AI-Resistant Course Design

Assignments Students Actually Have to Do

September 25 PD Day · Breakout led by Bill

How to Use This Page

A research-grounded framework for designing learning tasks that develop and demonstrate authentic understanding, grades 7–12. Based on peer-reviewed research in assessment design and academic integrity.

The Research Reality

AI Didn’t Create the Cheating Problem

~70%

of students reported cheating before generative AI tools existed (Stanford University, cited in ASCD, 2025)

“Cheating is not caused by technology — it is caused by culture.”

Students use AI when assignments feel irrelevant, when culture prioritizes grades over learning, and when anxiety is high.

Source: ASCD (2025) · Stanford Research

Why Detection Fails

AI Detectors Are Not a Reliable Solution

20%+

False-positive rate. Real student work flagged as AI-generated, with disproportionate impact on English Language Learners.

Equity harm confirmed

Non-native English writers and students from under-resourced backgrounds are flagged at higher rates.

Arms race dynamic

Each new detection method is quickly circumvented; humanizing tools are freely available online.

Better question: Instead of “Did AI write this?” ask “Does this submission show that the student genuinely learned?”

Source: Springer Nature (2025) · PMC AI-Resilient Assessment Framework (2025)

The Right Question

Five Root Causes of Academic Dishonesty

1

Irrelevant material

Design tasks connected to students’ real lives and futures
2

Weak integrity norms

Co-create expectations with students
3

Grade-focused culture

Shift emphasis to learning over performance metrics
4

High assessment anxiety

Use low-stakes checkpoints and mastery-based approaches
5

Student passivity

Offer choice in how students demonstrate understanding

Each root cause points to an assessment design fix — not a surveillance solution.

Source: ASCD (2025) · eSchool News (2025)

Research Framework

The AAAE Framework

Against · Avoid · Adopt · Explore

Against · No AI use

Traditional exams, oral presentations, in-class writing, paper-based tests.
Best for: skills requiring independent competency

Avoid · AI struggles here

Individual portfolios, personal reflection, self-awareness tasks, real experience required.
Best for: higher-order thinking, personal connection

Adopt · AI as tool

AI for brainstorming/drafting; student critiques, evaluates, and improves the output.
Best for: critical evaluation skills

Explore · AI as partner

Human-AI collaboration, co-design, creative exploration with full disclosure.
Best for: innovation, critical AI literacy

Source: MDPI Education Sciences, Rasheed et al. (2025), peer reviewed

Bloom’s Taxonomy & AI Vulnerability

AI Excels at the Bottom. Design for the Top.

Bloom’s levelAI capability
Remember & UnderstandExcellent
ApplyStrong
AnalyzeModerate
EvaluateWeak
Create (with situated judgment)Very weak

“AI dependency is highest in tasks requiring lower-order thinking. Higher-order tasks requiring critical reflection, ethical reasoning, and personal experience show significantly lower AI substitution rates.” (Springer Nature, 2025)

Practical implication: Redesign prompts to require students to apply, evaluate, and create in relation to their specific class context, experience, or community.

Source: Springer Nature, Bloom’s Taxonomy & Critical Thinking (2025) · Frontiers in Education (2024)

The Four-Pillar Framework

Design so that learning evidence cannot be faked. Each pillar is expanded below.

1

Authentic Task Design

Require situated judgment that generic AI cannot replicate — specific to this class, this community, this student
2

Process Documentation

Require staged submissions showing intellectual development — drafts, reflection logs, revision history
3

Verbal Demonstration

Short conferences, presentations, or Q&A where students explain their thinking in real time
4

Transparent AI Policy

Graduated expectations tied to learning goals — not blanket bans, but clear declared-use requirements

Key insight: No single pillar is sufficient. Circumventing one pillar is insufficient if the others still generate valid evidence of learning.

Source: PMC, AI-Resilient Assessment Framework, four-pillar model (2025), peer reviewed

The Four Strategies

1 · Authentic Task Design: make the task impossible to outsource
  • Anchor to class-specific content. Reference a text, lab result, discussion, or data set from your class that AI hasn’t seen.
  • Require personal experience or local context. “How does this apply to our school/neighborhood/community?” AI cannot answer for your students.
  • Add real audience or purpose. Write a letter to a real organization, design something for a classmate’s stated need, solve a problem you observed.
  • Use recent, specific data. Assign tasks using current local or school data, a recent event, or information from your unit that changes each semester.

Before & after

AI-vulnerable: “Explain the causes of World War I.”

AI-resistant redesign:

Using the primary sources from our Thursday class and your own family’s history in [era], explain which cause of WWI you find most compelling and why — and identify one argument from class discussion you disagree with.

Source: Frontiers in Education (2024) · PMC AI-Resilient Assessment (2025)

2 · Process Documentation: assess the journey, not just the destination
  • Require a minimum of 2 checkpoints with a dated draft, showing visible change and growth from one version to the next.
  • Assign a revision log. Students write 3–5 sentences explaining what changed between drafts and why. This can’t be faked without doing the work.
  • Weight process in the rubric (20–30%). Visible intellectual progression matters.
  • Use brief in-class brainstorm time. Even 5 minutes of documented planning in class creates a paper trail of authentic thought.

Sample checkpoint sequence

Checkpoint
D1In-class brainstorm + outline (Day 2)
D2Rough draft + 3-sentence revision plan (Day 5)
D3Revised draft + revision log explaining changes (Day 8)
FinalFinal submission + self-assessment (Day 10)

Process-based assessment was the most commonly adopted strategy in faculty training with 333 educators globally (Frontiers in Education, 2024).

Source: PMC Four-Pillar Framework (2025) · Frontiers in Education, Faculty Workshops (2024)

3 · Verbal Demonstration: students explain their own thinking out loud

A. 2-Minute Conference (1-on-1)

Ask 2–3 questions directly from their submitted work: “Walk me through your reasoning here.” “What did you originally think before you revised?” “What’s one thing you’d change now?”

B. Exit Ticket Explanation

When a take-home assignment is collected, students verbally state (or write in 2 minutes) their main argument and one challenge they encountered.

C. Small-Group Teach-Back

Students explain their work to 2 peers who ask questions. Peers evaluate whether the explanation matches the written submission.

Formal oral defense protocol (8–12 min)

TimeActivity
2–3 minStudent summarizes their key conclusions in their own words
4–6 minTeacher asks student-specific questions drawn from their process logs or draft choices
2–3 minStudent applies their reasoning to a novel scenario or extension question

Workload note: Formal oral defenses require ~17 hours per 100 students. Random sampling (20–30% of class) reduces time substantially while maintaining integrity.

Source: PMC AI-Resilient Assessment Framework (2025)

4 · Transparent AI Policy: “declare, don’t hide”

Level 1 · Prohibited

No AI use. Student demonstrates independent competency.
Disclosure: none.
Best for: in-class writing, exams, oral assessments

Level 2 · Guided Use

AI for delimited tasks (brainstorm, check grammar). Student critiques and improves AI output.
Required: AI-use log + prompts submitted with final work.
Best for: research papers, project proposals

Level 3 · Permitted

Full AI access. Assessment focuses on higher-order evaluation of AI-generated content.
Required: use acknowledged; student demonstrates critical distance.
Best for: AI literacy units, media analysis

Each assignment must state its level explicitly. Students should never have to guess what AI use is expected or allowed. Have students help define what “citing AI” looks like in your discipline. See also Example Language for AI Use.

Source: PMC AI-Resilient Framework (2025) · MDPI Education Sciences AAAE (2025)

Assignment Vulnerability Audit

Before you assign it, ask these six questions.

Could a competent LLM pass this without engaging?

If yes → add situated context, personal experience anchor, or class-specific data

Am I assessing only the final product?

If yes → add at least one draft checkpoint and a revision log

Is the task generic enough to use any student’s prompt?

If yes → specify a real audience, purpose, or constraint unique to this assignment

Does my rubric overweight polish over reasoning?

If yes → reweight: reasoning quality 25–35%, process engagement 20–30%

Is my AI policy stated explicitly on the assignment?

If no → add Level 1/2/3 designation before distributing

Will I be able to have a conversation with a student about this?

If yes → the assignment generates valid evidence of learning regardless of AI use

Source: PMC AI-Resilient Framework, Assessment Vulnerability Audit (2025)

Redesign Any Assignment in Five Steps

1

Run the vulnerability audit

Identify which risks apply. A single red flag = enough reason to revise.
2

Assign an AI policy level (1, 2, or 3)

Add one sentence to the top of the assignment:

This is a Level [X] AI task. [Specific expectation.]

3

Add ONE situated anchor

A reference to class content, a personal experience requirement, or local/current data that can’t be googled.
4

Add ONE process checkpoint

Even a rough outline submitted two days early creates evidence of genuine engagement.
5

Plan a brief verbal check

Will ALL students explain, or will you randomly sample 20–30%? Either is valid. Post the expectation up front.

Time cost of a full redesign

TaskTime
Vulnerability audit5 min
Add policy level label2 min
Rewrite prompt with anchor10 min
Add checkpoint to calendar3 min
Update rubric weights5 min
Total~25 min

Source: Adapted from PMC AI-Resilient Framework (2025) · Frontiers in Education (2024)

Test Before You Assign

Run your assignment through AI before your students do.

1

Paste your prompt into ChatGPT, Claude, or Gemini.

Use the exact wording you would give students. No modifications.
2

Would a student submitting this output receive a passing grade?

If yes, the assignment needs revision before it goes to students.
3

What is missing from the AI response that only a student could provide?

That gap is where your revision should focus.

What to add when AI passes

This test takes less than 5 minutes and tells you more about your assignment’s AI vulnerability than any detection tool ever will.

Source: Imagine Learning AI-Resistant Assessment Planning Guide (2025)

The FAIR Method

Four lenses for designing any resistant assessment.

F · Feedback

Students must respond to teacher or peer feedback in writing, explaining what they changed and why.

Read my feedback on your draft. Write 3 sentences explaining whether you agree and what you plan to change.

A · Assessment

Choose modes that require presence: in-class explanation, live demonstration, oral defense, a video log with the student on camera.

Record a 90-second video explaining your design decision. You must appear on camera.

I · Integrity

Set a Level 1, 2, or 3 AI policy on every major task. Integrity is built through transparency, not surveillance.

This is a Level 1 task. AI tools may not be used to generate any part of the written response.

R · Reflection

Every major assignment ends with a short structured reflection. Authentic reflection is among the most AI-resistant artifacts students can produce.

In 4-6 sentences: What was hardest? What changed in your thinking? What question do you still have?

Source: Imagine Learning AI-Resistant Assessment Planning Guide (2025)

Before and After: Three Redesigns

Argumentative essay → Process portfolio

Before: “Write a 5-paragraph essay arguing for or against social media regulation. Use 3 sources.” Vulnerability: high.
After: Same topic + annotated source log + two dated drafts + revision log + 5-minute in-class writing sample on a related prompt. Vulnerability: low. Added ~25 min teacher setup.

Research paper → Authentic investigation

Before: “Research the causes of the 2008 financial crisis. Write a 6-page paper with citations.” Vulnerability: very high.
After: Same event + interview one adult who lived through it + compare their account to 2 academic sources + explain the discrepancies in your own words. Vulnerability: very low. Added ~15 min teacher setup.

Math problem set → Explained solutions

Before: “Complete problems 1-20. Show your work.” Vulnerability: high; AI solves and shows steps.
After: Complete 10 problems + choose 3 you found hardest + explain each step in plain English + identify one step where you could have made a different choice and why you did not. Vulnerability: low. Added ~10 min teacher setup.

Source: Pertama Partners Five-Step Assessment Design SOP (2025)

Classroom and Homework Formats

Eleven assignment formats that build in resistance

Personal Narrative Analysis

Connect content to the student’s own experience or family history.

Annotated Source Log

Students find, annotate, and explain why each source was chosen.

Multimedia Portfolio

Podcasts, video essays, slide decks with recorded narration.

Primary Source Interrogation

Documents from your class only, defended in writing AND in discussion.

Problem-Based Case Study

Local/school-specific data or a real community problem.

Socratic Seminar + Write-Up

Written reflection must reference specific things said.

Lab / Field Observation Log

Timestamped real-time entries; analysis from genuinely collected data.

Comparative Revision Essay

Two drafts + 200-word explanation of every change.

Peer Teaching Segment

5-minute explanation to classmates who then ask questions.

Guided AI Critique

Students identify errors, bias, and omissions in an AI response using course content.

Portfolio Reflection

Students curate semester work and write about growth and what changed in their thinking.
In-classroom strategies

Low-stakes frequent checks

Brief in-class writes (5–8 min), think-pair-share, and whiteboard problems. Also the best formative data you’ll collect.

In-class brainstorm before take-home

5 minutes of initial thinking in class becomes a required reference in the final work.

Discussion-anchored assignments

Require the final product to reference a specific claim made by a named classmate. AI cannot know what Jordan said on Tuesday.

Quick-implementation habits

  • Begin each class with a 3-minute handwritten entry connecting to the previous lesson — no phones
  • Ticket out the door: name one thing you changed your mind about today
  • Collect brief in-class drafts of take-home assignments at intervals — grade effort, not quality
  • Ask students to present one surprising thing they learned while working on an assignment
  • Reserve 5 min for 3–4 students to read one sentence from their work and explain their choice
Out-of-class strategies

Staged multi-night assignments

Night 1: outline. Night 2: draft Sec. 1. Night 3: revise Sec. 1, draft Sec. 2. Final night: revise and reflect.

Interview or observation tasks

Interview a family member, neighbor, or community member and analyze it against course content.

Photo or artifact documentation

Document something in the student’s environment and analyze it using course concepts.

Reading response with quoted annotation

Two direct quotes and a response that could ONLY work with those quotes.

Deliberate position-switch

Argue the position opposite your own stated view, referencing class materials.

Curiosity log

Each night, one question the content raised and why it matters personally.

Source: Frontiers in Education (2024) · MDPI (2025) · PMC (2025) · ASCD (2025)

Common Mistakes to Avoid

Making tasks harder without serving learning

Longer papers, more sources, or more complexity do not reduce AI viability. Ask: does this addition require real understanding, or just more typing?

Assuming in-class work is AI-free

In-class tasks still need on-the-spot reasoning: a question posed that day, a document distributed in class, or a discussion that just happened.

Relying on citation format as proof of work

AI produces correctly formatted citations. Source selection, annotation, and explanation of relevance require genuine engagement.

Redesigning without a clear learning objective

If a verbal defense is added but students never practiced oral explanation, it measures anxiety rather than understanding.

Changing expectations without notice

Post checkpoints, verbal components, and AI policy levels with the assignment, not at submission.

The productive framing: The goal is not to catch students — it is to design tasks where catching isn’t necessary.

Source: Pertama Partners (2025) · ASCD (2025)

Sources

Peer-reviewed and practitioner sources
  • PMC / Springer Nature (2025) — “Designing AI-Resilient Assessment in Higher Education: A Four-Pillar Conceptual Framework”
  • MDPI Education Sciences (2025) — Rasheed et al., “Redesigning Assessments for AI-Enhanced Learning”
  • Frontiers in Education (2024) — “AI-Resistant Assessments in Higher Education: Practical Insights from Faculty Training Workshops”
  • Springer Nature (2025) — “Cultivating Independent Thinkers: AI, Bloom’s Taxonomy and Critical Thinking in Assessment Pedagogy”
  • ASCD (2025) — “The Real Way to Stop Cheating in an AI World”
  • Imagine Learning (2025) — AI-Resistant Assessment Planning Guide
  • Pertama Partners (2025) — Five-Step Assessment Design SOP

The goal is not to prevent AI use. It is to design learning tasks where genuine understanding is the only path to a meaningful submission.