Toolkit / Assignment Package: Independent First Pass + AI-Supported Revision
Assignment Package: Independent First Pass + AI-Supported Revision
A complete law-school assignment pattern that preserves evidence of independent analysis while teaching students to supervise AI critique and make defensible revisions.
Use this pattern when students need to demonstrate an independent first attempt and learn how to evaluate AI feedback. Students submit the protected first pass before opening an AI tool, then submit a revision and a short decision memo. The two artifacts answer different questions: What could the student do independently? What could the student improve while supervising assistance?
This is a design pattern, not Law School policy. Adapt it to the course, current Penn guidance, and any controlling exam, clinic, client, or placement rules.
Purpose
The assignment makes two capabilities visible without treating AI polish as learning:
- independent legal analysis or communication under stated conditions; and
- revision judgment—requesting useful critique, checking it, and deciding what to accept or reject.
It works best when the revision is consequential but bounded. It is not a reason to collect a full chat history unless that history serves a stated outcome.
Appropriate courses
- Legal practice skills, research and writing, drafting, and advocacy
- Doctrinal courses using short written problems or practice answers
- Seminars using an early thesis, abstract, or analytic section
- Clinics and simulations only when supervision and confidentiality rules permit the proposed tool and information
Avoid this pattern when the entire submission must demonstrate unassisted performance or when students cannot safely or equitably access an approved tool.
Learning outcomes
By the end of the assignment, students should be able to:
- produce an independent first-pass analysis or professional communication;
- identify the most important weaknesses in their own work;
- solicit critique targeted to audience, structure, reasoning, or counterargument;
- verify suggestions against facts, assigned materials, and legal authority;
- revise selectively rather than defer to fluent output; and
- explain consequential revision decisions.
Assignment sequence
| Stage | Student work | AI rule | Evidence retained |
|---|---|---|---|
| 1. Analyze | Read the problem and authorities; plan and draft | Prohibited | Student’s own notes if ordinarily required |
| 2. Submit first pass | Upload a time-stamped complete draft | Prohibited through submission | First-pass file |
| 3. Diagnose | Identify two or three revision priorities | Prohibited or permitted, at instructor’s choice | Brief self-diagnosis |
| 4. Seek critique | Ask AI for bounded feedback, not a replacement answer | Permitted in an approved tool | Useful prompts or concise use record |
| 5. Verify and revise | Check every substantive suggestion; revise selectively | Permitted | Revised product |
| 6. Explain | Identify important accepted and rejected suggestions | Permitted for proofreading only, unless stated otherwise | Decision memo and disclosure |
Student-facing instructions
Prepare and submit a complete first pass without generative AI assistance. Until that file is submitted, do not use AI to analyze the problem or to outline, draft, rewrite, translate, or polish your response.
After submission, you may use an approved AI tool to critique your first pass. Ask for feedback targeted to the assignment criteria—for example, missing counterarguments, unclear organization, audience fit, or places where the reasoning does not follow. Do not ask the tool to replace the response. Do not assume its legal or factual suggestions are correct.
Submit (1) the unchanged first pass, (2) the revised product, and (3) a short decision memo identifying the most consequential changes, one suggestion you rejected, and how you verified substantive advice. Your grade will reflect the capabilities listed in the rubric, not the volume of AI use.
AI rule
Before first-pass submission: no generative AI for the substance, organization, wording, or editing of the work. State separately whether ordinary spelling, citation-formatting, accessibility, translation, or assistive tools are permitted.
After first-pass submission: AI may provide critique within the announced boundaries. Students remain responsible for authority, accuracy, confidentiality, and the final work. Do not require uploading student work to an unlicensed tool. Check the AI Resources Portal and current Penn guidance before specifying a tool or data workflow.
Disclosure
AI-use statement: I used [tool] after submitting my independent first pass. I used it to [purposes]. I provided [general description of material, not sensitive content]. AI-generated wording [does/does not] appear in the revision. I verified substantive suggestions by [method]. The consequential suggestions I accepted or rejected are explained in my decision memo.
If no AI was used after the first pass, permit a one-sentence statement. Optional use should not carry a grade penalty.
Deliverables
- Unchanged, time-stamped first pass
- Revised product
- Decision memo (suggested length: 300–500 words)
- AI-use statement
- Only the prompt/output excerpts needed to support a disputed or graded process decision, if required in advance
Evaluation rubric
| Dimension | Weight | Successful evidence |
|---|---|---|
| Independent analysis | 30% | First pass identifies governing issues, uses relevant authority, and reasons from law and facts at the expected course level. |
| Final legal quality | 30% | Revision is accurate, well-supported, responsive to audience and purpose, and professionally communicated. |
| Revision judgment | 25% | Changes address important weaknesses; student accepts, modifies, or rejects suggestions for defensible reasons rather than following them mechanically. |
| Verification | 10% | Substantive claims, quotations, facts, and authorities affected by assistance are checked against reliable sources. |
| Explanation and compliance | 5% | Decision memo is specific and candid; required files and disclosure are complete. |
Change the weights to match the course. If independent performance is the principal construct, give the first pass more weight. If revision is formative, grade it lightly or provide feedback rather than points.
Class-size variants
Small class: conference on the first pass; individualized AI-critique plan; full decision memo.
Medium class: common prompt menu; peer discussion of accepted/rejected suggestions; rubric-coded feedback.
Large class: short first pass; automated time-stamped submission; 200-word decision memo; audit a disclosed sample rather than every chat transcript. Use teaching assistants only under a shared scoring guide and calibration examples.
Accessibility and equity
- Provide an equivalent human or structured feedback option when AI use is optional or inaccessible.
- Explain the protected first-pass conditions in advance, including permitted assistive technology.
- Do not infer weak independent ability from disability-related tools, language supports, or approved accommodations.
- Keep the decision memo concise; excessive process documentation can reward time and tool fluency rather than the intended capability.
- If oral explanation is added, provide practice, criteria, accommodations, and an equivalent alternative where needed.
Evidence and limitations
Backward design supports separating the capability to be learned from the evidence used to assess it. A legal-education experiment found that AI assistance can improve some work products, but it did not establish durable learning or independent competence. This package is therefore a design recommendation: no identified study directly compares learning outcomes from this precise sequence.
Use the pattern to make the two performances observable, not to claim that revision with AI necessarily improves learning. For the broader validity frame, see Dawson et al., “Validity Matters More Than Cheating”, and Schwarcz et al., “AI-Powered Lawyering”.