Why ChatGPT won’t replace lawyers

I keep hearing the same question in boardrooms and on law society and bar association panels: “Will ChatGPT replace lawyers?” As the CEO of a legal technology company, my answer is simple: you’re asking the wrong question.

The real disruption in law won’t come from some general AI chatbot spouting answers about everything and nothing. It comes from legal LLMs, which are large language models trained exclusively on legal data. The standard narrative that a one-size-fits-all AI will supplant lawyers is not just overhyped; it’s dead wrong. In reality, specialized legal artificial intelligence is the true game-changer, and it’s already pounding on the courthouse doors.

LLMs using legal databases

Consider how legal AI software is being engineered. It’s not just blindly relying on a single large model to do everything. The best systems combine LLMs with actual legal databases. For example, Casetext’s artificial intelligence assistant, CoCounsel, uses GPT-4 under the hood, but routes queries through specific legal tasks, like searching a database of case law, statutes, and regulations, before spitting out an answer​.

This means the model isn’t operating in a vacuum; it’s supported by a corpus of real legal knowledge when drafting a memo or analyzing a contract. The result is a far cry from the vanilla ChatGPT experience. In effect, the legal LLM behaves less like a chatty know-it-all and more like a supercharged legal researcher with instant access to the entire law library. It’s trained and tuned to understand legal jargon.

To put it bluntly, a general LLM is a jack of all trades; a legal LLM is a master of one.

The ChatGPT hype vs. the legal reality

Let’s cut through the buzz. OpenAI’s GPT-4 (the brains behind ChatGPT) famously passed the bar exam, stunning onlookers​. But passing a test in a lab and practicing law in the real world are very different beasts. In practice, general-purpose AI often falls short when it enters the nuanced, high-stakes realm of legal work.

The reason is simple: ChatGPT was trained to be a jack-of-all-trades, not a legal expert. It will happily offer plausible-sounding but false answers because its goal is to predict likely text, not accurate law. We witnessed this play out in a cringeworthy fashion when two New York lawyers relied on ChatGPT to write a brief. The AI fabricated six court cases out of thin air​.

The result was that they were sanctioned and publicly humiliated for filing a brief riddled with “bogus” citations. One later admitted he was “shocked” that the technology could make cases up “out of whole cloth​.” The lesson here? A general AI that “hallucinates” legal authorities isn’t a legal assistant; it’s a liability.

That said, lawyers should always review their work thoroughly before filing. It doesn’t matter if the information comes from Caseway, LexisNexis, Google, Bing, or a paralegal.

Contrast this with a legal-specific LLM, which is trained not on Reddit rants and Wikipedia entries, but on court rulings, statutes, regulations, and legal briefs. It doesn’t just talk like a lawyer; it knows the law, at least far better than any generic model. While ChatGPT might know a little about many things, a true legal LLM knows a lot about one crucial thing: law. No human lawyer could read 100 million court decisions and remember them all. But a legal LLM like Caseway can.

That specialization makes a world of difference. General AI might guess what a legal doctrine means; a legal AI can quote you the case. General artificial intelligence might spew a confident answer that’s completely wrong; a legal AI can be built to cite its sources and relevant precedents, reducing the risk of wild hallucinations. In other words, the generalist parrot is being upstaged by the specialist owl.

Legal LLMs are the real revolution

The ‘AI will replace lawyers’ narrative has it backwards. It’s not general AI like ChatGPT that will replace lawyers — it’s lawyers who leverage legal AI who will replace lawyers who don’t. The true disruption in law comes from within, not from some outside Silicon Valley experiment with no understanding of precedent or procedure.

It comes from teams of lawyers, engineers, and data scientists who train models on the actual substance of the law and wrap them in software that augments, rather than supplants, legal expertise. These legal LLMs are already starting to reshape our work, everything from research and drafting to how we think about legal services at scale.

We’re at the start of a new era. I predict that every competent lawyer will have an AI copilot in the coming years — not a gimmicky ChatGPT that might tell them the sky is green, but a trusted legal AI assistant that knows the corpus of law cold. Technology in hand, lawyers will still argue in court, exercise judgment, and apply creativity and empathy.

However, a massive chunk of the laborious analytical work and writing can and will be offloaded to machines that do it faster, cheaper, and in many cases better (or at least as well as a human). We consistently observe this in pilot after pilot and study after study. LLMs can deliver on core legal tasks with high accuracy and achieve significant efficiency gains. That means more access, lower costs, and yes, a jolt to the traditional law firm model.

The lawyers who thrive will embrace this software and redefine their role. Instead of spending five hours scouring case law, you’ll spend 30 minutes prompting AI and 30 minutes vetting and polishing its output. And guess what? You will produce a work product that is as good or better than before.

That leaves you more time to strategize with your client, think big-picture, and practice law instead of doing rote labour.

Meanwhile, those who refuse to use this software will struggle to keep up and likely produce work more slowly and expensively, pricing themselves out of the market or simply delivering subpar value.

Ultimately, this is a classic Darwinian situation: adapt or perish.