexisNexis, a division of London-based Relx Plc (the former Reed Elsevier), is hoping that a corralled, somewhat tamed version of generative AI, built atop models from OpenAI, Anthropic and others, can deliver big productivity gains for its customers without the so-called hallucinations. So is rival Thomson Reuters Legal Services, a division of Toronto-based Thomson Reuters Corp.

These legal AI services aren't flawless. Recently published research by a team of Stanford University scholars found that while the two firms' products were markedly better at avoiding hallucinations than ChatGPT, they still gave a lot of incorrect or incomplete answers, with LexisNexis' Lexis+ AI making fewer errors than Thomson Reuters' Westlaw AI-Assisted Research and Ask Practical Law AI. They do seem to be having success in the market: LexisNexis has gone from the slowest-growing part of Relx over the past five years to the fastest one, with what the company calls underlying revenue (excluding acquisitions, disposals and timing effects) up 9% in the first half of this year compared with the first half of 2024. Thomas Reuters Legal reported what the company calls organic revenue growth of 8% over the same period.

Law firms in general have been AI laggards, with a recent Thomson Reuters survey showing them well behind the corporate sector in organizational investment in AI. It is thus tempting for those in the legal profession to envision a future in which lawyers adopt this new technology deliberately but incrementally to become more productive but make no fundamental changes in how they do business. That would be good news for lawyers, Relx and Thomson Reuters - and good news for existing clients, who will presumably get better value for their money.

This sounds a lot like the first of two possible futures described a decade ago by father-and-son duo Richard Susskind and Daniel Susskind in their much-cited The Future of the Professions: How Technology Will Transform the Work of Human Experts. In this "reassuringly familiar" scenario, they wrote, "professionals continue working much as they have done since the middle of the nineteenth century, but they heavily standardize and systematize their routine activities." The second possible future - which the Susskinds saw as much more probable; Daniel has since written an acclaimed book called World Without Work - was one in which "the introduction of a wide range of increasingly capable systems will, in various ways, displace much of the work of traditional professionals" and "our professions will be steadily dismantled."

Which is it for the lawyers? This will hinge in part on whether LLMs can ever become reliable enough to replace them entirely, an outcome that remains very much in question. But even just the automation of rote tasks has the potential to upend the peculiar economics of the legal profession.

Those peculiar economics revolve around the practice of billing by the hour. At law firms, "productivity" means how many hours their attorneys bill. Everywhere else, it means how much workers get done in a given amount of time. Lawyers stand out, even among fellow professional service providers such as accountants, management consultants and tech consultants, for how little their output per hour has grown through the decades.

Adjust for capital spending, which for law firms is mostly on information technology, and what's called total factor productivity in legal services is, according to the US Bureau of Labor Statistics, substantially lower than it was in the 1980s.

This remarkable result is partly an artifact of how technological improvements are reflected in the BLS statistics - as massive deflation. A dollar spent on information technology, hardware and services in 2025 is deemed by the consumer price index to be the equivalent of nearly $15 in 1988, meaning that keeping current-dollar spending flat over that period counts as a gigantic increase in capital spending. But the fact that other high-end service providers have figured out how to increase total factor productivity despite this is an indication of how uniquely backward the law firm model is.

It may stay unique for a while yet. "I've been hearing about the death of the billable hour for a long, long time, and it hasn't died," Fitzpatrick said. By allowing lawyers to get more done in an hour, he added, the immediate effect of AI seems to have been an increase in hourly rates. "We're seeing high single-digit increases in rates per hour. I think part of that is attorneys are now delivering more value to their clients and they're doing it in less time." In a recent survey of big law firms, Robert J. Couture of the Harvard Law School's Center on the Legal Profession found a similar consensus that increased use of AI is bringing quality improvements for which clients are happy to pay a premium.

Maybe not forever, though. My friends Bruce MacEwen and Janet Stanton, who together constitute the legal management consultancy Adam Smith, Esq., think the decline in hours billed will eventually prove too steep for most law firms to bear. "This is going to kneecap the traditional hourly billing model of law firms," MacEwen said. "I think fixed-fee pricing is going to be the wave of the future." The problem for law firms, Stanton said, "is that they have really poor financial infrastructure so their ability to price things is not great." The likely result, MacEwen concluded, is "a vast winnowing of traditional law firms."

Will we get rid of the lawyers altogether? A major theme of the Susskinds' book is that the professional-services status quo isn't acceptable, in large part because "most people and organizations cannot afford the services of first-rate professionals." If normal people could just plug their legal queries into ChatGPT and get reliable answers, wouldn't society be better off?(1)

Maybe, but it's an awfully big if. In a series of papers with a variety of co-authors, Jonathan H. Choi of the Washington University School of Law has been testing the performance of LLMs on different legal tasks. They're great at simple ones with straightforward answers but struggle with complexity and ambiguity. In his most recent working paper, "Large Language Models Are Unreliable Judges," Choi describes how minor variations in the wording of prompts deliver enormous variations in answers to legal questions and how different models give much different answers, too.

Choi also suggests that these problems may be inherent to LLM technology and thus not something that will be fixed soon. The lawyers could still be with us for a while, then. It's just not clear how they're going to be paid.

---

(1) I realize one could say the same thing about journalists, but we get paid a lot less than lawyers and have already been through a pretty vast winnowing over the past two decades.

Justin Fox is a columnist writing about business. Prior to joining Bloomberg View, he was the editorial director of the Harvard Business Review. He is the author of "The Myth of the Rational Market."

(COMMENT, BELOW)

Previously:
• 04/09/25: If cameras don't slow serial speeders, 'limiters' can
• 03/16/22: If cameras don't slow serial speeders, 'limiters' can
• 11/22/21: Grown kids still stuck at home? Change is on the horizon
• 10/28/21: Fewer people going to college is good news
• 07/06/20: GET READY: This new coronavirus wave isn't like the old wave
• 01/16/20: Understanding the 'war on men' in the workplace
• 01/09/18: Why some cities get all the good jobs
• 01/04/18: If you want to 'change the world,' keep it to yourself
• 01/04/18: If you want to know the future, ask the humorists
• 12/04/15: What good retirement plans everywhere have in common
• 11/27/15: Sorry, you lost the right to have your day in court
• 07/01/15: Uber Is Lobbying for All of Us