From a Student POV: Essential and Forbidden: What Law Students Are Actually Doing With AI
Law students are getting two messages at once. One says AI fluency is essential to their careers. The other says using it on graded work could land them in front of an honor board.
That tension is now documented. Suffolk University Law School Dean Andrew Perlman recently published the AI in Legal Education: Law School Policy Archive, cataloging the AI policies of 128 U.S. law schools. Dozens of schools now require AI training, yet most schools with graded-work policies ban AI by default unless a professor opts in. Students are being told the same tool is both indispensable and off-limits, sometimes in the same semester.
Policies tell you what schools think students should do. We wanted to know what students are actually doing. So we asked our CEB student ambassadors, law students at top California law schools, three questions: how are you really using AI when nobody’s grading it, have you seen someone get burned, and where do schools go wrong? Their answers, in their own words, were more sophisticated than the policy debate gives them credit for.
How students actually use it: a starting point, never the finish line
Across every response, the same pattern emerged. Students use AI to get oriented, and then they leave it behind.
“I use AI to help give me a good starting point for my research, or an overview of a topic I’m not familiar with. I’ll look at how an issue has been ruled on recently, relevant statutes, jury instructions, secondary sources.”
“I mostly use AI as a starting point for research rather than as the research itself. I’ll use it to help me understand an unfamiliar area of law, brainstorm search terms, identify potential issues, or point me toward cases that might be relevant. But I always go back and pull the actual cases before relying on anything.”
That last habit isn’t paranoia. It’s experience.
The hallucination problem is real, and students have seen it firsthand
“I’ve definitely had AI give me a case that sounds completely legitimate but either doesn’t exist or doesn’t actually stand for the proposition it says it does.”
“I have had AI give me a fake citation from a real case, which is why I always make sure to verify any quotes or pincites generated by AI.”
Notice the second failure mode, because it’s the dangerous one. A fully invented case is easy to catch: you search for it and it isn’t there. A real case with a fabricated quote, a wrong pincite, or a holding it doesn’t actually support will survive a lazy check. It looks verified. It isn’t.
Our ambassadors haven’t watched a classmate get formally burned yet, but they’re watching the profession get burned in real time. Every one of them mentioned the sanctioned-lawyer headlines. One described the mess that unfolds even short of sanctions: work that was ‘clearly from AI’ triggering professor complaints and administrative investigations. The lesson has landed: the failure isn’t using AI, it’s submitting what AI gave you without checking it.
Where students say schools go wrong
“Teaching students to never use it is not entirely helpful. We were never taught how to use it for research purposes. Regardless of how bad or poor quality practitioners think it is, students are using it.”
“Schools sometimes go wrong by focusing so much on whether students should or shouldn’t use AI instead of teaching us how to use it responsibly. Learning how to verify citations, check the underlying authority, recognize hallucinations, and know when AI is inappropriate seems much more useful than just banning it.”
One ambassador is taking an applied AI course at their law school and described it as exactly the training the rest of the curriculum is missing. The students aren’t asking for permission to cut corners. They’re asking to be taught the skill the profession will demand of them: verification.
The skill nobody is grading: how to verify
So here it is, the verification workflow our ambassadors converged on, formalized:
- Treat every AI output as a lead, not a source. AI can orient you, suggest search terms, and surface issues. It cannot be cited, and it cannot be trusted about itself.
- Pull every case yourself. From an authoritative source, in full. If you can’t retrieve it, it doesn’t exist, no matter how plausible the citation looks.
- Read for the proposition, not the existence. The case being real is not the test. The test is whether it actually holds what AI says it holds. Check the quotes. Check the pincites. Check what happened to the case afterward.
- Confirm it’s still good law. Run every authority you rely on through a citator. In California, CEB’s TrueCite shows you at a glance whether a case has been questioned, criticized, or overruled.
- Start from authority when the stakes are real. For anything graded or filed, reverse the order: start with a current, expert-written secondary source and use AI, where permitted, only to help you digest it. CEB’s practice guides and OnLAW Pro exist precisely so that ‘getting oriented in an unfamiliar area of law’ doesn’t require gambling on a chatbot.
The students living inside the contradiction have already worked out the answer their schools are still debating: use AI for the map, never for the territory. The territory is the law itself, and it lives in verified, citable authority.
CEB student and recent grad resources, including free access through your law school, are available at https://ceb.com/resources-for-law-students-and-recent-graduates/
