AI for Speech Pathology in Australia: NDIS Reports and Scheduling

Use AI for the structure and the wording of NDIS reports, and for booking and reminders, but keep identifiable child information out of general tools. Draft from a de-identified outline, let the clinician check every statement against assessment evidence and sign it, and send reminders through the booking system with no clinical detail in them.

Two things shape everything below. A report is a professional opinion that funders and plan managers rely on, so a fluent sentence without evidence behind it is a liability. And a child's health information is the most sensitive data a clinic holds, so the privacy steps come before the prompts, not after them.

Follow me on Instagram@sagnikteaches

Start from the rules that apply to your clinic

Three sources cover most of what a clinic needs to check. None of them replaces advice from your professional body, your insurer or a privacy adviser, and all can change, so read the current versions.

Connect on LinkedInSagnik Bhattacharya

• The Australian Privacy Principles (APPs) under the Privacy Act 1988. The Office of the Australian Information Commissioner (OAIC) published guidance on privacy and commercially available AI products in October 2024. It advises, as best practice, not entering personal information, and particularly sensitive information, into publicly available generative AI tools, and it notes that typing personal information into a chatbot can be a disclosure to the tool's owner. It points to APP 6 (use and disclosure), APP 8 (cross-border disclosure) and APP 11 (security), with APPs 3, 5 and 10 also relevant to collection, notice and accuracy. Health information counts as sensitive information.
• The NDIS Quality and Safeguards Commission. Its February 2026 position statement on using AI in developing behaviour support plans says the legislative framework does not explicitly prohibit AI, and that the Commission does not endorse or approve AI tools. It expects that, if a provider uses AI, all information is appropriately de-identified and no personal information of participants is disclosed to AI systems. Its own example: disclosing sensitive personal information to a third-party platform such as ChatGPT without appropriate safeguards may breach the privacy requirement in the NDIS Code of Conduct. The statement is written for behaviour support plans, but the risks it lists (disclosure to third parties, processing overseas, inaccurate content, decisions without human oversight or clinical judgement) apply just as much to a speech pathology report. Check it and any newer material.
• Speech Pathology Australia. The association has published Artificial Intelligence (AI) in Speech Pathology: Ethical Considerations, which sets out the ethical principles and obligations of members who use AI in practice. Its resource page would not load when we checked, so read the current version on its site; some related material may be limited to members.

The practical consequence is a simple rule that this tutorial follows throughout: type the structure, not the participant.

Subscribe on YouTube@sagnikteaches

NDIS report drafting that starts from a de-identified outline

Funding-related reports tend to follow a pattern: background, assessment methods, findings, functional impact, recommendations with justification, and a request for supports. The sections are stable. The content is where your judgement lives. AI is useful for the first and must not be the author of the second.

The workflow in five steps:

1. Write your own findings in shorthand. In your clinical system, in your own words, as you normally would.
2. De-identify a working copy. Replace the name with [participant], the school with [school], dates with [date], and remove anything unusual that could identify the person. Delete scores and the names of any copyrighted tests; you will add them back in your own document.
3. Ask the assistant to structure and tighten the wording in the order your template uses.
4. Return to your system and reinsert the real details, test names and results yourself.
5. Review against the evidence and sign.

A prompt for step three, illustrative:

You are helping a speech pathologist draft a report section. Use ONLY the text between the
markers. Do not add findings, causes, recommendations or diagnoses. Rewrite it in clear,
professional language under these headings: Communication strengths; Areas of need;
Impact on daily life; Recommended supports with rationale. Keep every placeholder in square
brackets exactly as written. If anything I have written is unclear or unsupported, list it
under "Questions for the clinician" instead of resolving it yourself.
START
[participant] communicates mostly in 2-3 word phrases. Follows 1-step instructions in
familiar routines, struggles with 2-step. Uses gesture and pointing well. Frustration at
mealtimes when not understood. Parents report fewer outbursts when picture cards are used.
Recommend weekly therapy for [period] with a home programme, plus staff training for
[school] on using visual supports.
END

An illustrative output for the first heading: "Communication strengths: [participant] uses gesture and pointing effectively to express needs. [participant] follows one-step instructions in familiar routines." Then the second: "Areas of need: [participant] currently communicates mainly in two- to three-word phrases and finds two-step instructions difficult." That is a fair restatement. Look at what it does under recommended supports. A common failure is that the assistant inflates the justification: "Without intervention, [participant] is at risk of social isolation and reduced independence". You did not write that. Delete it, or replace it with evidence from your assessment, because funders look for reasoning that links evidence to the support, and an unevidenced claim weakens the request.

Where AI drafts go wrong in a report

• Invented causes. "Due to low muscle tone" appears from nowhere. Delete anything you did not supply.
• Stronger language than the evidence. "Severely impaired" for "difficulty". Match your assessment.
• Outcome promises. "Will achieve age-appropriate language." Remove; recommend, do not predict.
• Blended participants. When you draft several reports in one chat, details drift between them. Use one fresh chat per report.
• A fluent tone that hides gaps. If your notes lacked a baseline measure, the draft will not tell you. Ask it to list missing information, and decide.

What to swap out before you paste, with a filled-in example

De-identification is more than deleting a name. Work down this list on the working copy, and keep the key in your own system so you can reverse it.

Before, with capitals standing in for the real name and school: "FIRSTNAME, aged 5 years 3 months, attends the Year 1 class at SCHOOLNAME and scored 78 on the receptive subtest." After: "[participant], aged [age band], attends [school] and scored [test result] on the receptive subtest." The second version still gives the assistant everything it needs to write a sentence about receptive language, and nothing that identifies a child. In a small community, a combination of age, school and an unusual circumstance can identify a child even when the name is gone, so when in doubt, generalise.

Turning shorthand into a paragraph, then fixing it

Say your notes read: "Mealtimes hard. Points, no words. Dad says screams if not understood. Better with cards." An assistant told to write in report language might return: "[participant] relies on pointing at mealtimes and becomes distressed when a message is not understood. Picture cards appear to reduce this distress." Read it closely. "Appear to reduce" is fair, because the parent reported it. If the assistant had written "picture cards are effective", you would change it back, because one parent's report is not a measured result. That is the habit to build: check the strength of each claim against its source.

The review step: a sign-off checklist

Whatever tool helped, the same checklist applies before anything is sent. A report should not leave the clinic until each box is ticked by the clinician.

1. Every finding matches my assessment notes, test results and observations.
2. Every recommendation has a stated reason linked to a finding.
3. All placeholders are replaced, and no other participant's details are present.
4. Test names, scores and dates were entered by me, from the source.
5. Language is consistent with the level of evidence.
6. The report is in my voice, and I would be comfortable explaining each sentence to the family and a reviewer.
7. The file is saved in the clinical system with the date, version and reviewer.

For a worked comparison of this kind of report in a neighbouring profession, see the NDIS occupational therapy report template. For the privacy picture across providers, what is safe under the privacy principles is the companion piece.

A report, timed: where the hours go

Say a clinic writes eight reports a month and each takes about 2 hours 30 minutes from notes to sign-off, so 20 hours. The time splits roughly like this: 20 minutes organising notes, 70 minutes drafting and wording, 30 minutes for formatting and consistency, 30 minutes for review and sign-off. AI helps with the second and third blocks. If it cuts drafting from 70 to 35 minutes and formatting from 30 to 10, the saving is 55 minutes a report, or about 7 hours a month. The extra step of de-identifying and reinserting adds perhaps 15 minutes, so the net is about 40 minutes a report, or 5 hours 20 minutes a month.

These are illustrative figures. The honest takeaway is that the gain is real but modest, and it comes from structure and wording, not from skipping the review. If a clinic finds itself saving more than that, check that review is not being shortened.

Consent and notice for children's information

Children's health information raises the stakes. Who consents, and to what, depends on the child's age and capacity, the nature of the service and your own policies, so settle the specifics with a privacy adviser. The practical steps below help whatever the answer is.

• Say it in the privacy notice at intake. A short paragraph: whether you use AI tools, for what purposes (structuring and wording of documents, scheduling), whether identifiable information is ever entered, and who to ask.
• Update the privacy policy. The OAIC advises updating policies and notices with clear information about AI use, and a secondary use is more likely to be reasonably expected when it was outlined at collection.
• Ask for specific agreement for anything beyond ordinary care. If you ever want to use identifiable information in an AI tool, get express, informed agreement and record it. Better, do not.
• Know where data goes. APP 8 concerns cross-border disclosure, and many AI tools process data overseas unless the vendor offers local processing; the NDIS Commission lists overseas processing among its AI risks. Ask vendors where data is stored and processed, and write down the answer.
• Plan for breaches. Know your obligations under the notifiable data breaches scheme and have a one-page incident step in your policy.

An illustrative intake paragraph: "We may use computer-based writing tools to help structure and word our reports and letters. We do not put your child's name, details or results into these tools. A clinician checks and signs everything we send. If you have questions, please ask." Only write that if it is true of your clinic.

Scheduling and reminders: the low-risk use

This is where AI pays quickest and risks least. Most booking systems for allied health clinics can send reminders on their own. AI helps with the design: the wording, the rhythm and the rules around cancellations.

Reminder wording without clinical detail

Ask for a short message that confirms the time, place and clinic name, offers a way to change the booking and says nothing about why the person is attending.

Hi [first name], a reminder that you have an appointment at [clinic name] on [day] at [time]. Need to change it? Reply or call [clinic number] at least [x] hours before. See you then!

A subject line or message that says "speech therapy for [child's name]" is the kind of detail that can leak on a lock screen. Keep it generic.

Cancellation and waiting-list rules

Draft the rules yourself, then ask the assistant to write them in plain language. Decide: how much notice you need, what happens on a late cancellation, how you fill a gap, and how you offer a place from the waiting list. Many NDIS clinics have a cancellation policy in the service agreement, so make sure the plain-language version matches it exactly. Illustrative rule: "Contact the first two families on the list by phone within an hour of a cancellation, then offer by message to the third." Test it on a quiet week before relying on it.

Rota and caseload sums

A scheduling prompt can model a week without any personal data. Give it session lengths, clinician hours and travel time, and ask where gaps arise. Say a clinician works 30 client-facing hours a week with 8 hours of report writing and admin. If sessions are 45 minutes with 15 minutes of notes and travel, each slot takes an hour, so the clinician holds 30 slots. If reports take 2.5 hours each and she writes 2 a week, that is 5 hours, leaving 3 for the rest of the admin. A model like that shows quickly whether you can take on more participants or need to protect report time.

When a whole assessment summary went into a public chatbot

An illustrative case. A busy clinician, short of time before a plan review, pasted a whole assessment summary into a public chat tool, asked for a "funding-ready report" and received three polished pages. The pages looked excellent. Two things were wrong. The summary had included the child's full name, school and test scores, which went to a third party. On the privacy regulator's guidance that is a disclosure the clinic would struggle to justify, and it is close to the very example the NDIS Commission gives of a possible Code of Conduct breach. And the report contained a paragraph on "risk of regression" that the clinician had never assessed.

The clinic's response was to adopt the de-identified outline method above, add a line to its privacy notice, and make a second clinician check a sample of reports each month. The "funding-ready" instruction was banned from prompts: asking for persuasion invites the assistant to add claims. The prompt now asks for accuracy to the notes, and unsupported items go on a question list.

A reminder sequence that reduces missed sessions

For children's therapy, a reminder works best when it reaches the person who manages the diary, which is often a parent, carer or plan nominee. Ask the assistant for a three-step sequence and set the timings yourself.

1. Booking confirmation: date, time, place, clinic name, how to change it.
2. Two days before: the same details and a link or number to rebook.
3. Morning of the session: one line with the time and place.

If your system supports two-way replies, add a keyword such as "C" to cancel so the slot can be offered to the waiting list quickly. Check your messaging rules and consent for sending text messages with your software provider. Count the result after eight weeks: missed sessions per 100 booked, before and after. If the rate does not move, the problem is probably the cancellation policy or the time of day, not the wording.

A short policy your clinic can adopt this week

An illustrative one-page outline, to adapt with your own adviser:

1. Approved tools: list them. Nothing else is used for clinic work.
2. What may go in: structure, templates, generic wording, scheduling rules.
3. What may not: names, dates of birth, NDIS numbers, addresses, results, recordings, photos, notes.
4. Review: a clinician checks and signs every document, and enters all test details themselves.
5. Records: log date, task and reviewer for AI-assisted documents.
6. Incidents: tell the privacy lead the same day; delete the chat; follow the breach steps.
7. Review the policy every six months, or when tools or rules change.

A policy like this is short enough to be used. For tailoring a full set of provider policies, see NDIS provider policies and procedures. For a wider view of what to hand to AI first, the AI guide is the hub.

Signs the report and reminder changes are paying off

• Time a report from notes to sign-off before and after, for three reports each.
• Count the corrections you make at review. A rising count means the prompt or the notes need work.
• Check the no-show and late-cancellation rate for three months after changing reminders.
• Once a quarter, ask a colleague to audit five documents against the sign-off checklist.
• Reread the OAIC and association guidance twice a year, because both can change.

Questions clinics ask about AI and NDIS paperwork

Can I paste a participant's assessment notes into ChatGPT?

Not identifiable notes. The privacy regulator advises, as best practice, against entering personal and especially sensitive information into publicly available generative AI tools, and the NDIS Commission expects participant information to be de-identified before any AI use. Draft from a de-identified outline, and check your insurer and professional body too.

Does the clinician still sign the report?

Yes. The report is your professional opinion. AI can structure a draft, but you must be able to stand behind every sentence, including those it wrote. If a statement is not supported by your assessment, remove it, however well it reads.

Do I need parental consent for a child's data?

Consent and notice requirements depend on the child, the purpose and your own policies, so settle them with your privacy officer. In practice, many clinics explain AI use in the privacy notice at intake and ask the parent or guardian to agree to any use that goes beyond ordinary care.

Can AI handle scheduling and reminders?

Yes, scheduling and reminder wording is a good low-risk use, especially when the booking system sends the messages itself. Keep reminders free of clinical detail: a time, a place, a name of the clinic and how to change the booking, nothing about why they are coming.

What should a clinic check from Speech Pathology Australia?

Read the association's current guidance, Artificial Intelligence (AI) in Speech Pathology: Ethical Considerations, which sets out ethical principles and obligations for members who use AI in practice. Some related material may be member-only, so check what applies to your membership, and note the date you reviewed it in your practice policy.

Further reads

• How to Write NDIS Progress Notes: An AI-Assisted Audit Checklist — An audit checklist for AI-assisted progress notes.
• NDIS Software for Small Providers: Per-User Costs and What You Get — Per-user software costs for small providers.
• AI in Speech and Language Therapy Clinics: Booking and Waiting Lists — Booking and waiting lists for therapy clinics.
• NDIS Service Agreement Template: Plain-English Drafting With AI — Plain-English service agreements, including cancellation terms.
• NDIS Audit Checklist: Build the Evidence Folder With AI — Building the evidence folder an audit asks for.
• AI for Speech Language Pathologists: Home-Practice Handouts — Parent handouts with the same privacy steps.
• AI for Speech Therapy Reports: Draft, De-Identify, Review — De-identification and review steps first, then AI drafting of an evaluation report from assessment notes, with a worked example and sign-off list.
• AI in Occupational Therapy: Funding Letters and Parent Updates — A private-practice OT workflow for evidence-led funding letters and parent summaries that the therapist signs off.
• How to Get Clients as an NDIS Provider: AI for Enquiries, Referrers — A workflow for new NDIS providers: directory profiles, support-coordinator outreach and AI-drafted enquiry replies that a person always checks.
• Can AI Write NDIS Behaviour Support Plans? What the Commission Says — What the NDIS Commission's February 2026 statement says about AI in behaviour support plans, turned into a workflow a provider can follow.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Office of the Australian Information Commissioner, Guidance on privacy and the use of commercially available AI products (October 2024) and its Notifiable Data Breaches scheme page; NDIS Quality and Safeguards Commission, Position statement: Use of artificial intelligence in the development of behaviour support plans (February 2026); Speech Pathology Australia, Artificial Intelligence (AI) in Speech Pathology: Ethical Considerations (title as listed in ASHA's AI resources; the association's page would not load); OpenAI business data and Anthropic privacy centre pages. All checked 4 November 2026.