The Documentation Bottleneck

A progress note written straight after the session costs ten minutes you do not have. A progress note written at 8pm is written from memory, and memory is the first thing to go after eight sessions.

The AI Solution

Modern multimodal models have collapsed what used to be a multi-step pipeline. Rather than transcribing audio with one service and then analyzing the text with another, a single model can now listen to the session directly — separating the speakers, removing identifying details, and producing the clinical analysis in one pass. Fewer systems touch the recording, which matters more in a clinical setting than it does anywhere else.

The pipeline can:

• Transcribe audio sessions accurately.
• Draft the progress note in SOAP, BIRP or DAP format, from the session rather than from recall.
• Flag moments worth a second look, like avoidance or countertransference.

Reflection, Not Surveillance

The same analysis that drafts the note can show you what you missed in the room — a moment you became directive, a topic you steered away from. That is only useful if it is yours. A tool that reports those moments to someone else does not get honest material, because you will practise differently while it is recording.

The recording itself is destroyed once the analysis is done, so the material you are being honest about does not outlive the session.