Published: 24/4/2026 | Reading time: 4 minutes

AI Scribes Are Making Headlines. Patient Letters Still Need Getting Right.

smarter. faster. more organised.

There is no shortage of AI documentation tools in healthcare right now. Ambient scribes. Auto-generated clinical notes. AI that listens to your consultation and produces a summary before you’ve said goodbye to the patient.

The research on these tools is genuinely promising – and genuinely cautious in equal measure.

A landmark randomised controlled trial published in NEJM AI in December 2025 found that ambient AI scribes reduced time-in-note and showed improvements in clinician burnout across 238 outpatient physicians in 14 specialties. But the same study noted that occasional inaccuracies in either scribe tool tested require ongoing vigilance. This was the first RCT of ambient AI scribes ever published. The findings are encouraging. The caveats matter.

A systematic review published in BMC Medical Informatics and Decision Making in July 2025 – the most comprehensive review of its kind, drawing on MEDLINE, Embase, and the Cochrane Library across 29 studies – found that word error rates in AI transcription ranged from under 1% in controlled dictation settings to over 50% in conversational or multi-speaker scenarios.

Specialist consultations are conversational, multi-speaker scenarios.

That is precisely the environment your practice operates in every day.

A separate review published in Perspectives in Health Information Management across 129 studies concluded bluntly that moderately high error rates preclude the broad use of a comprehensive AI documentation assistant. A Lancet-published, NIHR-funded review in eClinicalMedicine (July 2025) – covering 524 healthcare professionals and 1,069 consultations – highlighted that transcription inaccuracies, especially with medication names, pose direct safety risks.

The research is not anti-AI. It is pro-accuracy. There is a difference.

Dict8ion Is Not an AI Scribe

This distinction matters more than it might first appear.

AI scribes listen to the consultation and generate structured clinical notes from the conversation. That is a fundamentally different task to what Dict8ion does – and it carries fundamentally different risk.

Dict8ion converts a doctor’s dictation – recorded after the consultation, in their own words, using their existing dictaphone, voice app, or the Dict8ion app – into a formatted patient letter. No ambient listening during the consultation. No AI making clinical inferences from a conversation. No output generated without the doctor first deciding what to communicate.

The doctor is already in control of the content before Dict8ion touches it. That is a meaningful distinction for clinical governance, for accuracy, and for the care team receiving that letter.

Why Human-in-the-Loop Is Not a Compromise – It’s the Standard

Every piece of current research lands in the same place: AI tools in clinical documentation require human review.

The Australasian Institute of Digital Health (AIDH) June 2025 guidance on AI scribes in healthcare is unequivocal: healthcare practitioners remain fully responsible for the accuracy of AI-generated documentation. Errors of omission – where AI misses something rather than gets it wrong outright – may be particularly difficult to detect.

A Yale multi-centre study across six health systems published in JAMA Network Open in October 2025 found real benefits from ambient AI scribes after 30 days of use, with burnout dropping from 51.9% to 38.8%. It also found that clinician review remained an essential component of responsible implementation. A rapid evidence synthesis published in JMIR AI in October 2025 noted that while AI scribes decreased self-reported documentation times, physician productivity via billing metrics was unchanged – and note length actually increased. More words is not the same as better letters.

Dict8ion is built around a different principle entirely. Every letter produced by the AI is ready for your practice team to proof – not delivered as a finished document. The clinician approves. The admin team finalises. The letter goes out under the doctor’s authority, not the algorithm’s.

AI speed. Human accuracy. That is not a marketing line – it is how responsible clinical correspondence works.

The Efficiency Gain Is Real. The Workflow Change Is Not.

One of Australia’s largest ENT practices processed 18,636 letters in 2025 using Dict8ion, achieving an 80% reduction in letter processing time. A second NIH-funded RCT published in NEJM AI – a stepped-wedge trial across 66 practitioners over 24 weeks – found that AI-assisted documentation time decreased without compromising diagnosis, billing compliance, or note quality. The efficiency case is solid. What does not need to change is how the doctor works.

The doctor did not change how they dictate. They did not sit differently, speak differently, or learn a new system. They dictated as they always have. Dict8ion handled the transcription and formatting. The admin team received a clean, structured draft ready to proof and send.

That is the model. smarter. faster. more organised. – and no one had to reinvent how they work to get there.

The Right Tool for the Right Job

If you are evaluating AI documentation tools for your practice, the questions worth asking are:

• Does this tool require the doctor to change their consultation behaviour?
• Who is responsible for accuracy when the AI gets it wrong?
• What is the output – a clinical note, or a formatted patient letter?
• Is there a human step before anything goes out?

Dict8ion is purpose-built for patient letter correspondence. It is not trying to replace clinical judgment, ambient listening, or the consultation itself. It handles the part of the workflow that consistently creates the most backlog: turning a dictation into a formatted, proofed-ready letter that goes to the full care team.

That is a smaller promise than some AI tools make. It is also a promise we keep.

Explore Dict8ion for your role: Clinicians | Practice Managers | Administration Staff

Read our case studies | How it works

Dict8ion.ai | AI speed. Human accuracy.

Further resources

• Frequently Asked Questions
• Dict8ion for Admin staff
• Dict8ion for Practice Managers
• Edify Medical – Knowledge Technology