In a mental health clinic in a quiet London suburb, a clinician listens attentively to a young person describe their mental health challenges. A scene repeated across the country, but with one crucial difference: there are no notes being taken and the clinician is free to fully and openly listen to the patient. The conversation concludes and, before the patient has left the building, a report has been written automatically, completing a process in minutes which used to take hours. Only a few years ago, this would have seemed like science fiction. Today, however, a small group of clinicians in North West London and Milton Keynes are already piloting Artificial Intelligence (AI)-based ambient voice technology (AVT), raising important questions about the way psychiatrists work and train.
Generative AI is being introduced broadly across all sectors of the economy and AVT is emerging as a key tool for improving clinician wellbeing and productivity.1,2 These systems work alongside clinicians in a variety of settings, listening to and transcribing clinical dialogue between them and their patients. The software then uses a type of AI called a Large Language Model to synthesise the conversation into coherent clinical documentation. Outputs can additionally be customised to fit the requirements of the service, requiring only limited proofreading before submission.
Experience of AVT use in mental health settings is limited, however early data from our pilot study in child and adolescent mental health services suggests it can substantially reduce time spent on paperwork.3 Additionally, whilst the current evidence base is limited, the pace of implementation in other clinical areas suggests that trainees are likely to encounter these systems in practice before their impact is fully understood. It is therefore important that both the potential benefits and associated risks are addressed within psychiatric training programmes.
Good clinical documentation remains a key skill as defined by the Royal College of Psychiatrists’ (RCPsych) curriculum, requiring trainees to ‘have an ability to synthesise complex information and communicate this succinctly and coherently in your written and verbal communications’.4 General concerns about de-skilling are beginning to surface in the literature, and were noted in focus groups within our pilot study: some felt that the ability to recall and synthesise information without technological assistance remained a valuable clinical skill.3,5 Skills potentially at risk include written formulation, proportionate risk documentation, summarisation, and explicit clinical reasoning; within the field of psychoanalysis, it has been proposed that the act of recording findings can help clinicians come to more reasonable conclusions.6 Training objectives should therefore include the ability to produce high-quality documentation without AVT, assessed through supervised case writing, workplace-based assessments, and portfolio reflections of assisted versus unassisted outputs.
Nevertheless, the potential benefits for improving the quality and consistency of letters and notes are substantial. In 2021, the RCPsych issued guidance on the quality and standards expected of psychiatrists in their clinic letters, with clear expectations for the content and tone of these documents.7 The guidance has been integrated into the model used in our pilot, and it could become feasible for consistent, comprehensive and sensitive letters to be written wherever patients receive care: free of medical jargon and superfluous information. This is particularly important in healthcare systems that increasingly value patients accessing their own data.
Known issues with AI implementation remain prevalent within clinical AVT and the maintenance of a ‘human-in-the-loop’ remains challenging in practice. Confabulations (known as ‘hallucinations’), omissions, and other performance dysfunctions, whilst rare, do occur. There remains a risk that bizarre or false information could be entered into patient notes, with potential ramifications for patient safety. This could be a risk for clinicians lower in confidence or digital competence who may be more willing to be led by an AI output, rather than assisted by it. It is crucial that clinicians carefully proofread outputs and take full responsibility for the content. Training may need to be provided to ensure none are left behind. Additionally, psychiatric consultations have particular sensitivities with regards to capacity, trauma histories, and safeguarding. Consent for AVT should be considered a core competency, requiring trainees to explain what is recorded, how and where data are processed and retained, available opt-out mechanisms, and to reassure patients that refusal will not affect care continuity, particularly given psychiatric sensitivities relating to trauma histories and safeguarding.
In the United Kingdom’s National Health Service (NHS), psychiatry training operates across regional programmes. These must evaluate whether the time saved by AVT enhances educational opportunities, rather than being solely directed toward increased productivity, with a knock-on effect on ringfenced training time. An evidence review from 2023 identified that less than 1% of the literature on the impact of new technology on staff time considered how the time was repurposed.8 The same review identified in a survey of clinicians that, if three hours of their time were saved per week, 40% of them would like to use this time on education, training, and continuing professional development. It is not yet known how health organisations will respond to any increase in the availability of clinical staff. The impact and redistribution of time will likely vary between services depending on clinical pressures, the cohort of patients being cared for, or the culture within the team. Without careful planning, AVT could inadvertently widen existing disparities in training.
As AI technologies become embedded in psychiatric practice, the role of trainees and the competencies expected of them must be established. Whilst the potential to reduce administrative burden and improve documentation quality and standardisation is substantial, the risks must not be ignored. Trainees should be considered key to the effective use of these tools: contributing to implementation, ensuring clinical oversight, and advocating for its inclusion in training. Preparing for a future where AI is routinely ‘in the room’ demands a curriculum that reflects the relative advantages and limitations of both human and artificial intelligence. Our challenge as trainees is not just to adopt new technology, but to lead responsibly in shaping its future in psychiatric care.
Author contributions
Dr Mark Brewerton took the lead on conception and design of the article, as well as co-ordination of collaboration between contributors. Dr Noah Stanton and Dr Salim Jakhra contributed equally via conceptual discussions around the topic, reviewing it critically for important intellectual content and final approval of the version to be published. All authors agree to be accountable for all aspects of the work.
Ethics statement
No ethical approval was required for the creation of this article.
Declaration of conflicts of interest
Since 11 August 2025, Dr Noah Stanton has been employed by an AI AVT provider, Anathem Ltd, in a part time role as a clinical prompt engineer, in addition to his clinical work as a psychiatric resident with Central and North West London NHS Foundation Trust. There are no other potential conflicts of interest to disclose from any of the other authors.
Role of funding
This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.