AI-Resilient Assessment: 7 Strategies

I open my faculty workshops with a small test. Three paragraphs on the same topic — the impact of social media on student well-being. Two were written by humans. One by AI. I ask the room to vote.

Here they are. Take a moment before you scroll.

A. Social media's impact on student well-being is honestly hard to pin down. Some of my students seem fine scrolling TikTok between classes, while others tell me it keeps them up at night comparing themselves to influencers.

B. Research indicates that social media platforms exert a multifaceted influence on student well-being, encompassing both positive and negative dimensions. Studies suggest that while social connectivity is enhanced, excessive usage correlates with increased anxiety and diminished academic performance.

C. I deleted Instagram for a month last semester and honestly? I slept better, but I also missed three study group invites and a deadline reminder. Social media is a mess — useful and toxic at the same time, and I'm not sure anyone has figured out the right balance yet.

It's B.

The room splits every time. Experienced academics, people who have marked thousands of scripts, disagree with each other. And that is the whole point:

If trained professors can't reliably detect AI text, how can an algorithm?

This is where most institutional conversations go wrong. They treat AI as a detection problem and go shopping for tools. The detectors have high false-positive rates — and they fall hardest on non-native English speakers, which in a Gulf or South Asian classroom means the students least able to absorb an accusation are the ones most likely to receive one.

AI hasn't just created a cheating problem. It has exposed a design problem.

What AI is actually telling us about our assessments

Here is the uncomfortable observation. If an AI can produce a first-class answer to your assignment in three minutes, the assignment was probably measuring recall and summary — not thinking.

That was always a weak proxy for learning. We used it because it scaled. AI has removed the option.

UNESCO's 2025 guidance pushes assessment away from measuring rote knowledge toward three areas:

Critical thinking — can students deconstruct AI outputs, question assumptions, spot errors, and adapt when evidence contradicts their position?
Creative problem-solving — can students integrate personal experience, local context and original synthesis that AI simply cannot replicate?
Ethical reasoning — can students use AI tools effectively and responsibly? This is a genuinely new learning objective that didn't exist five years ago.

If assessments test recall and summary, AI aces them. If they test these three areas, students have to show up.

The values layer: EAT

Before the tactics, the values. Rutherford et al. (2025) frame AI-enhanced assessment around three commitments I use as a design filter:

Equity. AI tools must not widen existing gaps. Account for unequal access — never require a specific paid tool. If AI is required, the institution provides access.

Agency. Students make informed choices about how and when to use AI. That means teaching when it helps, when it hurts, and how to use it to deepen rather than replace thinking.

Transparency. Mutual disclosure. Faculty disclose when they use AI. Students disclose when they use AI. Everyone knows who did what. This is the move from gotcha policing to professional accountability, and it changes the temperature of the whole conversation.

The 7 strategies

These are the tactics I take faculty through. Most assessments can absorb three or four without a redesign from scratch.

1. Contextualised assessments. Require students to connect concepts to their unique experiences, communities or observations. AI wasn't there. It cannot interview the shopkeeper on your campus road.

2. Process-based assessment. Evaluate the journey, not just the destination. Draft submissions, revision logs, reflection journals. The final artefact is easy to generate; a credible trail of revisions is not.

3. Oral assessments. Vivas, presentations with live Q&A, Socratic seminars. AI cannot speak for a student in real time. Ten minutes of questions reveals more than ten pages of prose.

4. Collaborative projects. Team projects with individual accountability — each member demonstrates their specific contribution and their understanding of the whole.

5. AI-integrated assessments. Don't ban AI — require it. "Use AI for a first draft, then critically revise. Submit both, with annotations explaining every change."

6. Authentic problem-solving. Real-world, messy problems requiring current data, field observations or community engagement. Local knowledge AI doesn't have.

7. Portfolio assessment. Cumulative portfolios showing growth over time, including metacognitive reflection on learning strategies — AI use among them.

Strategy 5 is the one I'd argue hardest for. It inverts the problem: the assessment now evaluates critical improvement rather than original generation. A student who can take a competent AI draft and explain precisely why each change makes it better has demonstrated something far more valuable than a student who produced a decent essay alone. And it is honest about the world those students are graduating into.

What this looks like in practice

Take a real assignment.

Before:

"Write a 2000-word essay analysing the impact of social media on youth mental health in the Gulf region. Cite at least 8 peer-reviewed sources."

AI does this perfectly, in about three minutes.

After:

"Design a mixed-methods research proposal investigating social media's impact on student well-being at your university:

Interview 5 students on campus (contextualised)

Submit two drafts with annotated revisions (process-based)

Use AI for the literature review, then critically revise (AI-integrated)

Present in a 10-minute viva with Q&A (oral)

Include an AI Collaboration Statement (transparency)

Reflect on how your understanding changed (portfolio element)"

Same learning objectives. Same topic. Six of the seven strategies applied.

AI cannot do this assignment. But AI genuinely helps with parts of it — and the student learns more.

That's the goal. Not AI-proof. AI-enhanced.

Where to start on Monday

Don't redesign your curriculum. Take the single assessment that carries the most marks in one module and ask three questions:

Is it equitable? (EAT)
Does it require what AI cannot do? (contextualised, process, oral)
Which of the 7 strategies could it absorb without a rewrite?

If the honest answer to question 2 is "not much", you've found your starting point. Adding a viva, or requiring annotated drafts, is often a single paragraph of change to an assignment brief — and it does more for academic integrity than any detection tool on the market.

The institutions handling this well aren't the ones with the strictest policies. They're the ones that stopped asking "how do we catch students using AI?" and started asking "what are we actually trying to measure, and does this task measure it?"

This material comes from AI-Based Methods for Effective Teaching and Evaluation, a three-day faculty development programme I delivered for faculty at the University of Technology and Applied Sciences (UTAS), Muscat, Oman. The full workshop covers prompt frameworks for educators, AI-assisted grading, AI ethics, and building a departmental action plan.

If you're responsible for faculty development at a university or college and want to run this programme with your staff, get in touch.

References

UNESCO (2025), guidance on assessment in the age of AI
Rutherford et al. (2025), AI-Enhanced Assessment Values — the EAT framework