BMJ Digital Health & AI review of 27 studies finds ambient AI scribes miss nonverbal cues and patient narratives

The University of Edinburgh team examined implementation risks of ambient AI scribes that combine speech-to-text with large language models. They identified consistent gaps in capture of nonverbal and psychosocial elements across primary care settings. Forty percent of UK GPs already use these tools, yet the review documented clinician reports of reduced memory recall and unrecognized notes after cognitive offloading during consultations.

Analysis links these documentation shortfalls to broader quality-of-care risks. Patients facing marginalization report hesitation disclosing substance use, domestic violence, or mental health concerns when aware of AI processing, potentially widening existing disparities. Manual note-taking, by contrast, supports iterative clinical reasoning; its replacement may accelerate deskilling, especially for trainees who lose opportunities to rehearse synthesis of verbal and nonverbal data.

Related observational work in JAMA Network Open (2025) on ambient documentation similarly noted 18% lower completeness scores for psychosocial domains versus traditional notes. A 2024 Lancet Digital Health commentary on the NASSS framework emphasized that technology designed in high-resource systems often underperforms when transferred to differently structured health services, a gap the Edinburgh authors flag for future study.

Next steps include longitudinal mixed-methods evaluations in varied national systems, with explicit metrics for patient voice preservation and clinician reasoning retention measured at 12- and 24-month intervals.