Related Experiment Videos
Cross-tool evaluation of artificial intelligence-drafted informed consent documents: A 3-level study
Shama U Rao1, Akuma Ifeanyichukwu1, Uma Kulkarni1
1Centre for Ethics, Yenepoya (Deemed to be University), Mangalore, Karnataka, India.
Background:
Informed Consent Documents (ICDs) comprising both the Participant Information Sheet and the Consent Form represent a cornerstone of ethical research involving human participants. The advent of generative Artificial Intelligence (AI) has introduced novel tools capable of drafting ethically oriented documents with high linguistic fluency. However, empirical evidence comparing their ethical robustness, completeness, and readability against institutional standards remains limited.
Aims And Objective:
This study aimed to compare the performance of five generative AI tools ChatGPT, Gemini, Copilot, Meta AI, and Perplexity AI in drafting ICDs across six simulated research scenarios, representing survey, observational, and interventional study designs.
Materials And Methods:
An experimental, cross-sectional, cross-tool comparative design was adopted. Each AI tool generated six ICDs (N = 30), which were evaluated by three trained assessors (blinded workflow) using a validated 4-point rubric (1 = poorly addressed; 4 = strongly addressed) covering completeness and ethical robustness. Readability was assessed using the Flesch-Kincaid Grade Level (FKGL) index, and inter-rater reliability was determined using Fleiss' Kappa.
Results:
Mean ethical robustness scores ranged from 2.8 to 3.1 and completeness from 2.3 to 2.7, indicating moderate adequacy. Meta AI demonstrated the highest ethical robustness (mean = 3.1), while Gemini achieved the highest completeness (mean = 2.7). ChatGPT showed balanced overall performance (mean = 2.9). Readability indices (FKGL 9.3-10.6) indicated that most AI-generated ICDs exceeded the plain-language level recommended for informed consent. Inter-rater reliability was low (κ = -0.063, p < .001), reflecting variability in ethical judgment.
Conclusion:
Generative AI tools can emulate the structural and linguistic characteristics of ICD templates with moderate ethical and procedural adequacy. However, they frequently omit context-specific ethical details, necessitating human oversight to ensure compliance and participant protection. A hybrid human-AI co-creation model is recommended for ethically robust and efficient ICD development.
