A hybrid activity combines stages where students use AI on purpose with stages where the learning must visibly come from them. Done well, it teaches students to use AI with judgement and gives you confidence in what they’ve learned. Done poorly, it becomes a confusing task with an AI step bolted on. These principles will help you land on the right side of that line.
1. Start from the learning outcome, then choose each stage’s lane
Don’t start with “where can I fit AI in?” Start with what students must be able to do by the end. Then, for each stage, ask whether AI would deepen that learning or replace it. If AI would do the thinking the outcome is about, that stage belongs in the AI-resistant lane. If AI would free students to focus on higher-order work, it belongs in the AI-integrated lane.
This is constructive alignment applied to a new tool. The outcome decides the lanes, not the technology.
2. Make the resistant stage check the integrated one
The most common design flaw is two unrelated tasks sitting side by side: an AI essay, then an unrelated quiz. A strong hybrid activity links them. The resistant stage should ask students to explain, critique, apply or defend what they produced with AI.
A simple test: if a student skipped the learning in the AI stage, would the resistant stage reveal it? If not, tighten the link.
3. Tell students exactly what’s allowed at every stage
Ambiguity is the main cause of accidental misconduct. Label every stage in plain language, such as “AI expected”, “AI permitted for editing only” or “No AI”. Add one sentence explaining why. A shared framework like the AI Assessment Scale can help keep labels consistent across a unit or program.
Students follow rules they understand far more reliably than rules they’re merely given.
4. Mark the judgement, not just the product
If your rubric only grades the final document, students will rationally optimise the document, and AI is very good at documents. Give real weight to the evidence of thinking: the change log, the critique, the oral explanation or the debrief. These are the parts that show learning, so they deserve the marks.
5. Design for your delivery mode and class size
A five-minute oral check is easy with 20 students. With 200, it’s around 17 hours of marking time. Be realistic before you commit:
- Large cohorts: use group defences, short in-tutorial checks, or randomly sample a portion of students for orals.
- Fully online units: favour timed drafting stages, recorded explanations and live video-call checks over in-room tasks.
- Placement-based units: anchor resistant stages in what students observe and do on placement.
6. Build in equity, access and privacy
Only require AI tools your institution has approved and that every student can access free of charge. Never ask students to enter identifiable personal, patient or client information into an AI tool. In health and other professional disciplines, model de-identification as part of the task itself, because it’s a professional skill in its own right.
7. Teach verification before you assess it
Students won’t spot AI errors just because you asked them to. Model it first: put an AI response on screen and critique it together, thinking aloud about what you check and why. Name the habits explicitly, such as checking sources exist, testing claims against guidelines, and asking what’s missing. Then let students practise before it counts.
8. Review the design every teaching period
AI capability moves quickly, and a stage that resisted AI last year may not this year. After each run, ask which stages still did their job, where students found shortcuts, and what the resistant checkpoints revealed. Small, regular updates keep an activity sound far better than a major redesign every few years.
Common mistakes to avoid
- Calling a task “AI-proof”. Nothing is, and the claim undermines trust when students prove otherwise.
- Banning AI in a stage where it can’t be enforced. Unenforceable rules penalise only the honest students.
- Adding AI for novelty. If it doesn’t serve the outcome, leave it out.
- Hiding the purpose. Students engage with checkpoints in good faith when they understand why they exist.
A quick pre-flight checklist
Before you run a hybrid activity, check that:
- Every stage is labelled with its AI rule and a one-line reason.
- Each resistant stage directly checks something from an AI-integrated stage.
- The rubric rewards judgement and process, not just the final product.
- The workload is realistic for your cohort size and delivery mode.
- Every student can access the required tools, and no private data goes into them.
- You’ve modelled the verification skills you’re about to assess.
Our Hybrid Activity Designer builds these staged activities for you, with each stage labelled and explained.