Prompt Coaching for Inclusiveness: A Media Literacy Approach to Increase Users’ Awareness of Algorithmic Bias and Prompting Efficacy

Paper Title

Prompt Coaching for Inclusiveness: A Media Literacy Approach to Increase Users’ Awareness of Algorithmic Bias and Prompting Efficacy

Publication Info

• Topic area: Media literacy intervention for combating algorithmic bias in generative AI.
• Keywords: Inclusive prompt coaching, algorithmic bias, generative AI, media literacy, trust calibration, user agency, design friction, prompting efficacy, user experience, cognitive elaboration.

Background and Problem

• Problem / challenge: Generative AI systems often produce biased or stereotypical outputs, reflecting systemic issues in training data and algorithm design. Users frequently fail to recognize these biases, and existing approaches, such as data documentation, are limited in addressing this challenge during real-time interactions.
• Significance: Addressing algorithmic bias is critical for reducing representational harm, particularly for minority communities, and for fostering trust and responsible use of AI systems.
• Motivation and related work: Prior studies have explored ethical linguistic interventions and data transparency but have not focused on user-centered strategies to enhance awareness of bias during AI interactions. This paper builds on research in media literacy and design friction to propose a real-time, user-focused intervention.

Solution

• Proposed approach: Inclusive prompt coaching—a media literacy intervention that guides users to write more inclusive prompts during their interaction with generative AI systems.
• Novelty:
  1. Introduces a user-centered design strategy to raise awareness of algorithmic bias through real-time interaction.
  2. Identifies cognitive elaboration as a mechanism for improving trust and trust calibration in AI systems.
  3. Provides a practical implementation of inclusive prompt coaching, adaptable to various generative AI contexts.
• Procedure and key techniques:
  • Developed three AI systems: inclusive prompt coaching, detailed prompt coaching (control), and no coaching (baseline).
  • Inclusive coaching flagged potential biases in user prompts and suggested more inclusive alternatives.
  • Conducted a user study with 344 participants to evaluate the effects of inclusive prompt coaching on algorithmic bias awareness, trust, prompting efficacy, and user experience.

Results

• Concrete findings:
  • Inclusive prompt coaching increased algorithmic bias awareness (M = 5.18) compared to no coaching (M = 4.82; p = .034).
  • Enhanced perceived prompting efficacy (M = 5.25) compared to no coaching (M = 4.95; p = .01).
  • Indirectly improved trust outcomes (cognitive, affective, behavioral) and trust calibration through cognitive elaboration.
  • Inclusive coaching caused higher frustration (M = 2.73) compared to no coaching (M = 2.06; p = .018).
• Advantage over baselines:
  • Inclusive coaching outperformed no coaching in raising bias awareness and prompting efficacy.
  • Cognitive elaboration mediated trust improvements, distinguishing inclusive coaching from no coaching.
• Experiments / evaluation:
  • Participants were randomly assigned to one of three conditions (inclusive, detailed, no coaching).
  • Measured outcomes included algorithmic bias awareness, trust, prompting efficacy, and user experience.
  • Quantitative analysis (ANCOVA, MANCOVA) and qualitative coding of user responses were conducted.
• Limitations and future work:
  • Small effect sizes and one-time exposure limit generalizability.
  • Lack of explicit definitions for inclusiveness and bias in meta-prompting.
  • Mixed user experience due to perceived loss of agency and frustration.
  • Future work should refine coaching mechanisms, tailor interventions to user goals, and test long-term effects.

Summary

This study introduces inclusive prompt coaching as a real-time media literacy intervention to raise awareness of algorithmic bias and improve prompting efficacy in generative AI systems. Through a user study, the approach demonstrated increased awareness of bias, enhanced trust via cognitive elaboration, and higher confidence in crafting inclusive prompts. However, it also revealed challenges in user experience, such as frustration and reduced agency. These findings highlight the potential of inclusive prompt coaching to empower users while underscoring the need for further refinement to balance usability and effectiveness.