Can you trust AI? Students generate answers with AI, fact-check them, and uncover errors, building critical thinking and verification skills.
- Group activity
- Gamified Verification task
- In class
- Generating and crosschecking AI outputs
- All disciplines (Research skills focus)
- Beginner
- approx. 60 minutes / 1 session
- Flexible
- GenAI tool
- External sources for verification
Short description
A gamified collaborative activity that trains students in critical evaluation and fact-checking. Students are divided into opposing teams: “The Generators” and “The Verifiers.” One team uses a Large Language Model (LLM) to answer specific inquiries, while the opposing team critically fact-checks the output using reliable human sources. Roles are then swapped to ensure all students practice both prompting and verification skills.
Competence Domain of the Didactic Framework: Critical Engagement
At the end of this activity, students can…
- compare AI and human sources by using search strategies to crosscheck information received from GenAI. (FLAIR Didactic Framework: LO 8)
- develop evaluation strategies for GenAI outputs. (FLAIR Didactic Framework: LO 6)
- judge the reliability of AI-generated text versus traditional academic sources.
- communicate and interact with GenAI effectively (formulate clear and effective prompts, iterate prompts and refine outputs). (FLAIR Didactic Framework: LO 16)
Instructions
The class is divided into an even number of Team A’s and Team B’s. Each team is then assigned a specific and verifiable topic provided by the instructor (e.g. historical dates, scientific facts, obscure literature summaries).
Team A (Generators) uses a GenAI tool to produce an answer or summary for the assigned topic. Team B (Verifiers) reviews the AI-generated output and checks its accuracy using reliable sources such as search engines, academic databases, or books. The students should then identify inaccuracies, missing information, or potential hallucinations.
Teams switch roles and repeat the process with a new topic. This ensures that all students practice both generating and verifying AI outputs.
In a final plenary discussion, the class reflects on the findings and the challenges of fact-checking AI-generated content. Guiding questions may include: “How often was the AI wrong/inaccurate?”, “Which types of claims were hardest to verify?”, and “How did the AI appear convincing even when incorrect?”. As an optional extension, students may summarize their findings in a simple verification checklist or guideline for future use.
Assessment
Assessment focuses on the quality of the verification process. Students are evaluated based on their ability to identify inaccuracies or hallucinations in AI-generated outputs and to support their findings with appropriate and reliable human sources. The assessment may be based on the submission of a “fact-check log” documenting the identified issues and corresponding sources as well as an optional verification checklist or guideline.
Possible challenges
- AI-generated outputs may be fully correct for simple or well-known topics, reducing the potential learning impact.
- Students may focus on finding errors only, rather than critically evaluating the overall quality of the output.
How to address them
- The instructor should prepare discipline-specific “trick” topics or obscure queries where hallucinations are more likely (e.g. asking for citations of non-existent papers or details on very recent local events).
- Encourage students to also evaluate completeness, source quality, and reasoning, not only correctness.
Hallucination Hunt: Instead of generating outputs themselves by only providing topics, students are given pre-generated AI texts that intentionally contain errors or inaccuracies. Working in teams, they identify and document as many issues as possible within a defined time period.
Using this resource
This resource is licensed under Creative Commons BY-NC-SA 4.0 license. Suggested citation: Flair Collaboration. (2025). FLAIR Toolkit. Teaching GenAI Competencies.
