Designing Accessible Audio Nudges for Voice Interfaces

Research Background and Issues

• Identified Problems or Challenges: Older adults (65 years and above) face difficulties in assessing the quality of information when using voice assistants (VAs) to search for high-risk health information due to the lack of visual cues. This results in information uncertainty, including doubts about the accuracy, completeness, practicality, and credibility of the information.
• Importance of the Problem: Uncertainty in high-risk health contexts (e.g., searching for information related to health issues) can lead to trust errors or misunderstandings, potentially endangering lives. Given the increasing reliance of older adults on voice assistants as information sources, mitigating information uncertainty is particularly urgent.
• Research Motivation and Related Work:
  • Extensive research has shown that nudging techniques in visual interfaces effectively guide users toward reflective decision-making, but studies on nudging techniques for voice interfaces remain scarce.
  • The lack of sufficient information quality cues (e.g., webpage design elements and information sources) in voice interfaces exacerbates uncertainty, and existing information verification strategies fail to effectively support older adults in critical thinking.
  • This study explores how to design audio nudges in voice assistants to help older adults reflect and resolve uncertainty during high-risk information searches.

Proposed Solution

• Proposed Solution: Design and evaluate four audio nudge prototypes, including two voice-based nudges (declarative and interrogative) and two non-voice-based nudges (based on sound characteristics, such as descending tones and valley-shaped sound effects).
• Innovations:
  1. Introducing audio nudges into voice assistants, expanding previous nudging techniques that primarily focused on visual interfaces.
  2. Conducting a preliminary exploration of the attributes of reflective audio nudges, emphasizing the importance of "explanatory" and "interactive" elements in reducing information uncertainty.
  3. Investigating the design challenges of balancing "accessibility" and "disruptiveness" in audio nudges.
• Implementation Steps and Key Techniques:
  1. Designing four audio nudge prototypes based on prior literature.
  2. Testing these prototypes using three health-related examples (sinus infection, kidney stones, and migraines).
  3. Conducting user studies with 34 older adults using the Wizard of Oz technique, including task operations and interviews to collect user feedback on various nudges.

Research Findings

• Specific Findings:
  1. Voice-based nudges, especially interrogative nudges, are more effective than non-voice-based nudges in prompting users to reflect on the credibility of information provided by voice assistants.
  2. Non-voice-based nudges were often ignored or misunderstood as system errors due to unfamiliarity or lack of clarity in their design.
  3. Older adults prefer audio nudges that explicitly guide them in verifying information credibility rather than providing vague hints.
  4. Well-designed audio nudges need to balance "disruptiveness" to reflect the severity of the issue while avoiding excessive user annoyance or cognitive burden.
• Advantages Compared to Existing Solutions:
  • Unlike traditional visual nudges, this study provides a unique non-visual perceptual solution for voice interfaces, supporting credibility judgments for older adults and visually impaired users.
  • Interrogative nudges effectively intervene in user behavior with voice assistants, encouraging deeper critical reflection.
• Experimental Results:
  • Users generally showed higher acceptance of voice-based nudges compared to non-voice-based nudges. While declarative nudges in voice assistants increased information credibility, they could lead to over-trusting, whereas interrogative nudges actively triggered questioning and further investigation.
  • Non-voice-based nudges had limited effectiveness, with many users perceiving them as background noise or system errors.
• Limitations and Future Directions:
  1. The range of nudge methods designed in this study is limited; future research could explore more types of audio nudges (e.g., hybrid nudges).
  2. Participants primarily consisted of older adults in good health; future studies should investigate the needs of vulnerable groups (e.g., visually impaired or cognitively impaired individuals).
  3. The study was based on short-term experiments; future research could examine the impact of long-term use on habit formation and the sustainability of nudges.
  4. There is a conflict between design diversity and user familiarity, requiring further exploration of how to balance specificity and universality.

Through this research, the paper provides design insights into effectively integrating audio nudges into voice interfaces while raising open questions about how sound-based interventions can balance "accessibility" and "disruptiveness."