An occupational therapist can draft NDIS reports with AI and stay accountable by drafting from de-identified notes and measures, using a fixed set of headings, checking every statement against their own assessment data, and signing only what they can defend. The tool writes words; the therapist owns the findings, the recommendations and the signature.
Search results for this topic are mostly template sellers offering a blank form. A blank form fixes the headings, which is the easy part. The hard part is knowing what each section must prove and catching the sentence a drafting tool adds that you never observed, because that sentence goes out under your registration, not the tool's.
Accountability comes first
Whatever tool you use, the report is yours. Your registration body, your insurer and the funder will all look to the named therapist as the author of the assessment and the recommendations. Three consequences follow. You must be able to trace each recommendation to an observation or measure you made. You must not include anything you did not assess or observe, however sensible it sounds. And you should keep your own record of how the report was prepared, for example the notes you worked from and the changes you made, in case the funder, participant or your professional body asks.
The risk with a fluent drafting tool is not obvious nonsense. It is plausible filler: a sentence about "reduced endurance" that you never measured, a recommendation that follows from the usual pattern rather than this participant. Plausible filler is harder to catch than error, and it is exactly what an accountable author must remove.
The template: headings and what each must prove
Use the funder's current requirements and your service's format where they exist, and adapt the headings below. The right-hand column is the test for each section.
| Heading | What goes in it | What it must prove |
|---|---|---|
| Participant and referral details | Name, identifiers, referrer, reason for report, date range of assessment | Who and why; filled by you, not the tool |
| Background (brief) | Relevant diagnosis and history in two or three lines, from documents you hold | Context only, no speculation |
| Participant goals | Their goals in their words where possible | The report is about their priorities |
| Assessment methods | Tools used, observations, interviews, settings, dates, who was present | The evidence is genuine and traceable |
| Findings by area | Function in the relevant domains, with measure or observation for each | Each claim has a source |
| Impact on daily life | How the findings limit participation, with concrete examples | Functional impact, not diagnosis |
| Supports provided and progress | For a plan reassessment report: supports delivered in the period, baseline and current measures, barriers met | Change is measured, not asserted |
| Recommendations | Supports, equipment, therapy, with frequency and duration | Each linked to a finding and a goal |
| Rationale and alternatives | Why this, why now, what else was considered, cost where required | A funder can see the reasoning |
| Risks and review | Risks of not acting; when you will review progress | Accountability over time |
| Summary for the funder | A half-page plain summary of need and recommendation | A decision-maker can act on one page |
| Declaration | Name, registration, signature, date | Yours alone |
The funder's decision turns on whether the supports are reasonable and necessary for the participant, so the rationale section carries the most weight. Read the funder's current guidance on what it wants in a supporting report before you finalise your format, because the wording of what is asked for changes.
Checking the headings against the funder's report guide
The funder's guide to report writing lists what an allied health plan reassessment report should do: outline your service details; summarise the goals you and the participant are working towards; summarise functional capacity at the beginning of the plan period; describe the assessment tools used to quantify it; summarise the supports provided in the reporting period; describe progress towards the goals, with the assessment measures that show it; describe barriers and risks and how they are being addressed; recommend supports for the next period; and justify any recommendation for more or different supports, with the proposed outcomes and any risks or impacts on other supports. Tick each item off against a heading in your template before you draft.
Two of those items catch AI drafts out. Functional capacity at the start of the plan period needs a baseline figure from your own records, which no tool can know, so a draft that describes it without a number has invented the comparison. And the justification for more or different supports needs the proposed outcome, not a restatement of the need. The same guide says to share the finished report with the participant, who can pass it to the funder, and that with their consent you can upload it through the provider portals. Write every line knowing the participant will read it.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowStep 1: prepare de-identified input
Create a working document with a reference code in place of the participant's name (for example "P-0417"). Write your own raw notes: assessment tool results, observations in point form, goals as stated, environment details, what worked in trials. Remove date of birth, address, funding number, names of family, carers and workers (use "carer", "support worker"), and employer or school names. Keep ages as a band if the age is clinically relevant.
This preparation takes 10 to 20 minutes and is the main privacy control. It also forces you to decide what the evidence actually is, which is useful in itself. Use a business plan with data terms you have reviewed, not a free personal account; on business plans ChatGPT Business, Claude Team, Microsoft 365 Copilot and Gemini in Workspace do not train on business content by default.
Step 2: the drafting prompt
The prompt does three things: fixes the headings, forbids invention, and demands a marker wherever the notes do not support a statement. Illustrative version:
You are helping an occupational therapist draft a report. Use ONLY the
notes below. Do not add findings, measures, diagnoses, equipment or
frequencies that are not in the notes. Use these headings in this order:
Goals; Assessment methods; Findings by area; Impact on daily life;
Recommendations; Rationale. After each finding write the source in
square brackets, e.g. [Assessment tool, item 4] or [Observation, kitchen
trial]. Where a section needs information the notes do not contain,
write [THERAPIST TO COMPLETE] instead of guessing. Plain English, short
sentences, no jargon without a short explanation. Use the
spelling of the notes.
NOTES (reference P-0417):
[your de-identified notes]The source-in-brackets instruction is the most important line. It makes every sentence checkable, and it exposes the sentences with no source, which are the ones you must remove or support.
Worked example: a first draft and what the therapist changes
Take an illustrative adult participant with a physical disability who wants to prepare simple meals independently. The therapist's notes cover a home visit kitchen trial, a standardised measure of upper-limb function, a fatigue observation and a discussion of equipment options with the participant. The goals note says: "make my own breakfast and lunch on weekdays".
The tool returns a draft. The findings section reads, in part:
"P-0417 can lift a lightweight kettle with two hands but cannot lift a full saucepan from the hob [Observation, kitchen trial]. Grip strength is reduced on the right [Standardised measure, item 3]. Fatigue is a significant factor limiting prolonged standing [Therapist note]. The participant has poor safety awareness around hot surfaces."
The first three sentences have sources the therapist can check against the notes. The fourth has none, and the notes say nothing about safety awareness; the tool has filled in a common pattern. The therapist deletes it. Had the sentence stayed, a report would have carried an unsupported claim about the participant's judgement into a funding decision, with the therapist's name beneath it.
The recommendations section proposes "a height-adjustable bench and kettle tipper". The notes mention a bench but not a kettle tipper; the therapist checks the trial record, finds a tipper was trialled and worked, and keeps it, adding the observation to the source bracket. A second recommendation, "an assessment by a physiotherapist", has no basis in the notes. It is removed.
In a draft like this one, most sentences survive a wording tidy and a handful need removing or sourcing. Your time goes into the review, not the typing, and the review is the part that makes you accountable.
Step 3: build the evidence table yourself
Before you sign, assemble a one-page table that lists each recommendation, the finding it rests on, the measure or observation and the goal it serves. You can ask AI to produce a first version from the draft, then check it. Illustrative rows:
| Recommendation | Finding | Evidence | Goal |
|---|---|---|---|
| Height-adjustable bench | Cannot work at standard height for more than a few minutes | Kitchen trial, observation, date | Make own breakfast and lunch |
| Kettle tipper | Cannot lift full kettle with one hand; two-hand lift unsafe at hob | Trial of tipper, outcome noted | Make own hot drinks |
| Six sessions of meal-preparation training | Needs strategy practice for energy conservation | Fatigue observation; participant report | Weekday meals independently |
Any recommendation without a row filled is not ready. That is a more reliable test than rereading the report for tone. It also gives you a ready answer if a reviewer asks how a particular item was justified.
Step 4: the funder-ready summary
Decision-makers often read the summary first and sometimes only the summary. After the full report is reviewed, ask the tool for a half-page plain-English summary from the finished text, not the raw notes, so it cannot introduce anything new. Illustrative prompt: "Summarise the report below in about 180 words for a non-clinical reader. Use only what is in the text. Start with what the participant wants to do, then what limits it, then what is recommended and why. Do not add anything." Then check it sentence by sentence against the report. A summary that says "urgently needs" when the report says "would benefit from" has changed the meaning, and you should fix it.
The review checklist, filled in as you go
Use this on every AI-assisted report. It takes 15 to 30 minutes, which is much less than writing from scratch, and it is the part you must not skip.
- Every finding has a source. Tick each bracketed source against your notes. Delete findings without one.
- Nothing assessed that you did not assess. Scan for measures, tools or test names you did not use.
- Numbers match. Scores, dates, frequencies and durations match your records exactly.
- Quotes are real. Anything attributed to the participant or family is in your notes.
- Recommendations link to goals. Each has a finding, evidence and a goal in the evidence table.
- Hedges are honest. "Appears to", "may" and "would benefit" reflect your confidence, not the tool's.
- Cost and quantity are yours. Quotes, equipment items and session numbers come from your own information.
- The tone is the participant's report, not a template. A reviewer should not be able to swap the name and reuse the text.
- De-identified references replaced correctly. The right name is in the right place and no placeholder is left in the text.
- You would be comfortable defending each line. If not, change it.
Ways AI-drafted reports go wrong
Boilerplate drift. Reports for different participants start to sound the same: the same opening, the same risk line, the same closing sentence. Reviewers notice. Edit for the individual, and add a detail only this participant has.
Over-confident prognosis. "Will significantly improve" appears where you wrote "may benefit". The tool leans positive. Replace the claim with your judgement and the evidence for it.
Diagnosis creep. The draft attributes a limitation to a condition the notes never name. Report function, not guesses about causes, unless you are qualified and have grounds to say so.
Mixed-up participants. A therapist drafts two reports in one chat session and details cross over. Use a fresh conversation for each report and delete the chat once the final is saved in your records, following your retention rules.
Funding language that overreaches. The draft uses the funder's phrases such as "reasonable and necessary" as decoration, without showing why. Make the rationale section do the work with specifics.
A second example: a child's report with a school component
Take an illustrative report for a child who struggles with handwriting stamina and classroom organisation, where the goals come from the parent and the teacher. The therapist's notes include a handwriting speed sample, a classroom observation, a parent interview and a trial of a pencil grip and a sloped board. The draft from the tool comes back fluent, but the review turns up two problems that are different in kind from the first example.
The first is attribution. The draft says "the teacher reports the child avoids written tasks". The notes record that the parent said the teacher mentioned it. That is second-hand, and the report must say so, or the therapist should ask the teacher directly. The second is a recommendation that stretches the evidence: "twelve weekly sessions" appears because that is a common package, but the notes support a short block of six, followed by a review. The therapist rewrites it with the rationale: six sessions to teach and embed strategies, then a review of whether handwriting stamina has improved on the same sample.
These examples show the same lesson from two angles. A drafting tool is good at structure and flow, and it is not a witness. Everything the report says about what someone did, said or can do must come from you.
A recommendation rationale: before and after
The rationale is where reports most often read as generic. An illustrative before and after for the same equipment recommendation shows the difference.
Before: "A height-adjustable bench is recommended to support independence and safety in the kitchen and is considered reasonable and necessary."
After: "In the kitchen trial the participant could work at a standard-height bench for about four minutes before needing to stop because of upper-limb fatigue, and could not reach the hob safely when seated at the lower bench height. At an adjusted height of [measure] she completed the full breakfast task in twelve minutes. A height-adjustable bench lets her prepare weekday meals without help, which is her stated goal. A fixed lower bench was considered and rejected because it would not suit other household members. Review in six months."
The after-version has a measure, a comparison, a link to the goal, an alternative considered and a review point. None of those came from a template. If the tool suggests the before-version, treat that as a prompt to ask yourself the questions in the after-version, and fill in the answers from your own records. You can also ask the tool to list what is missing from a rationale, not to rewrite it: "list the facts a funder would want that this paragraph does not contain." The answers are questions for you to answer.
What to ask your software vendor and your insurer
If a practice management system or scribe product offers report drafting, ask the vendor where data is processed and stored, whether inputs are used to train models, how long they are kept, whether you can delete them and whether staff log in individually. Read the answer in the vendor's own documentation, not a sales call. Then ask your insurer, in writing, whether using a drafting tool changes your professional indemnity cover. The reply goes in your policy file. For the buying questions in more detail, see the buyer's checklist for scribes, and for the wider question of safe practice, the tutorial on drafting reports safely.
Three more uses that stay within the template
Plain-language participant copy. After the report is final, ask for an easy-read summary for the participant, using the final text only, and check it with them. It helps them see what you have recommended and why.
Consistency check across a long report. Paste the final text with identifiers removed and ask for contradictions: a score that appears twice with different numbers, a date range that does not match, a recommendation that is not in the summary. This is a proofreader's job and a good one for AI.
Response to a funder query. If the funder asks for more information, draft the reply from your original notes and the report, and run the same checklist. For letters of this kind, the funding letters tutorial has more wording.
Keeping records of how the report was prepared
Keep, with the participant's file, the notes you worked from, the final report and a short line in the record: "Draft prepared with an AI tool from de-identified notes; reviewed and edited by [therapist] on [date]". Do not store the chat itself as the record. The funder's record keeping rules require complete and accurate records of supports delivered, including case notes, so the clinical record remains your own notes and reports, in your own system. If your service has an AI use policy, put the process in it; the monthly checks in the documentation audit tutorial will then pick it up.
When to write the report without a tool
Some reports are better written by hand: a complex case with significant risk, a report where the participant has specifically asked you not to use AI, one where you cannot properly de-identify the notes, or an urgent report where the review would take longer than writing. A therapist who can say "I did not use a tool for this one because the notes could not be de-identified safely" is applying the same professional judgement the rest of the report relies on.
If you would like help setting up a drafting workflow and checklist that suits your practice, that is what my AI implementation consultation covers; get in touch. Check your professional body's current guidance on AI and the funder's current requirements for reports, and ask your insurer if you are unsure what is covered.
Accountability questions for therapists
Who is responsible if an AI-drafted report contains an error?
The therapist whose name and registration are on it. Using a drafting tool does not move responsibility to the tool or the vendor. Read every sentence against your own assessment data and session notes before signing, and keep a record of what you changed so you can show how the final wording was reached.
Do I have to tell the participant I used AI?
Check your professional body's current guidance and your service's policy; many therapists choose to say so in their consent process and service agreement. At minimum, be ready to explain honestly how the report was prepared if asked, and never use a tool that your consent wording does not cover.
Can I paste assessment scores into the tool?
Scores and observations can be entered without identifiers, using a reference code, and only on a business plan whose data terms you have reviewed. Keep the name, date of birth, address and funding number out. Re-attach identifiers in your own document at the end.
Will a funder reject a report that looks AI-written?
Funders assess whether a report is clear, evidenced and ties the recommendation to the participant's goals and needs, not how it was drafted. Generic, repetitive or unsupported reports are the risk. Specific measures, real observations and a clear rationale protect you.
Further reads
- How to Write NDIS Progress Notes: An AI-Assisted Audit Checklist — Checking progress notes that feed your reports.
- How Can AI Help an Occupational Therapy Practice? Seven Admin Jobs — Seven admin jobs in an OT practice.
- ChatGPT for Occupational Therapy Goals: Drafting Without Client Data — Drafting goals without client data.
- How Can AI Help Physiotherapy Clinics Write Referrer Updates? — Referrer updates in a physio clinic, drafted and checked.
- AI for Speech Therapy Reports: Draft, De-Identify, Review — Draft, de-identify and review method for another allied health report.
- AI for NDIS Providers: What Is Safe Under the Privacy Principles? — What is safe under the privacy principles.
- NDIS Providers Automation Assessment Checklist: Which Jobs for AI First? — Score five admin jobs on time, repetition, checkability and harm if wrong, then pilot the winner for four weeks with clear stop rules.
- AI Accounting for NDIS Providers: Categorising Income and Claims — How an NDIS provider can use AI to categorise bank-feed income, match payments to claims and spot unpaid ones, with privacy rules and an accountant's review.
- AI for Speech Pathology in Australia: NDIS Reports and Scheduling — NDIS report drafting with clinician review, scheduling and reminders, and consent for children's data under the Australian Privacy Principles.
- 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: NDIS provider record keeping requirements (ndis.gov.au, current as of August 2026), the NDIS guide to report writing for allied health providers (fetched November 2026), vendor data-use pages for business AI plans; headings drawn from common allied health report practice, to be checked against the funder's current guidance.