As explored in our earlier article on burnout, documentation consumes approximately 35% of a clinician's working time. AI workflows that automate this process have demonstrated meaningful reductions in after-hours charting and improvements in clinician wellbeing across a range of specialties.
However, in psychiatry, the calculus is far more complicated. Mental health consultations are not simply medical interviews with emotional content. They are carefully constructed therapeutic relationships built on trust, safety, and the patient's willingness to be vulnerable. Introducing technology into that space requires a degree of deliberation that goes well beyond standard implementation guidance.
How Ambient scribes work and where they fall short
Modern ambient scribes use natural language processing to detect spoken words, transcribe them into text, and generate structured clinical output. The quality of that output depends entirely on what the transcript contains. In general medicine, this model works well: the words spoken in a consultation carry most of the diagnostic and clinical information needed.
In mental health, the relationship between what is said and what is clinically meaningful is far more complex. A patient who delivers the words "I'm fine" in a flat monotone, after a prolonged silence, following a question about their mood, is communicating something that the transcript cannot capture. Tone, volume, pacing, and silence are central to psychiatric assessment and current large language models interpret silence simply as an absence of input.
The therapeutic alliance as a clinical variable
The therapeutic alliance, the quality of the relationship between clinician and patient is one of the most robustly evidenced predictors of treatment outcome across psychiatric conditions. Anything that disrupts the formation of that alliance carries direct clinical risk.
For patients experiencing persecutory ideation, awareness that a consultation is being recorded and processed by an unseen system can actively reinforce delusional beliefs and make trust-building significantly harder. For survivors of trauma, recording technology may feel unsafe or intrusive, particularly early in the therapeutic relationship. Studies examining patient attitudes toward AI in healthcare have consistently identified confidentiality and data privacy as primary concerns.
The solution is not to abandon the technology, it is to approach consent differently. At Aisel Health, we believe consent for AI use in mental health settings should move well beyond a tick-box exercise. It should be an ongoing, transparent dialogue, one that explains what the tool does, what happens to the data, who can access it, and what the real risks are.
Clinician judgement matters here too. There are cases where the right call is simply not to use the tool at all. Aisel is built with this flexibility, recording can be paused mid-session, and clinicians retain full discretion over when the software is appropriate.
What the transcript cannot tell you
The second challenge is the degree to which AI language models can fully capture the texture of psychiatric dialogue. Silence, for instance, carries enormous clinical weight in mental health. A skilled clinician reads a patient's silence whether it signals dissociation, resistance, grief, or processing. Current LLMs interpret silence as an absence of input. The emotional undertone of a flat, monotone delivery of the words is invisible to a model reading a transcript.
Our approach to this is practical rather than theoretical. We encourage clinicians to verbalise clinically meaningful moments narrating a significant silence, noting a shift in affect so the scribe can capture what the transcript alone would miss. Our templates are built by clinicians specifically for psychiatric workflows, designed so the first output requires minimal editing rather than a full rewrite.
Protecting clinical reasoning
There is a subtler risk that sits beneath both of these challenges. Documentation tools that automate note generation may, over time, reduce the cognitive synthesis that underpins clinical reasoning particularly for early-career clinicians who are still developing their formulation skills. If the process of translating an encounter into written form is where much of the clinical thinking happens, removing that process entirely carries consequences that are not yet well understood.
Our approach
Aisel Health's clinical documentation tools are designed with psychiatric practice in mind. Our templates are built by clinicians for psychiatric workflows, with the goal of producing first drafts that require minimal editing rather than whole rewrites. We support clinicians in verbalising clinically significant moments narrating a meaningful silence, noting a shift in affect so that the scribe captures not only what was said but the clinical context surrounding it.
On data governance, no patient data is used to train third-party models. Our software is fully GDPR compliant, and recording can be paused mid-session at any point giving clinicians real discretion rather than nominal control.
Integrating AI into mental health practice is not a technical problem with a technical solution. It is a clinical challenge that requires the same consideration, sensitivity, and ongoing refinement that good psychiatric practice has always demanded. We are committed to working alongside clinicians on that journey.
Get in Touch 💙
We are always keen to connect with people working in mental health and feel free to reach out & book a demo on Aisel.co