Can psychologists use AI scribes in Australia? Learn what Ahpra requires around informed consent, privacy, accountability, clinical records, TGA regulation and human oversight.
AI scribes are not banned for psychologists in Australia.
That is the easy part.
The harder part is that an AI scribe can turn one private therapy hour into an audio file, a transcript, a draft note, a vendor data flow, a consent issue, a clinical record problem, and potentially a medical device question. If you treat it as a harmless admin shortcut, you are already behind the risk.
Psychologists in Australia can use AI scribes if they can meet their professional, privacy, consent, record keeping, and legal obligations.
That "if" is doing a lot of work.
Ahpra's current AI guidance does not say psychologists must avoid AI. It says practitioners remain responsible for how they use it. The Psychology Board of Australia also directs psychologists to Ahpra's AI guidance when considering AI and technology in practice.
So the question is not:
"Can I use an AI scribe?"
The better question is:
"Can I explain exactly what this tool does, obtain meaningful consent, protect the client's information, review the output properly, and remain accountable for the final record?"
If the answer is no, the tool is not ready for clinical use in your practice.
Ahpra's AI guidance is built around five practical principles:
These are not decorative values. They create a working standard for whether AI use is clinically defensible.
For a psychologist, that means:
The Psychology Board's Code of conduct also matters. Since 1 December 2025, psychologists are regulated under the Psychology Board of Australia Code of conduct rather than the APS Code of Ethics as the Board's primary regulatory code. That code requires good care, clear communication, privacy, confidentiality, culturally safe practice, accurate records, and appropriate use of technology.
AI does not sit outside those duties. It sits inside them.
If an AI scribe produces a note that misses a suicide risk disclosure, attributes a statement to the wrong person, invents a diagnosis, or stores sensitive information somewhere the client did not agree to, "the software did it" is not a defence. It is an explanation of the failure.
"AI scribe" is not one thing.
Some tools are essentially transcription products. They record or listen to an appointment and produce a transcript or summary. Others do more:
Those functions do not carry the same risk.
A tool that only converts speech into text has one profile. A tool that analyses a therapy session and suggests a diagnosis has another. A tool that stores audio overseas and uses content to train future models has another again.
Clinicians often talk about AI scribes as if they are a single admin product. Regulators do not need to treat them that way, and neither should you.
Before adopting a tool, ask what it actually does at each point in the workflow:
If the vendor cannot answer those questions in plain language, that is not a small procurement problem. It is a clinical governance problem.
AI-generated notes are drafts.
Not "mostly done" records. Not neutral summaries. Drafts.
The psychologist must review and correct the final record before relying on it. That review needs to cover more than typos.
Check for:
Ahpra's AI scribing case material specifically warns that AI tools can add diagnoses, omit clinically important information, or produce output that sounds confident but is inaccurate.
That matters because psychology notes are not ordinary admin files. They can shape continuity of care. They can be read in complaints. They can be requested in family law disputes. They can be subpoenaed. They can influence NDIS, insurance, or workplace decisions.
If you use AI to draft the record, your job is not to click approve faster. Your job is to make sure the record is clinically true, proportionate, and defensible.
For practical note-writing structure, pair this article with how to write psychology progress notes that are clear, fast, and defensible.
Consent is where many AI scribe workflows get sloppy.
If the tool records a session, you need consent before recording. Recording laws vary between Australian states and territories, so practices should check the law that applies in their jurisdiction. But the professional baseline is straightforward: do not record therapy without explicit consent.
Consent should not be buried inside a long intake form and treated as permanently solved.
Clients should understand:
A good consent conversation separates several choices that are often bundled together:
Those are different things.
A client might accept an AI-generated summary but refuse audio storage. A trauma client might refuse any recording at all. A young person might be unable to give meaningful consent without appropriate parent or guardian involvement. A separated parent might consent for themselves but not for information about the other parent to be recorded in an unmanaged transcript.
The client also needs a real option to decline. If the only alternative is "no appointment", consent starts looking less voluntary.
Health information is sensitive information under the Privacy Act.
That means an AI scribe workflow is not just a note-taking workflow. It is a health information handling workflow.
The risk is often in the steps clinicians do not see:
Privacy due diligence needs to happen before clinical use.
Ask the vendor for clear answers:
Data minimisation is also a clinical skill here. A full transcript can contain more sensitive information than the final note should ever hold: names of family members, allegations, sexual history, immigration details, workplace conflicts, details about children, and information about people who never consented to anything.
If raw transcripts and audio are retained, they may become discoverable, requestable, or relevant in complaints and legal disputes. That does not mean they can never be retained. It means retention needs to be deliberate, justified, disclosed, and governed.
General-purpose AI creates a related risk. If a psychologist copies session details into a general chatbot to "clean up" notes, that may still disclose client information to a third party. Removing the client's name is not always enough. Context can be identifying.
For technology-adjacent care risks beyond clinician documentation, see AI safety for young people, families, and clinicians in Australia.
Not every AI scribe is a medical device.
The Therapeutic Goods Administration draws an important distinction. Software that only transcribes or translates a clinical consultation is generally different from software that analyses or interprets health information and produces diagnosis, differential diagnosis, treatment recommendations, or other clinical outputs.
That line matters.
If a tool only creates a transcript for the clinician to review, it may sit outside medical device regulation. If it starts making clinical recommendations, interpreting symptoms, or generating diagnostic suggestions, it may fall into medical device territory and may need to be included in the Australian Register of Therapeutic Goods unless an exemption or exclusion applies.
Do not rely on marketing language.
Ask:
TGA has also been consulting on digital scribes. That area is still moving, so practices should keep a review date in their governance process rather than treating today's vendor answer as permanent.
Some sessions are simply worse candidates for AI scribing.
Not because the client is difficult. Because the stakes, privacy exposure, and potential downstream consequences are higher.
Use extra caution with:
In these settings, a transcript may capture details that are clinically sensitive, legally risky, culturally significant, or unsafe if mishandled.
This is also where power matters. A client may say yes because they do not want to disappoint the psychologist, look difficult, slow down the appointment, or lose access to care. That is not meaningful consent. That is compliance under pressure.
AI scribing may still be usable in some higher-risk contexts, but the threshold for consent, minimisation, and human review should be higher.
If a practice wants to use AI scribes, build the workflow first.
Do the boring governance work. It is the work that protects you.
Consent should be live enough to mean something.
Keep clinical control.
Treat the output as draft material.
This workflow is slower than pretending the tool is just a smart Dictaphone. It is also much easier to defend.
AI scribes may reduce documentation labour.
They do not reduce the standard of care.
For psychologists in Australia, the defensible position is simple: use AI only where you understand the tool, have meaningful client consent, can protect health information, can explain the data flow, can review the output properly, and can remain professionally accountable for the final record.
Convenience is a benefit. It is not the ethical test.
If you are building or reviewing your documentation systems, start with the parts humans still need to own: clear progress note structure, consent language, privacy practices, and clinical judgment. PsychVault's clinical documentation resources and resource creator tools are built around that same principle: reduce admin load without handing away responsibility.
Language note: This article uses "AI scribe" broadly because that is the term most clinicians use. In practice, tools vary widely. A transcription assistant, summarisation tool, and clinical decision support product can create very different legal, privacy, and regulatory risks.
Share your thoughts and experiences with this resource.
Sign in to leave a comment
Move from strategy into implementation with templates, handouts, and psychoeducation tools already live on the marketplace.
Publish clinician-grade templates, build trust signals, and start growing an evergreen library under your own brand.
A practical guide for provisional psychologists on continuity of care when therapy ends because a placement ends, including handover, consent, timing, and documentation.
A practice-focused guide to dual relationships in psychology: what they are, why they matter, when they are manageable, and how Australian ethical frameworks expect psychologists to think about risk.
A plain-English guide to Australia's proposed single 5-year psychology pathway, what has changed, what has not, and what current students and provisional psychologists should do now.
