Stop trying to AI-proof your assignments

Key points:

• Educators should design learning experiences in which technology can support the work, but students still have to do that work
• Students using AI: It’s not that scary and shouldn’t be banned
• Moving from AI detection to verification-first supports stronger academic integrity
• For more news on AI in assignments, visit eCN’s AI in Education hub

A few years ago, many of us in higher education were asking a fairly straightforward question about AI: How do we stop students from using it to do their work?

I was asking a different question.

In 2024, I wrote that students using AI was “not that scary” and argued that banning the technology was unlikely to be the answer. Instead, I suggested that educators needed clearer expectations for appropriate AI use and pointed to the AI Assessment Scale as one promising framework.

There was one important limitation to that argument at the time: I had not yet started using the scale consistently in my own college courses.

Now I have.

I teach undergraduate students preparing for careers in education and human services, including many adult learners. Over the past several semesters, I have become much more explicit about when students may use AI, what they may use it for, and what parts of their work need to remain distinctly their own. In many assignments, I use Level 2 of the AI Assessment Scale: Students may use AI for limited support such as brainstorming, organizing ideas, checking grammar, or improving clarity, but their ideas, experiences, analysis, and final voice must remain their own.

Using the scale has helped. But it has also taught me something important:

An AI policy cannot fix an assignment that no longer asks students to do meaningful thinking.

That realization has changed the question I ask when designing learning experiences.

Instead of asking, How can I AI-proof this assignment?, I now ask: What do I still need the student to do?

That question feels increasingly urgent.

A 2026 Pearson and AWS report on AI readiness in higher education found a substantial disconnect between the rapid adoption of AI in the workplace and the experiences students are having in college. Only 30 percent of learners reported that AI was covered extensively or “quite a lot” in their curriculum. The researchers describe a growing environment of “shadow AI,” in which students independently adopt tools without necessarily receiving guidance about transparency, ethics, data security, or appropriate use. 

Trying to eliminate AI from every assignment won’t solve that problem.

Instead, we need learning experiences in which the human contribution remains visible, necessary, and valuable.

Here are five ways I am thinking about that work.

1. Start with the thinking, not the AI

Before deciding whether students should be permitted to use AI, identify what you actually want them to demonstrate.

Is the goal to recall foundational knowledge? Develop an argument? Apply a theory? Reflect on an experience? Exercise professional judgment? Create something original? Evaluate competing possibilities?

Once that is clear, deciding AI’s role becomes easier.

This is where I have found the AI Assessment Scale particularly useful. Rather than having one blanket policy for an entire course, the scale encourages us to think about AI use in relation to the purpose of a particular assessment.

Some learning experiences should require students to work independently. Others might allow AI for brainstorming or editing. Still others might deliberately require students to use AI because evaluating and improving AI-generated work is itself part of the learning.

The Pearson/AWS report reinforces the importance of that distinction. It describes the AI-ready graduate as possessing four interconnected capabilities: functional AI proficiency, strategic intelligence, ethical stewardship, and critical human skills. 

If those are the capabilities our students will need, our assessments should give them opportunities to develop all four.

The question isn’t simply, Can students use AI?

It’s What am I trying to learn about what this student can do?

2. Ask students to bring something AI doesn’t have

One of the simplest changes we can make is designing assignments around inputs that are specific to the learner and the learning experience.

Consider a prompt such as:

Explain how identity can influence the work of a human services professional.

An AI tool can produce a perfectly respectable response to that question in seconds.

Now consider a different version:

Identify a specific experience, interaction, person, place, or challenge that shaped an important part of your identity. Connect that experience to two concepts from this course and explain how it could influence a decision you make as a future human services professional.

AI can still help.

But it doesn’t have the student’s experience.

That distinction matters.

This approach doesn’t have to be limited to personal reflection. Students can work with observations they’ve conducted, data they’ve collected, classroom conversations, field experiences, instructor-created cases, interviews, local problems, previous stages of a project, or feedback they’ve received.

The goal isn’t to devise increasingly elaborate ways of tricking students out of using AI.

It is to design assignments in which students possess something essential to completing the task.

3. Assess the decisions, not just the final product

For a long time, higher education has relied heavily on finished products as evidence of learning: the five-page paper, polished presentation, lesson plan, case analysis, or final project.

Generative AI complicates that assumption because it is remarkably good at producing polished products.

So perhaps we need to see more of the decisions behind them.

Ask students:

• Why did you choose this approach?
• What alternative did you consider?
• What evidence changed your thinking?
• What did you revise after receiving feedback?
• Where did you struggle?
• If you used AI, what did it suggest that you rejected?
• Why?

Those questions shift some of the assessment from production to judgment.

That may be one of the most important changes we can make.

The Pearson/AWS research found that employers see critical evaluation as a significant weakness among recent graduates; the report notes that 58 percent of employers rated graduates’ ability to critically verify AI outputs as their weakest competency. 

If AI can produce an answer, our assessments increasingly need to ask students to demonstrate why an answer is appropriate, what its limitations are, and what they would do with it.

In other words:

AI can generate a product. Students should still have to defend their decisions.

4. Sometimes, require students to use AI

There is another side to this conversation that we cannot ignore.

If we spend all our time figuring out how to prevent students from using AI, we may inadvertently send them into AI-enabled workplaces without knowing how to use it responsibly.

Sometimes the better assignment is to put AI directly in the middle of the learning.

Give students a complicated professional scenario and ask an AI tool to recommend three possible responses.

Then make the students’ work begin:

• Which recommendation is strongest?
• Which would you modify?
• Which would you reject?
• What did the AI overlook?
• What assumptions did it make?
• Does its recommendation align with the research, professional standards, or ethical principles we’ve studied?
• What would you ultimately do?

Now AI isn’t replacing critical thinking. It is providing something to think critically about.

This moves students beyond prompting toward evaluating, verifying, revising, and taking responsibility for AI-supported work.

And that resembles what they are increasingly likely to encounter professionally.

The Pearson/AWS study found that employers aren’t simply asking for technical AI skills. Employers emphasized communication and collaboration, adaptability, and the ability to combine human judgment with AI capabilities. 

Our students therefore need opportunities to practice being not merely AI users, but decision-makers who happen to have AI available to them.

5. Make discussions and projects harder to outsource by making them more human

We also need to reconsider one of the most common assignments in online and asynchronous higher education: the discussion board.

Consider:

What did you learn about diversity and inclusion from this week’s readings? Respond in 300 words and reply to two classmates.

We’ve all seen some version of it.

And generative AI is exceptionally capable of completing it.

But the deeper problem may not be AI.

The deeper problem may be the question.

What if instead we asked:

Identify one claim from this week’s reading that challenged, complicated, or changed your thinking. Explain why. Then find a classmate who interpreted the idea differently. What might explain the difference between your interpretations? After reading their perspective, what would you add to or revise in your original response?

Now the discussion depends upon other people.

It evolves.

It requires students to listen, compare, reconsider, and respond.

Projects can follow the same principle.

Instead of assigning one large product at the end of a module, create a sequence:

Proposal → feedback → revision → application → reflection.

Students leave evidence of their thinking along the way. More importantly, they experience learning as an iterative process rather than a transaction in which they submit a finished artifact for a grade.

That is good assessment design regardless of AI.

AI simply makes the need for it harder to ignore.

Stop trying to catch AI. Start designing for thinking.

There will still be times when students use AI beyond the boundaries we establish. Academic integrity still matters. Clear expectations still matter. And there are learning experiences in which students need to demonstrate that they can perform independently.

But I am increasingly convinced that our primary response cannot be detection.

The Pearson/AWS report makes an important observation about AI governance: Simply establishing rules isn’t enough to create transparent AI collaboration. Governance needs to help students develop the ethical judgment and responsible habits they will eventually carry into professional settings. 

That has been one of the most important lessons from using the AI Assessment Scale in my own teaching.

A few years ago, I was primarily interested in giving students clearer boundaries around AI.

I still believe those boundaries matter.

But I’ve become much more interested in what happens inside those boundaries:

• Are students making decisions?
• Are they applying what they’ve learned?
• Are they connecting ideas?
• Are they creating something?
• Are they reconsidering their thinking?
• Are they exercising judgment?
• Are they able to explain why they made the choices they made?
• And when they use AI, are they controlling the technology—or simply accepting what it gives them?

Those questions lead to a very different kind of assessment design.

Perhaps, then, the goal shouldn’t be to AI-proof our assignments.

Maybe we should human-proof them instead.

Design learning experiences in which technology can support the work, but students still have to bring the curiosity, creativity, experience, judgment, and thinking that make the work worth doing in the first place.