AI tools in clinical practice: beyond note-taking

AI has arrived in mental health from two directions at once. Our clients are already using it: more than one in three UK adults say they have used an AI chatbot to support their mental health or wellbeing, according to polling by Mental Health UK published in November 2025. And clinicians are using it in daily practice too.

I started using AI for my clinical notes in 2024, and I think this makes me an early adopter in the psychology world. Since then it has spread across my practice: local models for sensitive admin, agents for the technical side of my work, and purpose-built tools for clinical work.

What AI is genuinely good for in a clinical practice

The value sits in the work around the work, and there is far more of that than note-taking. In my own practice it shows up in three layers.

• Clinical-adjacent writing. Drafting letters and reports from my own notes, summarising long documents, drawing a formulation from material I already hold. I am editing rather than starting from a blank page.
• Sensitive admin, kept local. For admin that touches anything I would not want leaving my clinic, I run models locally, on my own hardware, so nothing is sent to a third party at all. This is the part most people miss: you do not have to choose between using AI and protecting data.
• The technical and the dull. I run agents to manage and optimise the technical side of my website, and to handle the administrative tasks that need none of my clinical input. This is where AI gives me the most time back, and it carries the least risk, because no client is anywhere near it.

None of this replaces clinical thinking. It clears room for it.

Is it safe to use AI for clinical notes?

Not by default, and not for everything. This is where I am strict, with myself and with the clinicians who ask me to help them adopt AI.

AI does not know your client. It cannot hold risk. It will write fluent, confident text that is sometimes wrong, and do it without a flicker of doubt. Two lines I do not cross:

• General-purpose tools. I do not put clinical material into consumer chatbots like ChatGPT, even with the identifying details stripped out. De-identification is weaker than it sounds: these tools can store and train on what you give them, and stripped-back clinical detail is easier to re-identify than people assume. For anything clinical I use a tool built for the job, with a clear data agreement and no training on what goes in, or I keep it on a local model. UK GDPR and your duty of confidentiality apply whatever a tool promises.
• Clinical judgement. Anything that touches a decision about a person needs your judgement on top of it. This is called having a 'human-in-the-loop' and it's something that is required for AI-assisted clinical work. The skill is working out which parts of your week are genuinely mechanical, and guarding the parts that are not.

Professional bodies are still catching up. The BPS has begun writing about what AI means for the profession, and it is worth following that guidance as it develops.

What comes after note-taking: agents and prompting

Note-taking is settled ground for me now. What holds my attention is what comes next, and some of it is already here.

I run agents that do not just answer a question but carry out a string of tasks for me, like maintaining and optimising my practice website. Watching them work has changed how I think about the skill that matters. The ability to prompt these systems well, to ask, to steer, and to push back on an answer, is starting to look more valuable than the engineering underneath them. If that holds, it is good news for clinicians: directing and interrogating something until it is right is close to what we do with people all day.

It is part of why I now supervise other clinicians on bringing AI into their practice. The questions are moving faster than the guidance, and too many of us are working through them alone.

---

Going further: at the AI in Psychology Summit, Natalie Stott speaks on how we can use these tools clinically in our own practice.