AI-Resistant Assignments: 5 Formats That Make Student Reasoning Visible

When students have access to generative AI, a polished submission may leave questions about their understanding. Assessment design can provide additional evidence through explanations, intermediate work, and discussion.

A useful alternative is to design assignments that make a student's reasoning easier to examine. AI-resistant does not mean AI-proof. The five formats below combine evidence, process, and discussion so a polished final answer is not the only basis for assessment. The examples are illustrative teaching designs, not reports of measured classroom outcomes.

What Makes an Assignment AI-Resistant?

An AI-resistant assignment asks for evidence of learning beyond a finished artifact. Three design principles can help:

• Personal context. Ask students to connect their answer to specific observations or evidence they collected.
• Process alongside product. Review intermediate work and the reasons for revisions.
• Live interaction. Use a discussion, demonstration, or follow-up question to explore understanding.

These principles can make assessment more informative, but students may still use AI. Explain permitted uses and assess their ability to account for the work.

Decision Simulations and Case-Based Reasoning

The most underrated AI-resistant format is the one where there is no single right answer, only better and worse arguments.

A decision simulation gives students a realistic scenario with competing constraints. For example, a business ethics student might receive a case about a manufacturing company with a defective part that would cost $400k to recall. The CEO wants to delay. The plant manager is being pressured by the board. The student must decide what to do, and, crucially, explain which stakeholder's interests they prioritized and why.

AI can produce and revise plausible ethical arguments. Follow-up questions are useful because they let you explore whether the student can explain the trade-offs and respond to a changed constraint. They do not automatically reveal who wrote the original answer.

Illustrative activity: Give students sequential updates to a political crisis over two weeks. Ask them to revise their recommendation after each update and explain which evidence changed their view.

Oral Defenses and Live Presentations

Ask students to defend submitted work in a short live discussion. This creates another opportunity to examine understanding, alongside the written submission.

Use follow-up questions tied to visible choices in the work, such as why a student selected one method over another. Allow thinking time and appropriate accommodations. Fluency or hesitation alone is not evidence of authorship.

Illustrative activity: After a capstone paper, hold a 15-minute discussion with questions specific to the student's argument. Use the same criteria across students and document how the discussion informs the assessment.

Scaling tip: For large classes, use small-group defenses (3–4 students present to each other with the instructor rotating) or asynchronous video responses where students answer 3 personalized questions on video with a 24-hour submission window.

Process-Based Assessments and Reflective Logs

If the final product is what AI can fake, grade the journey instead.

Process-based assessments ask students to show their work at every stage, not as busywork, but as evidence of authentic thinking. A lab notebook, a weekly reflection log, a design journal with dead ends documented and aborted approaches explained.

Use specific prompts: which source changed the student's hypothesis, what was the earlier hypothesis, and why did a discarded approach fail? Check these accounts against intermediate work. AI can fabricate plausible process narratives, so specificity should be paired with evidence.

Illustrative activity: Replace a single coding submission with a development portfolio containing:

• The original problem statement
• Three solution attempts
• An explanation of the revisions
• A short walkthrough of the final design

Use the portfolio and walkthrough together to examine understanding. Do not assume the format eliminates cheating.

Personalized and Contextual Assignments

Connect an assignment to evidence students collect and can explain.

Personalized assignments tie the task to the student's lived experience, location, or identity. Instead of "Write an essay on the effects of industrialization," assign "Write an essay connecting industrialization to changes in your hometown between 1900 and 1950." Instead of "Compare two philosophical arguments for free will," assign "Compare how free will is treated in a work of art or literature from your cultural tradition."

Students can give AI their local evidence, and AI may already know the location. Assess the quality of the source collection and the student's ability to justify the analysis, rather than assuming personal context makes AI use impossible.

Illustrative activity: Ask environmental science students to collect water samples under the course's laboratory and safety procedures, analyze the results, and explain how the measurements support their conclusions.

In-Class Data Tasks and Hands-On Work

When you need a controlled environment, bring the assessment into the classroom (physical or synchronous online).

In-class data tasks ask students to apply a method to a dataset during a supervised session. State which resources are permitted and provide appropriate time and accommodations. Novel data alone does not prevent AI assistance.

This format works across disciplines: a statistics student interprets a regression output they didn't generate; a literature student annotates a poem they've never read; a chemistry student runs a titration under observation. The assessment is the act of doing, not the artifact produced afterward.

Illustrative activity: Run a data workshop with an unfamiliar dataset and scaffolded analysis questions. If AI is permitted, require students to document its use, check its output, and explain their conclusions.

Putting It Into Practice: Start With One

Start with one assignment. Add an oral defense, an evidence-based reflection, or a supervised task, then review how well the new format reveals the intended learning.

For a decision-based activity, LiveCase can combine multimedia, questions, and AI character conversations in a case. Define the reasoning you want to observe, then use the responses as a starting point for discussion.

FAQ

What is an AI-resistant assignment?

An AI-resistant assignment reduces reliance on a finished answer as the sole evidence of learning. It can combine intermediate work, source evidence, and discussion. It does not guarantee that AI cannot assist.

How do I grade oral defenses fairly?

Use a structured rubric with clear criteria for comprehension, justification of choices, and ability to respond to follow-up questions. Record defenses so you can review them. For consistency, prepare a bank of follow-up questions but adapt them to each student's submitted work.

Can process-based assessments work in large classes?

Yes. Use templated reflection logs with specific prompts, and grade with a lightweight rubric. Peer review of process journals can reduce grading load while still making thinking visible. A "spot-check" approach, grading only 2–3 process entries deeply per student per semester, also works.

What is the difference between AI-resistant and AI-detection approaches?

Detection attempts to infer how text was produced. Assessment design focuses on the evidence needed to judge learning. Neither a detector result nor an assessment format should be treated as proof on its own.

How many AI-resistant assignments do I need to redesign?

Start with one assignment and evaluate the results. The number of changes needed depends on your course, learning objectives, and assessment policy.

Do AI-resistant assignments work for STEM courses?

Yes. Lab demonstrations, data analysis, and portfolios can reveal how students apply methods. Design the tasks around the intended skill, with clear resource rules and appropriate accommodations.