Can Psychologists in Australia Use AI Scribes? What Ahpra Expects

Yes, psychologists in Australia can use AI scribes, provided they keep full accountability. Ahpra expects you to tell clients you are using one, obtain informed consent (with an explicit consent step before any recording), check every note for accuracy, understand how the tool stores and reuses data, and confirm your professional indemnity insurer covers it.

Most pages on this topic review scribe products and skip the regulator entirely. Ahpra's guidance, first published in 2024 and, by its own description, reviewed and updated as AI develops, is the document your Board will point to if a complaint arrives, so it makes a better foundation for your checklist than any vendor's marketing. What follows maps each of its principles to a concrete action you can complete in an afternoon.

Follow me on Instagram@sagnikteaches

What Ahpra's guidance actually says about scribes

The guidance is titled "Meeting your professional obligations when using Artificial Intelligence in healthcare". It is not a new law and it does not ban anything. It explains how the existing codes of conduct apply when a practitioner brings AI into practice. Read it directly on the Ahpra website, because it is updated from time to time, and look at the case studies Ahpra publishes alongside it, which show how the principles play out with newer generative AI tools.

Connect on LinkedInSagnik Bhattacharya

It organises its expectations under a handful of headings: accountability, understanding, transparency, informed consent, and ethical and legal issues. Scribing is named specifically. It says tools like medical scribes are being used to manage workload and to develop or edit documents, and that generative AI scribing tools are generally general-purpose products without a therapeutic use, so they usually fall outside the Therapeutic Goods Administration's medical-device rules. That sounds like good news, but it cuts the other way: nobody external has approved the tool for you, so the testing and judgement fall on you.

Subscribe on YouTube@sagnikteaches

The table below turns each heading into the question a Board assessor would ask and the evidence you would want on file.

Accountability: the note is yours, however it was drafted

Ahpra is blunt that regardless of the technology, the practitioner remains responsible for safe, quality care and must apply human judgement to any AI output. For a scribe, it adds that you are responsible for checking the accuracy and relevance of records created using generative AI. Approval of a tool by the TGA, where that applies, does not change that, and tools should be tested before clinical use.

In practice, "checking" has to be a real step, not a glance. Psychology notes carry details a transcript engine gets wrong in predictable ways: a negation dropped ("denies suicidal ideation" becoming "suicidal ideation"), a medication name misheard, a family member's name attached to the wrong person, or a risk statement softened into something blander than what was said.

A review routine that takes three minutes

Say a solo psychologist sees six clients a day and drafts each note in the scribe straight after the session. A workable routine looks like this:

1. Read the risk and safety section first, every time, against your memory of the session.
2. Scan every number, date, dose and name. These are the items the model most often garbles.
3. Delete anything the client did not say. Scribes sometimes add a tidy interpretive sentence such as "client is motivated to change" because the pattern fits.
4. Add the clinical formulation yourself. A scribe summarises what was said; it does not know what you concluded.
5. Approve and lock the note, so there is a clear point when the AI draft became your record.

Time it for a fortnight. If review takes longer than writing the note from scratch for a particular client type (say, complex trauma sessions where precise wording matters), that is useful evidence for when to switch the scribe off.

Understanding: read the product information, then ask what is missing

Ahpra says that, at a minimum, you should review the product information for an AI tool: how it was trained and tested, its intended use, and its limitations and the contexts where it should not be used. It also says it is important to understand how data is used to retrain the AI, where data is located and how it is stored.

Vendor pages answer some of this and avoid the rest. A short list of questions to send before you sign up, with what a good answer looks like:

Our tutorial on AI scribe security questions goes deeper on retention and vendor agreements. Save a dated copy of each answer. If the vendor changes its terms next year, you want to know what you were relying on when you started.

Transparency: what to tell clients, in plain words

Ahpra says practitioners should inform clients about their use of AI and consider any concerns raised, and that the level of information depends on how the AI is used. For a tool that records consultations, it says you need to give more information about how it works and how it may affect the client's personal information. The guidance also notes that with public generative AI software, personal information can become public domain.

That last point is why a scribe built for clinical use and a general chatbot are not interchangeable. Pasting a session summary into a free chatbot to "tidy it up" is the scenario that sentence describes.

A client information paragraph you can adapt

How we write session notes

Our practice uses an AI scribe to help draft my clinical notes. With your agreement, it records the audio of our session, turns it into text, and drafts a note that I then check and correct myself. The note I approve is your clinical record.

The tool is [name]. It stores data in [region]. Audio is deleted [when]. It does not use your information to train its models [confirm with vendor].

You can say no at any point, including after we have started. If you say no, I will write notes myself and your care stays exactly the same. Tell me if you have any questions or concerns.

Fill the square brackets only with facts you have confirmed. The sentence "It does not use your information to train its models" is exactly the sort of line that should appear only after the vendor has put it in writing.

Informed consent: make the tool's consent step real

Ahpra says to involve clients in the decision to use AI tools that need their personal data, to obtain informed consent and ideally note the response in the health record. For recording tools it goes further: it warns there may be criminal implications if consent is not obtained before recording, and says the AI transcription software should include an explicit consent requirement as an initial step.

That has two practical consequences.

• Choose a tool with a consent gate. A scribe that starts listening the moment you press a button, with no prompt, puts the whole burden on your memory. If a product has a consent prompt or checkbox, switch it on. If it has none, build your own gate: a script you read aloud before you press record.
• Record the answer. One line in the note ("Client consented to AI scribe for this session, confirmed verbally") is enough to show the step happened.

Three situations that need extra thought

A client who consented at intake and seems unsure today. Consent is not a one-off signature. If a client hesitates or mentions a sensitive topic is coming, offer to pause or stop the recording. Ask again at the start of each session until the routine feels natural, then at least whenever the circumstances change.

Telehealth sessions. If the scribe records both sides of a video call, the consent step matters just as much, and you should check that the client is not being recorded in a way that picks up other people in their room. A short question such as "Is anyone else able to hear us?" covers it.

Couples and family sessions. Every adult in the room needs to agree, not only the person who booked. If one person declines, you write the notes yourself for that session.

For a wider look at easing clients into this, see how to roll out an AI scribe without losing trust.

Ethical and legal: privacy law, governance and your insurance

The guidance lists several obligations from the Boards' codes of conduct that touch AI use. Four matter most for a private psychology practice.

• Confidentiality and privacy. You must check that data is collected, stored, used and disclosed in line with privacy and health-records law. Ahpra specifically says to find out whether client data is also used to train the model, and whether identifiable data finds its way into that training data.
• Bias. Ahpra notes that data and algorithms can carry inherent bias, and asks practitioners to support the health and safety of Aboriginal and Torres Strait Islander people and clients from diverse backgrounds by understanding that risk. For a scribe, the visible form is accuracy: accents, speech differences and names are transcribed less reliably. Spot-check these notes more carefully.
• Governance. Be aware of the arrangements your employer or practice has for overseeing AI, including your role. In a group practice, that means someone owns the approved-tools list. In a solo practice, that someone is you, so write it down.
• Indemnity insurance. Ahpra expects appropriate professional indemnity insurance for all aspects of practice, and says to consult your provider if you are unsure that AI tools are covered.

The insurer email, ready to send

Subject: Professional indemnity cover and AI scribe use

Hello,

I am a registered psychologist (policy [number]). I plan to use an AI scribe that records sessions with client consent and drafts clinical notes, which I review and approve before they enter the record.

Please confirm in writing:
1. Is clinical documentation created this way covered under my policy?
2. Are there conditions (for example consent wording, review steps or approved vendors)?
3. Are there any exclusions relating to AI tools or data breaches by third-party software?

Thank you,
[your name]

File the reply with your practice policies. If the answer is vague, ask again; a "probably" is not cover.

Walking a four-clinician practice through it

Take a four-psychologist private practice with one part-time administrator. Everyone writes their own notes, averaging 20 minutes a day each on documentation after sessions. They are considering a scribe at roughly $40 per clinician per month (illustrative figure; check the current price). Here is how the Ahpra points become a plan.

1. Week 1, decide and ask. The practice manager sends the vendor questions and the insurer email. Nobody records a single session yet.
2. Week 2, write the paperwork. The client information paragraph goes into the intake pack and the website privacy page. A one-page procedure lists the consent script, the review routine and who can switch the scribe on.
3. Week 3, pilot with one clinician. One psychologist uses it with willing clients only, timing review against writing. Any error found is logged, for example "dropped 'not' in risk statement, caught at review".
4. Week 4, review. Check the error log, the time saved, and whether any client declined. Decide which session types stay manual.

Suppose the pilot clinician saves 10 minutes per day after review. Across four clinicians that is about 40 minutes daily, or around 14 hours a month at full rollout. Against a $160 monthly subscription, the arithmetic works only if that time is either billable or genuinely returned to the clinicians' day. If clinicians spend those minutes just finishing earlier, the benefit is wellbeing rather than revenue, which is a legitimate reason but a different business case. Compare it to the costs in our scribe cost comparison.

A before and after: one risk statement

This is an illustration of why the review step is not optional. Suppose a client says near the end of a session: "I haven't had any thoughts of hurting myself this week, but I did the week before."

Scribe draft (illustrative): "Client reports thoughts of hurting self." It collapsed a time-bound, mixed statement into a flat one.

Corrected by the psychologist: "Client denies self-harm thoughts this week; reports such thoughts in the preceding week. Risk reassessed; safety plan reviewed."

The draft is not exactly false, but it would send a future reader down the wrong path. A habit of reading the risk section first catches this kind of error. It is also the clearest example to use when you explain to a new colleague why "the AI wrote it" is not a defence.

When to leave the scribe off

Ahpra's emphasis on judgement means the tool is never mandatory. Reasonable cases for manual notes:

• The client declines, or seems uncomfortable but is too polite to say so.
• The session involves legal proceedings, where the exact status of records matters and you may be asked about retention.
• A crisis session, where the priority is the person in front of you and not the recording.
• Child and adolescent work, where consent involves guardians and the specifics of who agreed to what need extra care. Take advice from your Board's guidance, your insurer and your own supervision before recording minors.

Keep the exceptions in your procedure so the next clinician does not have to guess.

Two more cases: a supervisee and a switched-off tool

A provisional psychologist in a group practice. Suppose a supervisee uses the practice's scribe and the supervisor co-signs notes. The supervisor's co-signature is not a rubber stamp. Agree in the procedure who does the first accuracy check (the author, always) and what the supervisor samples, for example five notes a month chosen at random, with any scribe-caused errors logged. That keeps accountability visible without doubling everyone's workload.

A vendor that adds a feature. Suppose your scribe emails to say it now generates a "suggested treatment plan" from the transcript. Ahpra's point about understanding intended use applies again: a drafting aid that turns speech into text is a different product from one that proposes clinical content. Switch the new feature off until you have read what it does, how it was tested and where its limits are, and decide deliberately whether it belongs in your practice. Add the date and your decision to the evidence folder.

A practical habit for both cases: put a recurring calendar entry every quarter titled "scribe terms and features check". Ten minutes reading the vendor's change notes is far cheaper than discovering a retention change during a complaint.

Keeping the evidence tidy

A compliance folder for this does not need to be large. Six items cover it:

1. The vendor's answers to your questions, dated.
2. Screenshots or PDFs of the vendor's privacy and retention pages on the day you signed up.
3. The insurer's written reply.
4. The client information sheet and the consent wording, with version numbers.
5. Your one-page procedure with the date each clinician read it.
6. The error log from the pilot and a short note of what you changed.

Review the folder every six months and whenever the vendor emails about changed terms. If you are comparing platforms that bundle scribes with scheduling and billing, our practice software comparison lists what each includes. For a plain-English picture of what the tool does in the room, see what an AI scribe is and how it works in a consultation.

Treat this as practical guidance, not legal or regulatory advice. Ahpra and your Board are the authority on what they expect, your insurer is the authority on your cover, and privacy and health-records law can differ between states and territories, so confirm the details that apply to your registration and practice.

Questions psychologists ask before switching a scribe on

Do I need written consent, or is verbal consent enough?

Ahpra asks for informed consent and says to note the client's response in the record where you can. It does not prescribe a form. Many practices use a short written line in the intake pack and a verbal confirmation at the start of each session, recorded in the notes. Check your Board's code and your insurer's wording.

Can I use a free chatbot instead of a scribe product?

Ahpra's guidance warns that with public generative AI software, personal information can become public domain. A purpose-built scribe with clear data handling is easier to defend. If you do use a general chatbot, strip every identifier first and read its privacy settings.

What if a client says no?

Write the notes yourself, as you did before. Record that the client declined and make sure nothing about their treatment changes. A refusal should never cost the client a session slot or a worse service.

Do I have to keep the audio?

That depends on the tool and on your record-keeping duties. Ask the vendor whether audio is stored, for how long and where, and whether you can delete it. Then make sure the final approved note is what your practice treats as the clinical record.

How do I know if my insurer covers AI use?

Ask in writing: 'Does my professional indemnity policy cover clinical records created with the help of an AI scribe that I have reviewed and approved?' Keep the reply with your practice policies. Ahpra says to consult your provider if you are unsure.

Further reads

• AI Scribe for Psychologists: Notes, Reports and Consent Checks — Notes and reports with a scribe, and the consent wording.
• Best HIPAA-Compliant AI for Therapists: Which Vendors Sign a BAA? — Which therapy vendors sign data agreements.
• Best AI Software for Therapists in a Group Practice: Seat Costs — Seat costs when several clinicians share a tool.
• AI Dictation Software for Doctors: Dictation or Ambient Notes? — Dictation versus ambient scribing, compared.
• AI Medical Transcription Services vs Human Transcribers for Small Clinics — When a human transcriber still wins.
• Medicare Codes for Billing Psychology in Australia: AI Claim Checks — A pre-submission claim check for Australian psychologists: match MBS item numbers to times, referrals and session counts, flag mismatches and never upcode.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Ahpra, 'Meeting your professional obligations when using Artificial Intelligence in healthcare' (ahpra.gov.au).