Beyond the hype:
Navigating the challenges of ambient voice technology in mental health
Ambient voice technology - or ambient scribes - offers a compelling solution to clinician burnout, one of the most pervasive problems in the NHS. By passively listening to consultations and automatically generating clinical notes and downstream documents, these tools can save clinicians significant time and reduce burnout, whilst also improving the patient experience.
However, mental health care presents unique challenges. Unlike in physical medicine, the dialogue between clinician and service user in mental health is often the delivery of care itself - a multi-layered interaction rich with emotional subtext and nuance. Deploying scribes here is not a simple adjustment; it involves fundamental implications to therapeutic efficacy, patient safety, and ethics. This report goes beyond the promise of efficiency to explore the complexities of placing an artificial 'third party' into the therapeutic relationship, to ensure safe and responsible development and implementation of scribing tools.
We believe that an honest, transparent discussion about the unique challenges of using an ambient scribe in mental health is a crucial prerequisite to the safe and responsible development and implementation of the tool.
Two critical challenges
The impact on therapeutic alliance
While ambient voice technology can free clinicians from note-taking, the introduction of a non-human "third party" risks eroding trust, safety, and confidentiality, potentially leading to a "chilling effect" on patient disclosure, especially for vulnerable populations (such as those with paranoia or trauma).
The report explores how we can move consent beyond a mere tick-box exercise to an ongoing, transparent dialogue, and proposes three key ways to mitigate the potential negative impact on the therapeutic alliance.
The limitations of AI and effects on clinical practice
Mental health encounters are full of subtext, emotional nuance and meaningful silence. We dissect the limitations of large language models (LLMs) in such complex conversations, and discuss how to avoid “editing fatigue”, whereby clinicians spend too much time meticulously correcting verbose but clinically hollow notes.
Furthermore, we explore the potential long-term effects of ambient voice technology on clinical practice and proposes three key ways to overcome the limitations of AI in mental health.
.png){kind=small}
Accurx’s recommendations in the report:
Clinician focus and attention: Clinicians should provide context to the scribe before and a summary after the appointment, allowing natural conversation during the session.
.png)
Patient trust and safety: Clinicians must use clinical judgment to determine if scribe use is appropriate, even with consent, and avoid it when it may impact the therapeutic alliance. Accurx Scribe ensures patient data privacy by not using transcripts to train the AI model and allows for pausing the scribe mid-session for sensitive discussions.
Power dynamics: Consent must be clearly communicated as voluntary and revocable, with clinicians prepared to forgo scribe use if a patient shows discomfort, ensuring true patient autonomy.
Augmenting, not automating, clinical thought: Implement a robust human-in-the-loop (HITL) workflow where the scribe generates a first draft, but the clinician remains ultimately accountable for accuracy and clinical impressions. This requires new training for clinicians to adapt their cognitive synthesis processes.
Training on AI: Provide comprehensive training that goes beyond software mechanics, focusing on critically evaluating AI-generated text for errors of omission, hallucination, and substitution, and teaching clinicians how to effectively guide the scribe's output through prompts.
.png)
Avoiding 'editing fatigue': Design scribe tools to minimise editing burden. Accurx Scribe offers features like verbal instructions for note adjustment, customisable 'normal findings' and 'dot phrases,' and 'gold standard' templates to reduce manual editing and ensure high-quality documentation.