Is It Safe to Use AI for NDIS Reports? What Ahpra Actually Requires
Yes, with conditions, and Ahpra has written them down. Its guidance on using artificial intelligence in healthcare says the practitioner “remains responsible for delivering safe and quality care” and that “practitioners must apply human judgment to any output of AI.” Nothing in it prohibits using an AI tool to draft a report.
What it does is put the whole burden of accuracy on you. If a tool generates text, the practitioner “is responsible for checking the accuracy and relevance of records created using generative AI.” That single sentence is the entire compliance question, and it is a workflow question rather than a technology one.
Contents
What Ahpra says, in one section
Ahpra publishes standing guidance called Meeting your professional obligations when using Artificial Intelligence in healthcare. It is short, it is specific, and it is guidance on meeting obligations you already had rather than a new rulebook. It does not ban any tool and it does not approve any tool.
Its core position fits in two quotes. “Regardless of what technology is used in providing healthcare the practitioner remains responsible for delivering safe and quality care.” And: “Practitioners must apply human judgment to any output of AI.” Everything below is those two sentences applied to a report.
Which regulator is actually yours
One thing almost every article on this topic gets wrong: “allied health” is not one regulated block.
Ahpra and the National Boards register occupational therapists, physiotherapists and psychologists, among other professions. Speech pathologists are not Ahpra-registered: speech pathology is, in Speech Pathology Australia’s own words, a self-regulated profession, with SPA setting the professional standards.
If you are a speech pathologist, the Ahpra guidance is not your regulator speaking, but the obligations it restates, accuracy, consent, confidentiality, professional judgment, are the same ones your own professional standards carry. Read it as good practice rather than as your rulebook.
You remain the author, because you sign it
The accountability question is the one clinicians ask first: if AI drafted it and I sign it, am I still the author?
The guidance answers it without ever using the word author. Responsibility for safe and quality care stays with the practitioner regardless of the technology, and judgment cannot be delegated to a tool. You sign the report, so the clinical claims in it are yours, exactly as they would be if a student, a scribe or a template had produced the first draft.
That is not a technicality to hide behind. It is the reason the review step exists, and it is why a clinician who signs an unread draft has a problem that no tool caused.
You have to check what it wrote
This is the obligation that actually changes a working day: the practitioner “is responsible for checking the accuracy and relevance of records created using generative AI.”
For an NDIS report, checking means something concrete. Every assessment score against the score sheet. Every date against your notes. Every claim about function against what you observed. Every recommendation against what you actually concluded. AI tools, ours included, can be wrong, and the guidance assumes they will sometimes be wrong, which is why the checking obligation exists at all.
The practical implication: a drafting tool saves you the writing, not the reading. Budget review time for every report, and treat any flagged or estimated value as unconfirmed until you have confirmed it.
The participant has to know
Two quotes carry this one. “Health practitioners should inform patients and clients about their use of AI and consider any concerns raised.” And where a tool takes in personal data: “Health practitioners need to involve patients in the decision to use AI tools that require input of their personal patient data”, with informed consent obtained and ideally noted in the record.
How much explanation is proportionate depends on what the tool does. Recording a session is a bigger conversation than drafting from your own written notes. A privacy mode that keeps the participant’s identifying details out of the tool entirely changes the shape of that conversation too, though it does not remove the transparency expectation.
The simple, defensible habit: tell the participant you use a drafting tool, answer their questions, and note the conversation.
Where the data goes is your problem too
The guidance requires “ensuring confidentiality and privacy of your patient/client as required by privacy and health record legislation”, checked in practical terms: that data is “collected, stored used and disclosed in accordance with legal requirements.”
It also expects you to understand the tool you chose. The practitioner “should review the product information about an AI tool including how it’s trained and tested on populations, intended use, and limitations and clinical contexts where it should not be used.” So read the product documentation of any tool you use, ask where data is hosted and whether it is used for training, and keep the answers with your records.
And one question the guidance does not answer for you: your professional indemnity insurer may have a position on AI-assisted documentation. Ask them, in writing, before you rely on an assumption.
What none of this says
It does not say AI-drafted documentation is prohibited. It does not require a disclosure line on the face of the report. It does not distinguish between a dictation tool, an ambient scribe and a drafting tool: the same obligations apply to all of them.
Anything a page implies beyond that is that page’s opinion, not a rule. Articles about how to spot an AI-written report are answering a question the regulator does not ask. The regulator asks whether the clinician checked it.
And that reframing is the honest reassurance. Nothing here is a new obligation invented for AI. You were always responsible for the record, always required to check what went into it, always required to protect the data and tell the client how their information is used. The technology did not create a new duty. It created a new way to breach an old one, and the fix is the same as it always was: review before you sign.
What a compliant workflow looks like
Put your notes in: observations, scores, history, your clinical reasoning. Let the tool draft. Then review every clinical claim against your source material, edit in your own voice, confirm anything the draft flagged as estimated, and sign it as your own work.
That is the workflow the guidance describes, and it is the workflow Carearoo is built around: the draft comes back to you for review, flagged values are marked for your confirmation, a privacy mode keeps the participant’s identifying details out of the draft entirely, and nothing goes anywhere until you have signed it. Your notes and your clients’ information are never used to train AI models.
The clinician stays the author. That is not a limitation of the product. It is the point of it.
Frequently asked questions
Is it against Ahpra rules to use AI for NDIS reports?
No. Ahpra’s guidance does not prohibit AI tools. It restates the obligations that already applied: the practitioner remains responsible for safe and quality care, must apply human judgment to any AI output, and is responsible for checking the accuracy of records created using generative AI.
Do I have to tell the participant I used AI?
Ahpra says practitioners should inform patients and clients about their use of AI and consider any concerns raised, and where a tool takes in personal patient data, involve them in the decision and obtain informed consent, ideally noted in the record.
Am I still the author if AI drafted the report?
You sign it, so the clinical claims in it are yours. Responsibility stays with the practitioner regardless of the technology used, exactly as with a scribe, a student draft or a template.
What if the report contains an error?
The checking obligation exists because AI tools can be wrong. A practitioner who reviewed the draft against their source material and corrected it has met the obligation. One who signed unread has not, and that would be equally true of an unread human-typed draft.
I am a speech pathologist, not Ahpra-registered. Does this apply to me?
Speech pathology is a self-regulated profession in Australia, with Speech Pathology Australia setting professional standards, so the Ahpra guidance is not your regulator speaking. The underlying obligations on accuracy, consent and confidentiality are the same ones your own standards carry.
Does using a compliant tool make me compliant?
No tool can. The obligations sit with the practitioner: checking the output, informing the client, protecting the data. A tool can make the compliant workflow easier, and that is all it can honestly claim.
How this guide was written
Carearoo researches and drafts its guides with AI assistance, working from the primary sources linked in the text, and a person checks every fact against those sources before it is published. Last checked 5 September 2026.
Draft the report. Keep the authorship.
Carearoo drafts NDIS reports from your notes, flags anything estimated for your confirmation, keeps identifying details out with privacy mode, and never trains on your data. You review every line and sign it as your own work.