Penn Carey Law You can build strong legal foundations in your 1L year without using AI as a study partner. For some students, waiting is the right call, and nobody will think less of you for it.
Fluency with these tools will matter by the time you graduate. The 1L question is when and how to start, not whether. This guide is for students who want to start now.
Companion to the 1L talk. The slides cover the essentials in fifteen minutes. This page keeps the same traffic light and study routines, adds more prompts, explains how to extend Claude, and, for those who want it, lays out the learning science behind the advice. It will keep growing; check back.
Course rules come first. Nothing here overrides a syllabus or an assignment instruction; see section II.
IYour AI Tool Access
Every 1L has a Penn Claude seat. Penn also runs its own portal, PennChat, which is useful as a second opinion.
The seat includes the Claude chatbot, Claude Code, and API credits. Sign in at claude.ai with your PennKey@upenn.edu address, not your law school address, and complete the short required training before first use. Under Penn’s enterprise agreement, Anthropic does not train its models on Penn data. Setup details are on the Penn Law ITS Penn Claude page.
Go to pennchat.upenn.edu on PennNet, or on the GlobalProtect VPN off campus, and sign in with your PennKey. It offers both OpenAI and Anthropic models inside Penn’s own environment. Its main use for studying is a second read: ask a different model the same question and compare.
Penn Claude is cleared for low- and moderate-risk data; PennChat for low, moderate, and most high-risk data. For studying, the rule is simple: never put Social Security numbers, credit-card numbers, passwords, health information, or another person’s private information into any AI tool. Your own notes and outlines are yours to use. Whether you may upload a professor’s materials depends on that professor’s rules. Fuller guidance: Penn Law ITS AI Guidance.
IICourse Rules Come First
- Rules are set course-by-course, and they can differ by assignment within a course.
- Read the syllabus and the assignment instructions.
- Permission in one course says nothing about another.
- Permission to use AI as a study partner never implies permission to use it on graded work, writing projects, or exams.
- If anything is unclear, ask before you use the tool.
IIIThe Key to Learning with AI: Who Does the Work?
Think of the gym. If someone lifts the barbell for you, the barbell moves and nothing happens to you. A trainer does the opposite: hands you the weight, watches your form, and adds resistance when it gets easy.
Claude can play either role. Ask it to summarize a case before you read it and it will, and you will have learned nothing. Ask it to quiz you after you read the case and it becomes a patient study partner that is available at 2 a.m. and willing to push back. Same tool. The only question that matters is who is doing the work.
The question bites harder in law school, because the documents you produce have no value of their own. Nobody will ever pay you for a 1L Contracts outline or a brief of Pennoyer v. Neff. Those documents matter only because of what producing them does to you. If the tool produces them, the point is gone.
- Summarizes the case before you read it.
- Writes your outline.
- Drafts your practice answer.
- Hands you the “right answer” to a doctrinal ambiguity.
- The barbell moves. Your muscles don’t.
- Quizzes you after you read.
- Pokes holes in your rule statement.
- Grades the answer you wrote.
- Argues the other side against you.
- You do the reps. The resistance builds muscle.
IVThe Traffic Light for 1L Study
The test is simple: did you do the thinking first?
- Quiz me on what I read. After you have read and briefed the case, have Claude question you on the holding and how the facts drove it.
- Poke holes in my analysis. Write your own rule statement or argument, then ask where it is imprecise, overbroad, or open to attack.
- Give me a new hypo to work out. A fresh fact pattern on a doctrine you have already studied, with no answer until you commit to one.
- Argue the other side against me. Take a position, then have Claude play opposing counsel.
- Critique the answer I already wrote. Write a practice answer under timed conditions, then ask for feedback. If your professor has released a model answer and allows you to upload it, ask Claude to grade against that.
- Audit my brief, after I read the case. Compare Claude’s read against your brief only once you have read the opinion and written your own.
- Explain it, after my own try. Asking for an explanation is fine once you have tried to explain the doctrine yourself and got stuck. Treat the answer as a hint to check against the casebook, not as the rule.
- Compare my outline to the syllabus. Look for omissions and inconsistencies in a structure you built yourself.
- Untangle the procedural posture. Multi-party appeals and jurisdictional maneuvers, after you have made your own attempt.
- Any case, quote, or citation it names: verify it before it goes anywhere.
- Summarize the case before I read it. This removes the struggle that teaches you to read a case.
- Build my outline for me. Synthesizing forty cases into one framework is the hardest and most valuable work of 1L. Outsource it and you arrive at the exam with someone else’s map.
- Draft my practice answer. Your first attempt at any practice problem is your only honest measure of what you know.
- Cite anything it gives me without checking. Cases, statutes, Restatement sections, quotations.
- Anything the course rules bar. And anything that puts sensitive data into the tool.
VStudy Routines to Start With
Copy these into Penn Claude and adapt the bracketed parts. Every one starts with your own work on the page, and every one carries a boundary line: “Do not summarize it,” “Do not rewrite it,” “Use only those cases.” That last one matters. Without it, Claude will import doctrine from cases your professor never assigned, and you will not know the difference yet. Three routines you actually use beat thirty you don’t.
I just read [case]. Do not summarize it. Ask me one question at a time about the holding and how the facts drove it. Push back when my answer is thin.
Here is my three-sentence statement of [doctrine], based on [cases]: [paste]. Where is it imprecise, overbroad, or open to counterargument? Use only those cases.
Here is my section on [topic]: [paste]. Do not rewrite it. Point out gaps, contradictions, and missed exceptions, using only the assigned cases.
More routines
Once the three above are habits, these extend them.
I am a 1L studying [course]. We are covering [doctrine]. Do not explain the doctrine or summarize it. Act as a demanding law professor: give me a two-paragraph fact pattern that tests the edge of this doctrine and ask me to take a position. Wait for my answer. Then critique it, name any element I missed, and add a factual twist that makes it harder.
I wrote this practice answer for [course] under timed conditions, without help. The question: [paste] My answer: [paste] [Optional: the professor's model answer or rubric: [paste]] Grade it the way a law school exam grader would. Did I spot the major issues and the secondary ones? Did I organize around clear rule statements and counterarguments, or jump to conclusions? Where did I fail to use specific facts from the question?
I am preparing an argument on [topic] for [course]. My position and supporting cases: [paste]. Act as opposing counsel. What is the most vulnerable premise in my theory? Which of the assigned cases cut against me, and how would you distinguish mine? Give me the strongest policy argument against the rule I am advocating. Use only the assigned materials.
VIFour Habits That Keep You in the Green
Never start from a blank screen. If you have not put your own analysis in words, you have nothing for the tool to work on. Write the holding, the rule statement, or the exam paragraph first, then ask Claude to test it.
Make Claude the demanding professor. Questions, hypotheticals, and critique, not summaries. If you catch yourself reading an explanation and nodding, stop and try to reproduce it without looking.
Anything it names, you confirm in the casebook, on Westlaw, or on Lexis before relying on it. The Legal Practice Skills program will teach you the research tools built for this later in the term.
Some things stay fully human: your first read of every case, class itself, and your practice exams. Those are where you find out what you actually know.
VIIYour Toolkit: Apps, Projects, Skills, and More
The chat window at claude.ai is the starting point, not the limit. Everything below is optional, and everything below still follows the traffic light: you do the work first, then the tool tests it.
1. The apps
Your Penn Claude seat works everywhere Claude does: the web at claude.ai, desktop apps for Mac and Windows, and mobile apps for iPhone and Android, all from claude.com/download. Sign in with the same PennKey@upenn.edu address and your chats and projects follow you across devices. The desktop app is the one to use for long study sessions; the phone app is fine for a quick quiz on the train.
2. Models
Penn Claude offers more than one model. Sonnet 5.5 is fast and the everyday choice; Opus 5.5 is slower and stronger on hard reasoning. For quizzing and rule-checking, Sonnet is plenty. Switch to Opus when you want a serious critique of an argument or a practice answer. The model picker is at the top of every chat.
3. A Project per course
Projects group chats together with a standing instruction and a shared set of files, so you stop re-explaining yourself every session. Set up one per course. Give it a short tutor instruction, something like: “You are my study partner for Civil Procedure. Ask questions rather than give answers. Require my attempt before you respond to any substantive question. Flag uncertainty. Rely only on the materials in this project and on the cases I name.” Then add your own notes, briefs, and outline as project knowledge as the semester goes. Your own notes and outlines are yours to upload. Whether you may upload a professor’s materials depends on that professor’s rules.
4. Skills
A skill is a packaged set of instructions Claude loads when a task calls for it, so a routine you use often becomes a single command instead of a pasted prompt. Skills live under Customize → Skills in settings, but Penn has not yet turned them on for the Penn Claude organization, so you will not see that menu for now. When it appears, you will be able to package your own: the three study routines above, written up as a skill, become “quiz me on Palsgraf” with no setup each time. The faculty portal’s skills catalog is written for faculty, but it shows what a finished skill looks like. I’m planning a student study-skill set for this guide.
5. Claude Code
Your seat includes Claude Code, a command-line version of Claude that works directly with files on your computer. For students comfortable with a terminal, it is a good way to search a semester of notes, check cross-references, compare two versions of an outline, or build a local study workspace. It is not required, and this is not a coding tutorial. If the chat window does what you need, stay there.
6. PennChat, for a second read
When Claude critiques your rule statement, ask a different model in PennChat the same question and compare. Where they disagree is usually where you should look harder. Access details are in section I.
7. Connectors, and what to expect
Connectors let Claude read from an outside source rather than from memory. Penn Claude does not currently allow you to add connectors, so none of this is available to you yet. A legal-source connector (CourtListener, which holds full-text opinions) would be the natural first one; Westlaw and Lexis connections may follow, with no date. When one arrives it will change what the tool can look up, not what you have to verify. Case name, citation, court, date, and whether the opinion is still good law remain your job.
8. More guides and training
Where to go next, grouped by who wrote it.
- Anthropic. Claude Academy, free courses; start with “AI Fluency: Framework & Foundations,” about four hours. Help-center articles on Projects and Skills, and the Claude Code quickstart.
- Penn Carey Law ITS. The Penn Claude setup guide and the school’s AI guidance index, including data rules.
- Penn ISC. The Anthropic Enterprise FAQ, the PennChat FAQ, Penn Generative AI Tools & Resources, and the data risk classification the tool clearances refer to.
- Penn Libraries. AI Tools and Best Practices, a research guide with a good section on evaluating AI output.
- Penn Carey Law AI Project. The Prompt Guide & Scenarios, with worked examples for class prep and practice questions; the earlier 1L Guidance on AI Use, which covers how the models work and the human-AI-human workflow (you draft, the tool responds, you check); and the Legal AI Tool Guide for when you meet Westlaw’s and Lexis’s AI features in Legal Practice Skills.
- The faculty portal. AI Resources is written for faculty, but its tool catalog, model notes, and policy pages apply to you too.
VIIIThe Science
You do not need this section to use the traffic light. It is here for students who want to know why the advice runs the way it does.
The illusion of competence
When Claude explains personal jurisdiction or promissory estoppel, the explanation is clear and easy to follow. You nod along and feel that you understand. Cognitive psychologists call this the illusion of competence: recognizing a clear explanation is not the same as being able to generate, apply, and defend the analysis under pressure. On a cold call or a timed exam the tool is gone, and if you never struggled to build the understanding yourself, the understanding is not there either.
Desirable difficulties
Research on learning, including Robert and Elizabeth Bjork’s work on “desirable difficulties,” indicates that durable retention and flexible understanding depend on effortful work: retrieving from memory before checking, and resolving contradictions before seeing the answer. When you read a nineteenth-century opinion, the confusion is where the learning happens. Separating holding from dicta and reconstructing the procedural posture is what builds the understanding you will actually use. A summary removes the struggle, and much of the retention with it.
The validation tether
In practice, lawyers will work with AI, and the work will only be as reliable as the lawyer checking it. Nolan Lovett, in “The Tragedy of the Cognitive Commons” (Human Resource Development Review, 2026), distinguishes internalized mastery (deep knowledge built through sustained, unassisted practice) from distributed mastery (the skill of directing AI systems). His point is that the second depends on the first, and he calls that dependence the validation tether: your ability to check the machine is tied to what you know yourself. Anyone can check whether an AI memo is well formatted and confident in tone. Only someone who knows the law can see that it cited an overruled case, missed a statutory exception, or conceded a dispositive procedural point. Use AI to skip the foundations now and you will be an operator who cannot evaluate what the tool produces. That dependence is why the foundational work happens in 1L.
Law is judgment under uncertainty. Used well, AI can quiz, critique, search, and suggest. It does not decide. Attempt first. Verify everything. Follow each professor’s rules.
Questions or suggestions for this guide: Polk Wagner, pwagner@law.upenn.edu.