The promise and the peril of AI legal help

Artificial intelligence is widely pitched as a way to help people who cannot afford a lawyer. The idea is intuitive: most people who face a legal problem never speak to a professional, and a tool that is available at any hour, at little or no cost, seems like an obvious way to close part of that gap. Even the Chief Justice of the United States has noted that, for those who cannot afford a lawyer, AI can help.

The appeal is strongest where the task is explanation rather than judgment. AI can take a dense notice, a court form, or a contract clause and restate it in plain language. It can summarize a long document, define unfamiliar terms, and lay out the general steps in a process so that a person at least knows what they are looking at and what usually comes next.

It can also help people organize. Sorting facts into a timeline, drafting a first version of a letter, or assembling the pieces of a file are tasks that consume time and confidence, and where a tool can genuinely lower barriers. Used this way, AI does not replace legal help so much as make the surrounding work less intimidating.

None of this is a small thing. For someone who would otherwise face a problem with no support at all, plain-language explanation and basic organization can be the difference between acting and giving up. The promise is real, which is exactly why the limits below deserve equal attention.

The evidence urges caution. A Stanford study found that general-purpose AI tools hallucinate on legal questions between roughly 58 and 88 percent of the time. In other words, asked a legal question, these tools produced confident answers that were wrong, or that cited things that do not exist, far more often than not.

The picture improved but did not become safe when researchers looked at specialized legal-research tools, which were found to hallucinate between about 17 and 33 percent of the time. That is better than a general chatbot, yet it still means that a meaningful share of answers cannot be trusted at face value. In one test, about three-quarters of an AI tool's answers about a specific court ruling were incorrect.

What makes this especially worrying for access to justice is where the errors cluster. Mistakes were most pronounced at the lower-court level, which is exactly where self-represented people are most likely to be operating, and most likely to rely on a tool because professional help is out of reach.

The practical lesson is not that AI is useless, but that its output is a draft to be checked, never a final answer to be acted on. A confident, well-written response can be completely wrong, and the person least equipped to catch the error is often the one depending on the tool the most.

Beyond accuracy there is a legal line. Applying the law to a person's specific facts and telling them what to do is legal advice, and giving legal advice is something non-lawyers generally may not do. That rule applies to tools as much as to people: a service that crosses from explaining the law in general to directing a particular individual's case is on the wrong side of it.

This is not theoretical. Some AI legal services have faced unauthorized-practice-of-law challenges. The rules exist to protect the public, because advice comes with accountability and professional duties that an automated tool cannot carry, and a wrong recommendation can do real harm.

The safe design follows from this. A responsible tool informs and organizes: it explains how a process works, surfaces deadlines, and helps a person assemble their materials. It does not tell the individual what they should do in their specific dispute, and when the stakes or the risk are high, it routes the person to qualified help rather than pretending to stand in for it.

Built this way, AI complements professional help instead of imitating it. The goal is to leave the person better informed and better organized so they can make their own decisions or seek advice with confidence, never to substitute a machine's guess for a lawyer's judgment.

This is research and educational material on the use of artificial intelligence for legal help. It is general legal information, not legal advice, and it does not describe any individual's matter.