Randomized trial tests three AI prompts for citation accuracy, suggesting effective strategies.
Key Points
The study aims to evaluate the effectiveness of three AI prompts for verifying citation accuracy and analyze their quality based on the CLEAR framework.
Tested 45 altered citations from 15 source citations across three prompt strategies: Verify, Correct, and Fit.
Analyzed results using quantitative chi-square analysis and the CLEAR framework.
Generated a total of 135 prompts for Microsoft CoPilot.
The Fit prompt provided the highest accuracy in citation correctness.
The Correct prompt generated incorrect citations due to 'corrective hallucinations'.
Only the Fit prompt met all criteria of the CLEAR framework: concise, logical, explicit, adaptive, and reflective.