For people who cannot afford a lawyer, an AI chatbot may offer a starting point that the legal system often does not: a clear explanation of what to do next. But its usefulness comes with a serious caveat—the technology can confidently get the law wrong.

At the Everlaw Summit in San Francisco, former Legal Services Corporation president Jim Sandman argued that the relevant comparison is not always between artificial intelligence and a lawyer. For many people, he said, the alternative is no help at all.

Sandman, who teaches at the University of South Carolina’s Joseph F. Rice School of Law, described an exercise in which students submitted ordinary people’s legal questions to two general-purpose AI models and two legal research tools. One prompt asked: “My apartment is freezing. What can I do?”

ChatGPT and Gemini responded with numbered steps, including contacting the landlord, sending a demand letter and seeking help from housing authorities. The legal research tools got the law right, Sandman said, but suggested going to court first—a potentially impractical response for a tenant seeking immediate heat.

The gap in access to legal help is substantial. According to figures cited at the panel, landlords have lawyers in 84% of eviction cases, while tenants have representation in about 4%. Only 1% of the nation’s roughly 1.3 million lawyers are paid legal aid attorneys. Vanderbilt Law’s Cat Moon said Tennessee has fewer than two legal aid attorneys statewide for every 2,000 eligible people.

Sandman cautioned that AI cannot fix an inaccessible system by itself. “If you automate a bad process, you end up with an automated bad process,” he said. Verna Williams of Equal Justice Works called the access-to-justice crisis an “all-hands moment.”

Legal technology company Clio is developing tools aimed at another part of the problem: courts facing heavy workloads and growing numbers of self-represented litigants. The company recently hired Casetext cofounder Pablo Arredondo and former Michigan Supreme Court Chief Justice Bridget McCormack for its judiciary business and acquired Learned Hand, an AI startup used by Los Angeles County Superior Court.

Arredondo described his approach as “ambient error detection”—assistance that brings relevant legal research to judges as they work, rather than replacing judicial judgment.

A separate tool used by a Massachusetts court illustrates the potential. Former Appeals Court Chief Justice Mark Green said it reduced review of emergency eviction-stay materials from four and a half hours to 45 minutes, with citations allowing judges to check the underlying record.

Sandman said the ideal system would combine accessible guidance with legal-grade accuracy, though “we’re not there yet.” His concern is that insisting AI match a lawyer before it can be useful overlooks people who otherwise receive no assistance.