Which AI tools are actually worth your time?
Sort tools by whether they are grounded in a real document set or generating from open memory. Grounded tools retrieve from a database and cite what they retrieved; they still misdescribe holdings, but they rarely invent case names. Ungrounded tools produce plausible text and are far more likely to fabricate authority. Use grounded tools for anything that will end up in a brief, memo, or exam answer.
Your tuition already covers the expensive ones. Both major research vendors give law students full academic access through school credentials, and both now ship generative features at no extra cost to you.
- Legal research: Lexis+ AI and Westlaw Precision with CoCounsel. Free with your school ID. Use them for issue framing, finding a starting case, and summarizing long opinions — then read the opinion.
- Your own materials: NotebookLM. Upload your class notes, your outline, and case briefs; it answers only from what you uploaded and links back to the source passage. Best tool in the category for Q&A over your own work.
- Explanation and hypo drilling: ChatGPT, Claude, or Gemini. Excellent at restating a doctrine three different ways and at generating unlimited practice fact patterns. Unreliable on jurisdiction-specific rules and on anything decided recently.
- Spaced repetition: Anki, with AI used to draft cards from your own outline. Review every card before you add it; a wrong card drilled 40 times is worse than no card.
- Transcription: Otter or built-in phone transcription, but only if your professor allows recording. Many do not, and recording without permission is itself an honor code problem at some schools.
- Commercial supplements with AI layers (Quimbee, bar-prep platforms): fine as supplements, useless as substitutes for reading the assigned cases.
What does AI actually do well in law school?
It is very good at compression and at generating volume. Feeding it a 60-page opinion and asking for the procedural posture, the holding, and the dissent's core objection gives you a reliable scaffold for your own reading — not a replacement for it. It is equally good at producing twenty issue-spotter hypos on personal jurisdiction so you can practice recognition until it becomes automatic.
It is also a patient Socratic partner. Ask it to quiz you on the elements of adverse possession and to refuse to give you the answer until you have tried. Ask it to argue the opposite side of a case you think is obviously right. That kind of adversarial practice is hard to get from a study group at 11 p.m.
It is worst at exactly what law school grades: applying a rule to messy facts with judgment about which facts matter. Generated exam answers tend to be rule-dumps with thin application — the structure that earns a B-minus. If you outsource the analysis step, you never build the skill the exam tests.
How do you stay on the right side of your school's rules?
This genuinely varies by school and by professor, and the variation is not cosmetic. Some schools adopted a default prohibition — generative AI is barred on graded work unless a professor affirmatively permits it. Others permit it with a disclosure statement. Many leave it entirely to individual syllabi, which means you may be allowed to use AI in Contracts and forbidden in Legal Research and Writing during the same week. Legal writing faculty are the most likely to ban it outright, because the course exists to build the drafting skill AI would perform for you.
Three rules will keep you safe. First, read the honor code and every syllabus, and when a syllabus is silent, email the professor and keep the reply. Second, treat exams as a hard line: exam software like Examplify locks down the machine, and using a second device during a closed exam is straightforward cheating regardless of the AI question. Third, never paste confidential material — a clinic client's facts, a summer employer's documents — into a consumer chatbot.
The stakes are higher than a bad grade. An academic misconduct finding is reportable on every bar application's character and fitness section, and you will be explaining it in writing for the rest of your career.
How do you verify output so you don't end up sanctioned?
The cautionary cases are real and they keep multiplying. In Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), lawyers filed a brief containing citations ChatGPT had invented and were sanctioned under Rule 11. The Second Circuit referred an attorney for disciplinary consideration over a fabricated citation in Park v. Kim, 91 F.4th 610 (2d Cir. 2024). The Missouri Court of Appeals has dismissed an appeal and awarded damages where the brief cited nonexistent authority. Courts have not been sympathetic to the explanation that the software made it up.
Build a mechanical verification habit now, before you have a supervising attorney. For every citation an AI gives you: pull it in Westlaw or Lexis by citation, confirm the case exists with that name and reporter cite, open the opinion and find the proposition on the actual page, and run KeyCite or Shepard's for negative treatment. If any step fails, the cite does not go in your document.
Verify holdings, not just existence. The most dangerous failure mode with grounded research tools is a real case, correctly cited, described as standing for something it does not hold — often a proposition from the losing party's argument or from dicta.
How do you cite AI in academic writing?
The short answer is that you usually should not need to. AI output is not authority. If a model points you to a statute or a case, you read that source and cite the source, in ordinary Bluebook form. Citing the chatbot signals that you did not verify.
If a professor or journal requires disclosure — for example, on a seminar paper where you used AI to generate an initial outline — follow the format they specify. Absent a specified format, a footnote or an appended note stating which tool you used, the date, and what you used it for is the safe approach. The Bluebook's internet-source rule (Rule 18) is the closest analogue for a non-persistent online source, but editions differ on how they treat AI output, so confirm which edition and which house style your journal follows before you improvise.
One practical caution about the research platforms: your academic Lexis and Westlaw credentials come with use restrictions. They are for coursework and law-school-related activity, not for paid work at a summer job, externship, or clerkship where the employer has its own subscription. Violating that is a contract problem, not just an etiquette one.
What does a sensible weekly workflow look like?
Read the case yourself and brief it yourself. Then use AI as the second pass: ask it to explain the part you did not follow, or to summarize the opinion so you can check your brief against it. The order matters. Reversing it means you read the summary and never form your own read of the facts.
For outlining, do the synthesis by hand and use AI to stress-test it. Paste your rule statement for promissory estoppel and ask what a hostile grader would say is missing. Ask it to generate a hypo where your rule statement produces the wrong answer. The outline's value is in the making, so never adopt a generated outline as your primary document.
For exam prep, use it as a feedback loop. Write a timed practice answer with no assistance, then give the model the fact pattern, the professor's model answer or rubric if you have one, and your response, and ask for a structural critique: did you spot the issues, did you apply facts or just recite rules, did you address counterarguments. Trust its structural feedback more than its substantive corrections, which may not match your professor's jurisdiction or emphasis.