Claude for Law Students: Outlines, Briefs, and Honor Codes

The Short Answer

Claude is a strong study tool for law school when you use it to test your understanding rather than replace it, and whether that use is permitted depends entirely on your school's and your professor's written policy, which you must read before you paste in a single assignment. As of September 2026, the published policies at law schools that have them range from Berkeley's near-total restriction to Columbia's learning-aid model, and each of them restricts AI in exams by default (Berkeley and Columbia bar it outright; Stanford bars using it to substantially complete one). Below: what Claude does well for a 1L through 3L, what each of those policies says, how Claude compares with ChatGPT for this specific use, what students can actually pay, and a short honest section for readers who are not students at all but self-represented litigants searching for "the best legal AI for pro se."

What Claude Is Actually Good For in Law School

Outlining, the right way. The value of an outline is in making it, not having it. Use Claude as a Socratic partner: give it your reading notes for a unit (check the policy on uploading course materials first; some schools prohibit it) and ask it to quiz you on the rule, the elements, the exceptions, and how the cases fit together. Ask it to find the gaps in your synthesis. Do not ask it to write the outline; a downloaded outline you did not build teaches you nothing, and depending on your school, submitting or sharing it may be a violation.

Case briefing as a check. Brief the case yourself first. Then paste the opinion (public court opinions are not course materials) and your brief, and ask Claude what you missed or mischaracterized: is that the holding or dicta, did you state the procedural posture correctly, what is the best counterargument the dissent makes. This is the same holding-versus-characterization check practicing lawyers use in a cite-check, described in our research and cite-check workflow, and it builds exactly the skill exams test.

Exam practice. Ask Claude to generate issue-spotter hypotheticals in the style of your course, write your answer under time pressure, then have it grade you against a rubric you supply (IRAC structure, issues spotted, rule statements, application depth). Ask it to explain what a stronger answer would have added. This is legitimate, policy-permitting, and one of the most effective uses of the tool, because the model is generating questions and feedback, not answers you submit.

Learning Mode. Anthropic's education offering includes a mode that behaves like a tutor, asking questions that lead you to the answer rather than handing it over. If your school licenses Claude for Education, use it; Anthropic also brought a Learning style to all Claude.ai users in August 2025, and you can reinforce it with a standing instruction in a Project: "Do not give me answers to legal questions; ask me questions until I get there, then confirm."

What it is bad for. Citing cases from memory (it fabricates, as every practitioner has learned), telling you whether a case is still good law (see why Claude is not a citator), and anything that goes into an exam.

The Honor Code Is the Whole Ballgame

Read your school's policy, then your professor's syllabus, then ask if anything is unclear. Three examples of published policies, chosen because they span the range:

• UC Berkeley Law (published policy): AI may not be used for any purpose in any exam situation; students may not upload course materials (assignments, readings, slides, recordings) into generative AI systems; AI may not be used to conceptualize, outline, draft, revise, translate, or edit work submitted for credit, including correcting grammar; the sole permitted default use for papers is identifying sources such as cases, statutes, and secondary materials. Instructors may deviate in writing.
• Columbia Law School (2026-2027 default policy): AI may serve as a learning aid for studying concepts, research, and improving clarity of written work, but students may not rely on it to perform the legal analysis or judgment, absent advance written permission from the instructor, AI may not be used in exams and every word of an exam answer must be the student's own, submitted work must be student-composed, and students are required to disclose any use of AI that contributes to a written submission or other work product. Faculty may adopt alternative policies in writing in their syllabi.
• Stanford (university-wide Honor Code guidance): absent instructor guidance, AI use is treated like assistance from another person; using AI to substantially complete an assignment or exam is not permitted; students should acknowledge non-incidental AI use and default to disclosing when in doubt. The Stanford Law Juelsgaard clinic goes further and requires express approval before any AI use on clinic work.

Two patterns hold across all three: exam use is restricted by default, and the professor's written rule overrides the school default. Many schools publish less than these three; where your school has no AI-specific academic policy, the operative rules are your syllabus and the general honor code's plagiarism and unauthorized-assistance provisions. (Do not confuse those with admissions rules: a 2026 survey of 89 law schools' application requirements found only 13 with a published J.D. AI rule for applicants, which governs application materials, not coursework.) The stakes are not academic. An ABA Journal report quotes Richard Trachok, chair of the Nevada Board of Bar Examiners, saying that a candidate with an honor-code violation could potentially not be allowed to take the bar exam, and Belmont law professor David L. Hudson Jr. calling an academic misconduct charge "one of the major red flags" in character and fitness review. When in doubt, ask in writing and keep the answer.

Claude vs ChatGPT for Law School

For the study uses above, both work, and the differences are modest. In our experience, Claude's default writing register is closer to legal prose, it handles long documents (a full casebook chapter, a 60-page opinion) comfortably on paid plans, and Projects let you keep one persistent workspace per course with your standing "tutor, not author" instruction. ChatGPT has a broader consumer feature set, strong voice mode for oral practice, and a larger ecosystem of study-oriented custom GPTs. Our practitioner comparison goes deeper on the writing and document differences; for a student, the honest advice is to pick the one you can get cheapest through your school and use it within policy. Whichever you choose, treat every case citation as unverified until you pull the opinion.

What Students Can Pay (October 2026)

• Claude Free is free and, per Anthropic's pricing page, includes chat, web search, memory, file creation, and connectors. Enough for briefing checks and exam hypotheticals at modest volume.
• Claude Pro is $20 per month or $17 per month billed annually ($200 up front), and adds higher usage, Claude Code, and Projects beyond the Free plan's limit of five. Anthropic does not list an individual student discount on its pricing page as of October 2026; its education product, Claude for Education, is a university-wide plan that institutions license (partner logos on the page include Northeastern, LSE, Dartmouth, UVA, and Syracuse). Ask your law library or IT office whether your school has a license before paying.
• ChatGPT is running a back-to-school offer of four free monthly billing periods of ChatGPT Plus for verified students at eligible U.S. degree-granting institutions, claimable through October 31, 2026, with enrollment verified through SheerID; after that it reverts to the standard $20 Plus price. Some universities also license ChatGPT Edu.

Two cautions on data. Consumer plans on both services may use your conversations for training unless you change a privacy setting (Anthropic's consumer privacy policy describes the opt-out and its safety-review carve-out). That matters less for a torts outline than for a clinic client file: clinic work is real client work, and it belongs on whatever plan and policy your clinic director specifies, which at many clinics is "none without approval."

A Short, Honest Section for Pro Se Readers

If you searched for the best legal AI for pro se and found this page, here is what a lawyer-facing site can tell you truthfully. Claude can help you understand a court form, explain what a legal term means, organize a timeline of events, and draft a plain-English letter. It is not a lawyer, it is not a substitute for one, and it will invent case citations that look real. Courts have been sanctioning self-represented litigants for filing AI-fabricated authorities in 2026; a May 2026 roundup lists dismissals with prejudice, stricken motions, and monetary penalties, and several of the courts said pro se status did not excuse the fabricated citations. If you use AI to help write a filing, verify every citation in a free source such as CourtListener or your state's court website, and look for your court's self-help center or a legal aid organization, which can tell you things a chatbot cannot. Anthropic's Claude for Legal toolkit includes a Legal Clinic plugin and, per Anthropic's announcement, connectors to free and low-cost public-interest tools such as Courtroom5 and CourtListener, which we cover in MCP connectors for legal; those support clinics and self-represented litigants, and they do not replace counsel.