Self-serve toolkit
Assignment Toolkit
Make an assignment more AI-resilient, or build AI in on purpose, while keeping your learning goals. You don't need every idea here; one or two changes can make a real difference.
Is this assignment AI-vulnerable?
A quick check
- Test it. Paste your assignment prompt into the NU Claude Portal. Would the answer earn a passing grade?
- Could a student answer well without doing the reading, or from general knowledge instead of your course material?
- Does the prompt only ask “what do you think?” or “what resonates with you?”
If the answer to any of these is yes, pick your assignment type in the tabs above for concrete changes.
Before redesigning, decide your AI level. Is AI prohibited, permitted, encouraged, or required on this task? See assignment labels. If it's encouraged or required, go to .
Discussion prompts & qualitative work
The Four Levers, from Linda Kowalcky's Discussion Prompt Redesign Checklist
1 · Anchor to something specific: a local policy, a campus event, a dataset, or a class discussion.
“Pick a real policy that affects your community and apply this week's principle to it.” Give students a non-personal option if the anchor involves their own life.
2 · Trace the “why”, not the “what.” Ask for the reasoning, not a summary.
“Pick one design element; explain why that one had to change.”
3 · Build in peer dependency. The follow-up needs classmates' specific posts.
“Compare examples across your group; report the strongest pair and why.”
4 · Require a textual reference: a quoted passage or term from the reading in the pre-discussion post.
Try it: use the printable Discussion Prompt Redesign Checklist, and see the levers applied in three weeks of redesigned public policy prompts.
Papers & essays
Works even without in-class drafting time
Connect the text to class discussion: require named passages and a point raised in your class sessions. Class discussion is harder to reproduce than the reading itself.
Add a short process memo (½–1 page): the thesis, what changed as they read, the turning-point passage. Quick to skim, and it rewards genuine engagement.
Add a brief oral check-in: a 3–5 minute “walk me through your argument” in office hours, a class slot, or a 2-minute video.
Grade drafts and revision: a draft, feedback, and a note on how the student used the feedback.
Quantitative / data assignments
Grade the thinking, not just the output
Grade interpretation: AI can run the regression; students must interpret it, critique it, and justify their choices.
Anchor to a chosen dataset or case, plus a decision log (“why this method, why this variable”).
Check live: a short in-class replication, or a moment where students explain their code.
Moves that work on any assignment
Separate practice from assessment: keep practice low-stakes, and put high stakes where you can see the work (in class, oral, or proctored).
Grade process as well as product: drafts, reflections, version history, checkpoints.
Protect the early thinking: students do their first independent thinking before AI enters.
Say what's expected, and why: name the skill the task builds and why the shortcut skips it.
Mix individual and team work: individual work first, then a team stage that builds on it.
Build AI in on purpose
For assignments labeled Permitted, Encouraged, or Required. The university standards recommend that any AI-integrated task include verification and critique.
Five skills to build across a course
- Ask good questions: practice framing clear, specific prompts.
- Check the answers: verify AI output against sources instead of trusting it.
- Argue with the AI: find its errors, gaps, and weak reasoning.
- Know when not to use it: recognize when AI would skip the learning.
- Ask students: gather feedback on when and how they actually use AI.
The focus is judgment, not usage. Keep the structure and swap in your discipline's content. The designs below put these skills into practice.
Critique the AI: students ask an AI tool a question about your course material, then find what it got wrong, missed, or oversimplified.
- Goal
- Close reading, and seeing AI's limits firsthand
- Students submit
- The AI's answer, annotated, plus a short corrected version citing the source
- Grade on
- Accuracy of the critique and use of evidence, not the AI's answer
- Example
- Argue with the AI (economics, adaptable)
Do it yourself first, then with AI: students complete the core task on their own, then use AI and compare.
- Goal
- Build the foundational skill, then learn when a tool helps
- Students submit
- Their own version, the AI-assisted version, and a reflection on the differences
- Grade on
- The first version and the quality of the comparison
Compare and test: students give the same prompt to two tools, or rephrase it, and evaluate the results.
- Goal
- Understand how AI output varies and what makes it trustworthy
- Students submit
- Outputs side by side, with checks against course sources
- Grade on
- The criteria students use to judge quality
Document the process: students keep their prompts or chat transcripts and an AI journal.
- Goal
- Transparency and reflection on appropriate use
- Students submit
- A disclosure statement or transcript with the assignment
- Grade on
- Completeness and honesty of the reflection
Have students use university-supported tools such as the NU Claude Portal so no one needs a paid account.
Check for authentic work
Designs that make students' own understanding visible, without AI detectors
Short oral conversation: an 8–15 minute talk where students walk you through their own work. It shows understanding and doubles as interview practice.
How to run one: logistics, questions, and a rubric
Logistics
- When: the week after the project is submitted.
- Format: one-on-one in office hours or on Teams, booked with the Canvas scheduler.
- Weight: part of the project grade (for example, 70% the work, 30% the conversation), so a polished submission alone can't carry a student who can't explain it.
- Large classes: shorten to 8 minutes, talk with a random subset each cycle, or start with a 2-minute recorded walkthrough.
- Framing: “This is how professional interviews work: walk me through your work.” A normal part of the course for everyone, not an accusation.
Questions (pick 4–6, easier to harder)
- In one sentence, what question does your project answer, and what did you find?
- Walk me through how you got from your sources or data to your main conclusion.
- Why did you choose this approach, method, or set of sources? What else did you consider?
- What's the key assumption or interpretive move your argument depends on? Why is it reasonable?
- What's the strongest objection to your conclusion, and how would you answer it?
- Point to one piece of evidence and explain what it does and doesn't show.
- If AI was allowed: show me where it helped. What did you ask, and how did you check it?
- Where did AI get something wrong, or where did you decide not to use it, and why?
Rubric (score each 1–5)
| Dimension | 1 · Not yet | 3 · Competent | 5 · Strong |
|---|---|---|---|
| Process | Can't reconstruct the steps | Explains the main steps | Fluent, including small choices |
| Reasoning | States the approach, not why | Names the key assumption | Defends it against real objections |
| Evidence | Repeats the conclusion | Interprets evidence correctly | Adds limits and uncertainty |
| Limits | None identified | Names one limitation | Anticipates where it could fail |
| Judgment | Can't explain own work or AI use | Explains most choices | Checks AI output; knows when not to use it |
Adapted from the replication interview Richeng Piao uses in his data and econometrics courses.
Anchor part of the grade to something you can see: pair take-home work with a short in-class checkpoint or quiz.
Project plus a brief defense: a real-data, archival, or replication project that students present and answer questions on.
Live in-class activities: polls, simulations, and games that produce class-generated material students then build on.
Keep it fair
- Offer alternatives for students with accommodations (for example, a written reflection instead of a live oral), and coordinate with Disability Access Services when needed.
- Frame oral check-ins as a normal part of the course for everyone, not an accusation.
- If you suspect misuse, the university asks you to start by asking the student to explain their process. See the rules.
Grounded in Northeastern's AI Readiness Framework, the Assessment in an AI-Enabled World faculty guide, and the university's AI standards. Discussion-prompt levers adapted from Linda Kowalcky's checklist. Want help applying these to your course? Request a consult with CATLR.