Prep your law class like it's 2026

This post will show law faculty how to prep their classes better and faster using modern AI methods. I'm going to make this concrete by preparing a class session on the unconstitutional conditions doctrine, which I feel I need to add to my forthcoming 2026 1L Constitutional Law class.

The Black Letter Law CustomGPT

The first thing I want to do is have AI help me get a lay of the land by providing the black letter law. As it happens, I feel I know this area tolerably well, but there may still be insights that AI can contribute. I'm going to use a CustomGPT that I built months ago and that has been used by over 300 people since that time. I go to ChatGPT and search for black letter law.

I now enter that CustomGPT and make sure it is using ChatGPT 5. I ask it for "Black letter law of the unconstitutional conditions doctrine. make sure to emphasize key Supreme Court cases." Here's a sample of the results I obtain.

The AI has conceptualized the unconstitutional conditions doctrine more broadly than had I, bringing in a larger number of cases. ChatGPT identifies not just the core cases of Perry v. Sindermann (1972) or Rust v. Sullivan (1991) but also cases such as Trinity Lutheran Church v. Comer (2017), which it conceptualizes not so much as a free exercise of religion case but as conditioning the opportunity for state government playground funding on a school's non-religious identity. I might not teach all of the cases the CustomGPT identified as part of this unit but I can make connections between the core cases and these other cases when I teach each of these units. So, AI has already improved my teaching.

Based on this analysis, I'm going to center my class around three cases: Perry v. Sindermann, Rust v. Sullivan, and Agency for International Development v. Alliance for Open Society International, Inc. (2013). Unfortunately, none of these cases are emphasized in the book I am assigning. Ordinarily, I might scrounge around other casebooks to see if they have copies or look for edited versions in Harvard's H20 open casebook initiative, or spend hours editing each of the cases down to a reasonable length while making sure to preserve coherence. But, as shown in a prior blog post, I now have an AI tool that can produce a custom edited version of the case for me. This way, I can make sure that each case has the length and emphases I want; I can also make sure that the cumulative reading burden on the students is sensible. Subscribing to legaled.ai is so valuable!

Have AI edit the key cases

I tell Claude to load up its Case Editing skill that I discussed in an earlier blog entry. I now give Claude three prompts. I'm targeting about 6,000 words of reading for the students.

Use your casebook editor skill to give me a 2000-word version of Perry v. Sinderman. It is part of a unit on what will come to be known as the unconstitutional conditions doctrine.

Find Rust v. Sullivan on the web. Then use your case editing skill to create Rust v. Sullivan at 2,200 words while preserving discussions distinguishing Perry v. Sindermann and related unconstitutional conditions cases.

OK. Now use that same skill to produce an 1800-word version of Agency for Int’l Development v. Alliance for Open Society Int’l (AOSI) (2013) – anything distinguishing Rust v. Sullivan is important. You can find the opinion here. https://supreme.justia.com/cases/federal/us/591/19-177/

It takes Claude maybe 10 minutes to produce my documents and make them available for download as either Markdown or PDF. I'm going to use Markdown so I can more easily do some custom edits on the work.

Combine two LLMs to create a note to follow the cases

Next, it's time to have AI bring in all the peripheral cases it identified. I want a note I can give the students to complement their case readings. Here's the prompt I give the CustomGPT that earlier identified the key cases in the field.

I regard the core UCD cases as being Perry v. Sinderman, Rust v. Sullivan and AOSI. What I would like you to do is write a 750 word essay for law students connecting up the other areas you cited earlier. For example, I thought it was interesting how you conceptualized Trinity Lutheran as a disguised UCD case. Discuss these other areas and show how they arise in different contexts – land use, abortion funding – but how they have a UCD flavor attached to them. Also mention how the context in the UCD cases and the doctrine differs, for better or worse, from the South Dakota v. Dole and the NFIB v. Sebelius case in which the 10th amendment rights of states is implicated. Each case should be identified by name and year but there is no need for a volume, reporter, page citation.

The result out of ChatGPT is passable, but parts of it reek of AI slop; overall, I feel the response could be sharper. I use a standard technique to improve AI prose: give it to another AI to critique and improve. AI's are sometimes better at finding problems with other AI prose than they are at generating great AI prose themselves. Because the task is intellectually challenging, I enlist Claude's most sophisticated current model: Opus 4.1.

Critique and improve this essay. Make it intellectually sharper. Eliminate AI writing tropes and cliches. But do not lengthen it. Audience is law students.

Opus is smart enough to see exactly what was wrong with the ChatGPT variant.

Main Problems:

Formulaic opening: The three-case setup reads like a Wikipedia introduction

Vague metaphors: "radiating into areas," "restored balance," "cousin to the UCD"

Mechanical transitions: "Two decades later," "Similarly"

Underdeveloped tensions: The essay glosses over the genuine doctrinal contradictions between cases

Conclusory language: Phrases like "enduring tension" and "single constitutional value" substitute for actual analysis

After Opus is done with revisions, we have a more sophisticated take on the doctrine that weaves in the other cases. Here's how the revised version starts. I feel fine about giving this material to my students as a note following the main three cases.

The unconstitutional conditions doctrine prohibits government from achieving indirectly what it cannot do directly—yet this simple formulation conceals profound disagreements about when persuasion becomes coercion. The modern doctrine emerged from employment retaliation (Perry v. Sindermann), evolved through funding restrictions (Rust v. Sullivan), and reached its current form in compelled speech cases (AOSI). But the doctrine's real interest lies in its contradictions.

Have an app prepare a lesson plan

Now it's time to prepare a class unit on the subject. I give the cases to an app I have built and now make available free of charge to my subscribers: law-professor-class-prep. Here's a screen capture of the app.

I give the app the following prompt in the "Additional Instructions" section. (I preserve my typos; AI is very forgiving):

A 45 minute class that give the class the background they need, notes the tension between the Perry v. Sinderman view and the Rust case, discusses how the AOSI case can make sense of the two but then turns to the many modern applications of the doctrine. Have the class think about this recent proposal: %%% The Trump administration's new rule amends the Public Service Loan Forgiveness (PSLF) program, allowing the Education Department to deny loan relief to public workers at organizations with a "substantial illegal purpose." This includes human trafficking, gender-affirming care for transgender youth (termed "chemical castration"), aiding illegal immigration, or supporting terrorism. It targets nonprofits and public entities on immigrant or LGBTQ+ issues, potentially affecting teachers, doctors, and firefighters qualifying after 10 years.
Denials can be based on court rulings, settlements, or the secretary's "preponderance of evidence" determination. Effective post-July 1, 2026, barred groups face 10-year exclusions. Officials claim it prevents subsidizing lawbreakers, impacting under 10 groups annually. Critics call it ideological overreach that harms essential workers and invites abuse. PSLF, established in 2007, has helped over 1 million.

The app produces a lesson plan. It's not genius, but it's a start.

Lesson Plan

Class: Unconstitutional Conditions and the Modern Administrative State (45 minutes)

Goal: Understand the tension between the government's power to spend and define programs, and its inability to condition benefits on the relinquishment of constitutional rights, applying this framework to a contemporary policy debate.

Phase 1: Orientation & Perry (10 minutes)

Opening puzzle:
Imagine you’re a public school teacher and love your job. Next year, your contract isn't renewed. Your principal tells you it's because you publicly criticized the school board’s curriculum choices. Can the government do that?

Teachable tension:
The government provides myriad benefits (jobs, funding, licenses, tax breaks)—but how far can it go in conditioning those benefits on someone giving up their constitutional rights?

Instructor: (Briefly introduce "unconstitutional conditions" as the core problem.)

"Today, we explore the 'unconstitutional conditions' doctrine. The government gives out lots of benefits – jobs, funding, permits. Can it attach any condition it wants to those benefits, even if the condition requires you to give up a constitutional right?"

Perry v. Sindermann (1972): Public Employment and Free Speech

Setup: Professor Sindermann, a public college teacher, criticizes his administration. His contract isn't renewed. No formal tenure, no hearing.
Question to class (Perry Part I): Sindermann claims his non-renewal was retaliation for his speech. The college says, "we just chose not to rehire a non-tenured employee – no 'right' to the job." Who wins that argument, and why?
Key takeaway (Instructor): The Court says even if you have no "right" to a benefit (like a non-tenured job), the government cannot deny that benefit for unconstitutional reasons, specifically retaliating against your protected speech. To do so would "penalize and inhibit" free speech. "It may not deny a benefit to a person on a basis that infringes his constitutionally protected… freedom of speech even if he has no entitlement to that benefit." (Perry, quoting some older cases).
One-fact delta: What if Sindermann was fired for repeatedly showing up late to class and failing to teach required material?
- Expected answer: That's a legitimate, non-speech related reason. The government can fire for valid job performance issues.

Phase 2: Rust vs. AOSI (20 minutes)

Instructor: "Alright, so Perry establishes a baseline: government can't punish protected speech by denying a benefit. But what if the government is just choosing to fund certain activities consistent with its policy goals? Does that always count as penalizing speech?"

Rust v. Sullivan (1991): Defining the Scope of a Program

Setup: Title X provides federal funds for family planning. Congress says, "No Title X funds 'shall be used in programs where abortion is a method of family planning.'" HHS regulations go further: no counseling, referral, or advocacy for abortion even as a medical option within Title X projects. Grant recipients, usually health organizations, want to offer comprehensive advice.
Question to class: Does this violate Perry? Is the government denying a benefit (Title X funding) because doctors want to engage in constitutionally protected speech (abortion counseling)?
Key takeaway (Instructor): The Court (Rehnquist) says NO. This is different. The government, as a funder of a distinct program, can define the scope of that program. It's not stopping doctors from talking about abortion elsewhere or with other (non-Title X) funds. It’s simply saying, "we will only fund these specific activities through this program." The key is the ability to maintain physical and financial separation. "When the government appropriates public funds to establish a program, it is entitled to define the limits of that program."
Teachable Tension (between Perry and Rust): Perry says the government can’t pressure you out of your speech. Rust says the government doesn’t have to pay for your speech. Where’s the line?

AGENCY FOR INTERNATIONAL DEVELOPMENT v. ALLIANCE FOR OPEN SOCIETY INTERNATIONAL (AOSI) (2013): Affirming Beliefs

Setup: The Leadership Act funds anti-HIV/AIDS NGOs. It has two conditions:
1. No funds "may be used to promote or advocate the legalization or practice of prostitution or sex trafficking." (the "use condition," like Rust)
2. No funds may be used by an organization "that does not have a policy explicitly opposing prostitution and sex trafficking." (the "Policy Requirement")
Question to class: How does the "Policy Requirement" compare to the Rust condition? What’s the critical difference?
Key takeaway (Instructor): The Court (Roberts) strikes down the Policy Requirement. This isn't just defining the program's activities (like the "use" condition, which is fine); it's forcing the organization itself to adopt a particular belief, even in its privately funded activities outside the scope of the federal program. This crosses the line from defining a program to compelling speech/belief.
- Roberts quotes Barnette: "If there is any fixed star in our constitutional constellation, it is that no official, high or petty, can prescribe what shall be orthodox in politics, nationalism, religion, or other matters of opinion or force citizens to confess by word or act their faith therein."
- Mechanism: Rust worked because you could separate your Title X activities from your other activities. Doctors could still discuss abortion with private patients, or in their private time, without violating Title X rules. Here, the "Policy Requirement" applies to the organization's entire identity and advocacy, making separation impossible.
Teachable Tension Refined: Government can define what it funds (Rust), but not who you are or what you believe outside that funding (AOSI). The critical element is whether the condition imposes a belief or restricts activity that cannot be confined to the scope of the government-funded program.

Phase 3: Application to Modern Hypo (15 minutes)

Instructor: "Let's apply this framework to a recent real-world proposal."

Hypothetical: PSLF "Substantial Illegal Purpose" Rule (The Trump Administration Proposal)

Context: The Public Service Loan Forgiveness (PSLF) program offers student loan forgiveness to public and non-profit workers after 10 years of payments.
Proposed Rule: The Education Department gets to deny loan relief if the public worker is at an organization with a "substantial illegal purpose." Examples given: human trafficking, gender-affirming care for transgender youth (termed "chemical castration"), aiding illegal immigration, or supporting terrorism. Denials can be based on court rulings, settlements, or the Secretary's "preponderance of evidence" determination.
Counter-narrator moment: A lawyer for the Education Department defends this rule using Rust.
- Education Department's Argument: "The PSLF program was designed to encourage public service that benefits society. We, as the government, are simply defining what kind of public service we want to incentivize. We are choosing not to subsidize service at organizations whose mission we find problematic or contrary to public policy. We're not telling people what to believe; we're just setting eligibility criteria for our loan forgiveness program. This is just like Rust defining the scope of Title X funding. We don't want to use federal money to indirectly support what we deem 'illegal purposes' through loan forgiveness."
Class Discussion:
1. Which precedent applies most directly here: Perry, Rust, or AOSI? Why?
  - Guide students: Is it a direct punishment for speech (Perry)? Is it defining the scope of how money can be used within a program where separation is possible (Rust)? Or is it compelling a belief or regulating the entire organization's existence beyond the scope of direct federal funds for specific projects (AOSI)?
  - Think about: The "organization that does not have a policy explicitly opposing X" language from AOSI versus "funds may not be used for Y" from Rust.
2. What if the rule said, "Loan forgiveness will not be granted for time spent directly engaging in gender-affirming care for transgender youth"? Is that different?
  - Guide students: This gets closer to Rust – defining the activity that qualifies for the benefit. But even then, the controversy is about "illegal purpose" and the vagueness of that, and singling out specific types of care that are not universally deemed illegal.
3. What if a public defender works for a nonprofit that also, on its own dime, advocates for looser immigration policies (which the Administration deems "aiding illegal immigration")? Could their PSLF be denied?
  - Guide students: This pushes into the AOSI territory of regulating the organization's broader mission or "policy" rather than the specific activity being subsidized by PSLF (i.e., the public defender's day job). The "illegal purpose" here is not directly the public defender's work itself, but perhaps their employer's other activities or stance.

Exit Bridge:
"If you are concerned about this proposed PSLF rule, what legal argument, informed by these cases, would you develop for a lawsuit challenging it?"

This lesson plan forces students to grapple with the nuances of the unconstitutional conditions doctrine, moving beyond simple slogans to the precise mechanisms the Court uses to distinguish permissible program definition from impermissible compelled speech or penalized association. The modern hypo grounds the abstract judicial distinctions in a timely and politically charged policy debate.

Use a Claude Skill to generate a black letter law/bar review Word File

Great. But we're not done yet. First, I want to create a better version of the "black letter law" for myself. I don't think I will give it to the students. Better they figure it out for themselves or at least go through the trouble of using AI. I could recycle what I produced at the beginning of this blog, but I have a somewhat more sophisticated Claude Skill that is likely to do a better job. (By the way, subscribers, that skill file might prove very useful even if you are not a law professor but are a law student, practitioner or just curious).

Use the black letter law skill to summarize the unconstitutional conditions doctrine. Here are some materials that you should emphasize.

The result is a nice Word document. Here's a screen capture of part of its first page.

Create a class presentation

Have Claude create a PowerPoint

I'm old school enough to believe that a PowerPoint or similar presentation is useful in class. It gives the students something to focus on and helps me stay organized in the classroom. So let's use our lesson plan, the cases and the black letter law to create one. We'll try it two different ways and see which one I like better. First, let's use Claude's ability to directly produce PowerPoints. I load it up with the edited cases, the note weaving in the other cases, the lesson plan, the black letter law word file, and the article on the proposal from President Trump to limit PSLF funding based on the ideology of the employer.

After Claude's first effort crashed, I coaxed Claude to try again. The result is a useful document. It might not be exactly what I would produce but, because it is in an editable document, I can take a strong foundation and tweak it rather than suffer the first world problem of producing a PowerPoint from scratch. Here's a screen capture of a slide.

Have Gemini create a Beamer presentation

I really like Claude. As I write this blog entry, it is the best large language model for law professoring. But I also recognize that some faculty may not want to shell out $20 a month (or even more) for paid Claude and have invested in ChatGPT or Gemini. Plus, in the months ahead, one of those other AIs may well leapfrog over Claude in capability. (To avoid the bet-on-the-wrong-horse problem, sacrifice a few dollars to purchase AI plans by the monthly rather than making an annual commitment). So, I also examined whether a professor without paid Claude could produce a useful presentation. The answer is yes. I gave ChatGPT the same materials and asked it to produce a 15-slide beamer presentation. After doing a cruddy job and being corrected – I dislike progressive disclosure slides – it ultimately came up with something acceptable. Again, because beamer presentations are editable, I can tweak the output in Overleaf or other LaTeX editors. Here's a screen capture of a slide.

Generate a NotebookLM notebook and derivative products

I now want to create interactive and multimedia materials for the students. I'm going to use the one best tool for legal education: NotebookLM. I upload all the materials we've produced thus far but I then supplement by asking NotebookLM to itself discover sources relevant to the unconstitutional conditions doctrine. You can see the result here.

With the sources in place, I ask for three products: (1) a video overview; (2) a debate-style audio-overview, and (3) a quiz that the students can take.

Of course, once I have materials inside NotebookLM, the sky's the limit. Indeed, what I do is give my students a somewhat minimalist notebook that contains the source materials and one or two derivative products but suggest that they produce their own derivative works. They can produce the flashcards on their own, they can generate custom reports or their own blogs. As discussed in a prior blog entry, NotebookLM is this factory that lets the student produce whatever they think useful either directly or by treating the output from NotebookLM as input to a specialized AI.

Conclusion

At the end of the day I have a complete packet for the students. It's all contained in a ZIP file that I share here. I have cases edited to the length I desire and emphasizing the points I think are important. I have one intellectual note following the case and could easily have generated more. I have a Word document summarizing the black letter law in the field. I have PowerPoint slides for class. And I have a NotebookLM notebook with which the students can interactively engage and that contains a debate on the public loan forgiveness example that should make the doctrine very relevant. All of this took about two hours, but a lot of it was spent watching AI toil away. That enabled me to multitask and get other work done (like writing this blog entry).

All this use of AI to prepare does not mean it is yet possible for the Associate Dean to press a button and say, "create materials for a section of constitutional law that addresses the unconstitutional conditions doctrine." I still have a job as a human law professor. Someone, after all, had to identify the unconstitutional conditions doctrine as an area that needed to be added to the syllabus for a class in 2026. Someone needed to use a little background knowledge of constitutional law to confirm the key cases. Someone needed to ensure that the AI edits of those cases were sensible. Someone needed to realize that the initial note looking for tensions and contradictions in the unconstitutional conditions doctrine wasn't very good and identify a vehicle for improvement. The work still needed a human orchestrator with informed judgment.

It may well be that in two or three years there will be video avatars that can deliver classroom instruction awfully well and actually interact with the students. Until such time, however, someone (me) needs to get this material into their head and stand and deliver before the students. I can't be replaced yet. Neither can the students. Still, I very much like the idea of having something better than a casebook and teaching manual on a custom topic prepared to my specifications within a few hours. That's how AI can work to our advantage in legal education.

Notes

I used a Claude Skill I made a few weeks ago, Skill2App (itself made using the amazing Claude Skill Building Skill), to go from a Claude Skill that produces lesson plans to an app that produces lesson plans. In other words, I used the Claude Skill Building Skill to create a Skill – call it law-professor-preparation-skill) that would produce lesson plans. That law-professor-preparation-skill is just a long Markdown file. I then fed the law-professor-preparation-skill file into Skill2App and asked Claude to write Python or JavaScript code that would emulate the capabilities of the law-professional-preparation-skill but would do so outside of Claude. Claude complied in defeating its own exclusivity and wrote code that got me 90% of the way home. I then imported the Claude draft app into Visual Studio and used Claude Code to debug and refine the app. I have placed the results of that process on GitHub. How long did this all take, you ask? Probably 2 1/2 hours, though because of the way AI's work, there were opportunities while the AI was thinking to work on other tasks (or watch the World Series as it extended deep into the night).
Confession: The lesson plan produced by the app is in Markdown not in HTML as shown in the blog. I had an AI convert to HTML before showing it to you because I wanted the plan to stand out on the page and Markdown does not understand background colors.