How to Redesign Assignments for the Age of AI

No assignment is AI-proof, but faculty can redesign assignments so a student's own reasoning is on the record: tie reflection to specific course work, stage the project, and add a short conversation about what was submitted. This guide is for instructors and academic technology teams deciding which redesign moves are worth the effort in a real course. It describes each move, what it does and does not show, and where an assessment platform helps and where it does not.

Why redesign is a different question from detection

Detection asks whether a text was machine-written. Redesign asks a more useful question: what evidence would convince me this student understands the material? A polished document answers neither question on its own, because a clean draft is now cheap to produce.

Redesign has limits. Any written task a student completes unobserved can be assisted by a tool. The aim is not to make cheating impossible. The aim is to make understanding visible, so that a student who did the thinking has an easy path to show it and a student who did not has a harder one.

Redesign moves that put the student's thinking on the record

Most creative assessment ideas fall into a few families. Each can be used alone, and they combine well.

• Anchor the task in specific course material. Ask students to work from a particular reading, class discussion, dataset or local case instead of a general topic. Specific material is harder to answer from generic text, and easier for you to check.
• Stage the project. Collect a proposal, an outline or draft, and a final version at separate points. The sequence shows development, though dated drafts can be fabricated, so treat staging as context and not proof.
• Add reflection tied to the work. Ask what the student decided, what they tried that failed, and what they would change. Reflection is strongest when it points to concrete choices in the submitted work.
• Add a short conversation about the submission. A brief, planned exchange in which the student explains and extends their own work. This is the move that most directly tests understanding.
• Have students conduct an interview. The student questions a practitioner, a source or a character, and submits what they learned and how they decided what to ask.
• Use in-person or observed work for the highest stakes. Some outcomes justify a supervised component. That is a policy and logistics decision for your department.

Each move costs time. Choose one or two for the assignment where understanding matters most, and leave the rest of the course alone.

Making reflection worth grading

Reflection is often added as an afterthought and graded on completion, which invites generic text. A few design choices raise its value.

• Point to artifacts. Require each reflection to cite two or three specific places in the student's own work, such as a paragraph, a figure or a line of code, and explain the decision behind each.
• Ask for the counterfactual. "What would change your conclusion?" is harder to answer in general terms than "What did you learn?"
• Connect it to something said in class. Reference a discussion, a guest talk or a feedback comment that only your section had.
• Keep it short. A tight prompt with a word limit produces more usable evidence than an open essay.
• Follow it with a question. A reflection is a claim about the student's thinking. A short follow-up conversation lets them stand behind it.

Adding an interview or short conversation

There are two different things people mean by an interview, and they suit different goals.

A conversation about the student's work. You, a teaching assistant or a platform asks questions drawn from the student's own submission and follows up on the answers. The value comes from the follow-up: a second question that depends on the first answer is hard to prepare for in advance. Keep it short and publish the rubric beforehand.

A conversation where the student does the questioning. The student interviews a client, patient, witness or stakeholder and has to get what they need. This tests how a student gathers information and adapts, which a written task cannot. See AI simulations and role-play for how such scenarios are built.

Both formats raise workload questions, so time a small sample before you commit. The pilot below shows how.

Where Rocketproof fits

Rocketproof is an authentic assessment, simulation and course intelligence platform. It is not an AI detector, it does not proctor students, and it does not judge whether a student used AI. Faculty judgment stays central.

For the moves above, it supports the following.

• A conversation about a submission. For upload-based assignments, the questions are prepared from the submitted document, and you can add your own questions before students begin. Students respond by recorded audio, live voice or text chat, depending on how you set up the assignment, and approved accommodations such as extended time apply. See AI oral exams and oral defense.
• Student-led interviews. You define an AI character, such as a patient, client or witness, and the student has to question it. See AI simulations and role-play.
• Practice rounds. An assignment can be marked as practice so it stays out of the gradebook and out of grade passback.
• Review you control. Each session leaves a transcript. Rocketproof drafts rubric-aligned scores with evidence from the transcript. You review them, change any you disagree with, and release the result. Nothing is final until you release it.

What it does not do: it does not decide what counts as reflection or process evidence in your course, it does not replace staged drafts or in-class work, and it does not issue grades on its own. If a conversation raises a concern, follow your institution's normal process, with the transcript as one piece of evidence and not a finding.

Choosing a redesign for your course

Match the move to the outcome you care about.

• If the outcome is explaining and defending an argument or method, add a short conversation about the submission.
• If the outcome is professional communication or information gathering, use a student-led interview or role play.
• If the outcome is self-awareness about process, add reflection tied to specific artifacts.
• If you only need context for an integrity conversation, stage the project, and keep the records you collect proportionate.

Check your campus policy on oral assessment, recording and AI tools first, and involve disability services early. For the broader framing, see authentic assessment in the age of generative AI.

A small pilot

1. Pick one assignment. Choose the project where explaining the work is part of the learning outcome.
2. Choose one added component. Reflection tied to artifacts, a short conversation, or an interview, not all three.
3. Write two to four outcomes. State what the added component must show.
4. Publish the rubric. Students should see the criteria before they start.
5. Run a practice round. Let students see the format with nothing at stake.
6. Plan accommodations. Decide in advance how approved accommodations apply.
7. Time your review. Measure how long it takes you to review a handful of sessions end to end, then multiply by your enrollment.
8. Ask students. Collect feedback, adjust, and then decide whether to expand.

Frequently Asked Questions

How can instructors redesign assignments for the age of AI?

Anchor the task in specific course material, stage the project, tie reflection to concrete parts of the student's work, and add a short conversation in which the student explains what they submitted. Start with one assignment and one added component, then adjust after you see the results.

Can an assignment be made AI-proof?

No. Any task completed unobserved can be assisted by a tool. Redesign makes a student's understanding easier to see and shallow work easier to notice. It does not make misuse impossible, and observed or supervised work remains an option for the highest-stakes outcomes.

How do you make reflection assignments harder to fake?

Require students to cite specific places in their own submitted work, ask what would change their conclusion, and reference class-specific material. Then follow the reflection with a brief conversation so the student can stand behind it.

What is the difference between an oral assessment and an interview assignment?

In an oral assessment, someone asks the student about their own work and follows up on the answers. In an interview assignment, the student does the questioning, for example of a practitioner or a simulated character. The first tests understanding of a submission. The second tests how a student gathers information and adapts.

Does Rocketproof detect AI use or grade the work?

No. Rocketproof is not an AI detector and does not proctor students. It conducts structured conversations, keeps a transcript, and drafts rubric-aligned scores with evidence. The instructor reviews the evidence, decides, and releases the grade.

See a conversation about student work

Try the public demo as a student or instructor, or request a demo to talk through a pilot for one assignment in your course.