Metacognitive Demands and Strategies While Using Off-The-Shelf AI Conversational Agents for Health Information Seeking

Paper Title

Metacognitive Demands and Strategies While Using Off-The-Shelf AI Conversational Agents for Health Information Seeking

Publication Info

• Topic area: Metacognitive demands in health information seeking via AI conversational agents
• Keywords: AI conversational agents, health information seeking, metacognition, user strategies, think-aloud study, design considerations, digital health literacy, prompt engineering, user experience, health decision-making

Background and Problem

• Problem / challenge: Off-the-shelf AI conversational agents, such as ChatGPT, are increasingly used for health information seeking but impose significant metacognitive demands on users. These systems are not specifically designed for health contexts, leaving users to manage tasks like evaluating information trustworthiness and structuring queries without adequate support.
• Significance: Health-related queries are sensitive, and inaccurate or incomplete information can lead to poor health decisions. Understanding and addressing the metacognitive demands of these systems is critical for improving user experiences and ensuring safer health outcomes.
• Motivation and related work: Prior research has explored health information seeking behaviors and the use of AI conversational agents but has paid limited attention to the metacognitive skills required for effective interaction. Existing systems provide little scaffolding for managing these demands, leaving a gap in understanding how users cope with challenges like prompt formulation, evaluation, and workflow adaptation.

Solution

• Proposed approach: A think-aloud study with 15 participants using a custom-built conversational agent interface powered by the ChatGPT-4o API to identify metacognitive demands and user strategies in health information seeking.
• Novelty:
  1. Identification of metacognitive demands across key stages of interaction: prompt formulation, evaluation, iteration, and workflow adaptation.
  2. Analysis of user strategies to manage these demands, including task decomposition, confidence adjustment, and workflow flexibility.
  3. Development of design considerations to reduce metacognitive demands and improve health information–seeking interfaces.
• Procedure and key techniques:
  • Conducted a think-aloud study with 15 participants using six simulated health scenarios.
  • Analyzed transcripts to identify metacognitive demands and coping strategies.
  • Mapped findings to the GenAI Human Workflow Framework to structure insights.

Results

• Concrete findings:
  • Participants faced challenges in prompt formulation (e.g., structuring symptoms), evaluating responses (e.g., assessing trustworthiness), and adapting workflows (e.g., managing agent-led shifts).
  • Metacognitive demands included self-awareness of health goals, task decomposition, confidence adjustment, and flexibility in workflow strategies.
  • Participants used strategies like breaking down prompts, testing agent trustworthiness, and cross-checking responses with external sources.
• Advantage over baselines: The study provides empirical evidence of metacognitive demands and strategies specific to health information seeking, which is underexplored in prior work.
• Experiments / evaluation:
  • Participants interacted with a conversational agent in a controlled environment using six health scenarios.
  • Data collection included think-aloud transcripts, demographic surveys, and a shortened Metacognitive Awareness Inventory.
  • Analysis focused on thematic coding and mapping to a structured workflow framework.
• Limitations and future work:
  • Small sample size (15 participants) limits generalizability.
  • Simulated scenarios may not fully capture the urgency and emotional stakes of real health situations.
  • Future work should explore longitudinal use, diverse populations, and evolving AI capabilities.

Summary

This study investigates the metacognitive demands users face when seeking health information using off-the-shelf AI conversational agents and the strategies they adopt to cope. Through a think-aloud study with 15 participants, the authors identified key challenges in prompt formulation, response evaluation, and workflow adaptation. Findings were mapped to a structured framework, and design considerations were proposed to reduce metacognitive demands and improve user experiences. These insights are critical for developing AI systems that better support health information seeking while minimizing user effort and risk.