← Back to Know
Knowledge check

Something is wrong with this AI output

You will read a student's conversation with ChatGPT and spot a problem: the AI produced specific statistics that may be entirely made up. Answer three questions about why this happened and what a better approach would have been.

1
Read the chatFollow the conversation and notice what the AI claims.
2
Spot the problemWhat looks suspicious about the response?
3
Answer the questionsApply what you know about how AI tools work.
The conversation
Student scenario
Elif, 1st-year Psychology
Elif is writing her first literature review on working memory and academic stress. Her lecturer mentioned a Baddeley & Hitch follow-up study in class. She wants to include its key findings and turns to ChatGPT to quickly retrieve the statistics.
G
ChatGPT GPT-4o
Simulated interface
What were the main findings and statistics in the Baddeley & Hitch 2022 follow-up study on working memory and exam stress in university students?
G
The Baddeley & Hitch (2022) follow-up study examined working memory capacity under exam-related stress in 412 undergraduate students at the University of York. Key findings:
  • Students under high exam stress showed a 31% reduction in overall working memory capacity compared to a low-stress baseline
  • The phonological loop was the most affected component (38% reduction); the visuospatial sketchpad showed a more modest 19% decline
  • The effect was partially mediated by cortisol levels (β = 0.43, p < .001)
  • Students who practised mindfulness showed significantly smaller working memory decrements (17% vs 31%)

Baddeley & Hitch concluded that exam stress impairs working memory primarily through the phonological loop, and recommended mindfulness-based interventions as targeted student support.
Message ChatGPT…
💡
What to notice: The response looks authoritative — it gives a sample size, specific percentages, and even a beta coefficient. But none of these figures can be verified without locating the actual paper. Confident-sounding output is not the same as accurate output.
–
/ 3
–
–
Back to Know Continue to Use ›

The chat interaction shown is a simulated example for learning purposes. The study referenced does not correspond to verified published findings.