Students are using AI tools in their academic work, and that’s not going away. What’s worth paying attention to is what happens when students use those tools uncritically — accepting AI-generated summaries, explanations, and responses without checking them against the actual source material. AI outputs can be fluent, confident, and wrong in ways that aren’t immediately obvious, which makes verification a skill worth teaching directly rather than assuming students will pick it up on their own.
This activity gives instructors a concrete way to do that. It doesn’t require a new tool, a complicated setup, or a dedicated unit on AI. It fits into existing coursework, takes roughly one class session, and leaves students with something transferable: the habit of going back to the source.
The Setup
Start by selecting or generating a short passage that contains intentional errors — fabricated citations, misleading claims, or statements that sound plausible but don’t hold up against the source material. Upload it to your LMS alongside the original reading it’s based on.
With Hypothesis integrated directly into Canvas, Blackboard, D2L, or Moodle, students can annotate both documents without leaving the course environment. Hypothesis LMS Integrations has setup guidance for each platform.
The Activity, Step by Step
Step 1: Students annotate the AI-generated passage as they read. Ask them to highlight anything that seems off — suspicious claims, statements that feel unsupported, language that’s vague where it should be specific. They add comments explaining their reasoning directly in the text, not in a separate reflection document. The goal is to make their thinking visible while they’re still inside the material.
Step 2: Students verify against the source. They go back to the original reading and find where each flagged claim either holds up or falls apart. Annotations from step one stay visible as they cross-reference, so they’re building an argument in context rather than working from memory.
Step 3: Small groups compare findings. Students review each other’s annotations, confirm or challenge what peers identified, and discuss why certain errors were harder to catch than others. Because the annotations are anchored to specific passages, the conversation stays grounded in the text rather than drifting into generalities.
Step 4: Class debrief. Reveal the intentional errors, discuss what students caught and what slipped through, and spend time on the errors that were hardest to detect. Those are usually the most instructive — the places where AI output sounded authoritative enough that critical reading required real effort.
Why This Works
The activity works because it doesn’t just tell students that AI can be wrong. It puts them in a position where they have to prove it, using evidence from the source material, in front of their peers. That’s a different kind of learning than a lecture on AI limitations.
It also scales. Group-based annotation means instructors aren’t reading every individual response in isolation — they can see patterns across the class, identify which errors generated the most discussion, and use that as a starting point for the debrief. More than 300 colleges and universities use Hypothesis to support this kind of activity directly inside their LMS. You can explore how institutions are applying it at Hypothesis Education.
Unlike detection-based approaches, this one doesn’t become obsolete as AI tools improve. The skill being taught — returning to the source, evaluating evidence, explaining your reasoning — applies regardless of how good the AI gets. Students aren’t learning to spot AI. They’re learning to read critically, which is the same thing they needed before AI existed.
If you want a ready-to-use version of this activity with structured prompts and faculty guidance, the AI Literacy Course Pack has everything you need to get started. And if you’re new to Hypothesis, getting started takes less than a class period to set up.
Frequently Asked Questions
How long does this activity take?
It can be completed in a single class session or adapted for a longer asynchronous assignment.
Does this work in online courses?
Yes. Students can complete annotation and discussion asynchronously within the LMS, with no need for a synchronous session.
Can this scale to large classes?
Yes. Group-based annotation lets instructors manage participation and encourage peer learning without reviewing every response individually.
Do students need prior experience with AI?
No. The activity introduces AI evaluation and verification as part of the process itself — no background required.
How does this work inside the LMS?
Hypothesis integrates directly with Canvas, Blackboard, D2L, and Moodle, so students annotate and collaborate without leaving their course environment.
Related blogs
How to Design Reading Assignments That Work in the Age of AI — How to redesign assignments so students engage directly with texts instead of relying on AI summaries.
Combating AI-Generated Essays with Collaborative Annotation Assignments — How annotation-based assignments help students analyze and verify AI-generated content.
Teaching the Process, Not the Product — Why focusing on visible thinking helps students build deeper understanding in AI-driven classrooms.