AI is most defensible in a therapy practice where it touches the business, not the clinical work: scheduling, invoicing, drafting routine correspondence, and structuring your own administrative writing. It becomes risky precisely where it becomes attractive — anything involving identifiable client material. This piece is about the practice-management side only.
Scope note: this article deliberately does not cover AI journaling or client-facing reflection tools. That is a separate subject with a separate set of considerations, and mixing the two tends to blur a distinction that matters — a tool that helps you write an invoice is not operating in the same territory as a tool a client confides in.
What AI can realistically help with
The honest version of the current capability is narrow but real. Language models are competent at reformatting, summarising text you supply, drafting routine correspondence, and turning rough notes into structured prose. Speech-to-text is accurate enough on clear audio to save meaningful time. Scheduling and practice-management software increasingly bundles automation that is described as AI but is mostly rules — reminders, waitlist filling, invoice chasing.
None of this improves clinical work. It reduces the unpaid hours around it, which for many practitioners is the actual constraint.
Scheduling, invoicing and the inbox
This is the safest ground, because it need not involve clinical content at all. Automated booking and reminders reduce non-attendance administration. Invoice generation and chasing removes a task most therapists dislike and therefore defer. Template correspondence — waiting-list updates, fee-change notices, holiday cover, referral acknowledgements — can be drafted once with assistance and reused thereafter.
Two rules make this uncomplicated. Keep names and clinical detail out of any general-purpose AI tool; draft in the abstract and fill in the specifics yourself. And check that any scheduling system holding client contact details is one you have assessed as a data processor, whether or not it markets itself as AI.
Transcription and note support: the highest-risk area
Session transcription and automatic note generation are the most heavily marketed applications and the ones requiring the most scrutiny. Recording a session means creating a new, highly sensitive record that did not previously exist, and sending it to a third party for processing.
Before considering it, the questions are not about accuracy: they are about consent that is genuinely informed and freely refusable; where audio and transcripts are stored and for how long; whether content is used to train the supplier's models; whether sub-processors are involved and in which jurisdictions; what your data protection assessment concluded; whether your indemnity insurer is aware; and what happens to the record if the supplier is acquired or fails.
Accuracy matters too, in a specific way: a plausible-sounding summary of something a client did not say is more dangerous than an obviously garbled one, because it survives review. Any generated note is a draft you are professionally accountable for, not a record.
Confidentiality and data handling
Client material is special category personal data. Using a tool that processes it makes that supplier part of your data-protection position, with the documentation that follows: a lawful basis, a privacy notice that reflects reality, a processor agreement, a retention period, and a plan for breaches. The Information Commissioner's Office publishes guidance for small organisations covering each of these.
A practical discipline that removes most of the difficulty: treat general-purpose AI tools as public. If you would not paste the text into a public forum, do not paste it into a chatbot. Nearly every safe administrative use survives that rule.
What your professional body expects
Registers set expectations about confidentiality, record-keeping, informed consent and competence that apply regardless of the technology involved, and guidance specific to AI is developing. Check your own body's current position — BACP, UKCP, or HCPC for regulated titles — rather than relying on a supplier's claim that a product is compliant. Compliance is a property of your practice, not of a purchase.
Bring it to supervision as well. A change in how session material is recorded is a change in practice, not a procurement decision.
What AI should not be asked to do
- Clinical judgement. Formulation, risk assessment and decisions about treatment are yours and cannot be delegated.
- Risk detection. Do not rely on a tool to notice deterioration or danger.
- Standing in for you between sessions without an explicit, agreed framework about what it is and is not.
- Producing records you have not read. An unreviewed generated note is still your record.
- Writing claims about your competence. Marketing copy that overstates scope is a professional issue whoever drafted it.
A cautious way to start
Begin where no client data is involved: correspondence templates, website copy, structuring your own reflective or supervision-preparation writing, and administrative summaries of your own text. Run it for a term. Note what time it actually saved rather than what it felt like it saved. Only then consider anything touching session material, and only with consent, a documented assessment and your insurer informed.
Where this fits
Administrative efficiency is usually the second problem. The first is being findable, which is covered in how clients actually find therapists. The foundations underneath both — registration, CPD, supervision and insurance — are mapped in therapist resources worth knowing about. If some of the time saved is destined for a second income stream, the pillar guide to side hustles for therapists is the place to start, and writing prompts for professional reflection covers the part of the writing that is worth keeping human.
Practitioners can list a practice in the Echo practitioner directory, and the practitioner workspace sets out how Echo supports the professional side of practice.
This article is general information for professionals and not legal, data-protection or regulatory advice. No products are endorsed or compared here; confirm your own obligations with your professional body, insurer and data-protection guidance.