NDIS Incident Report Template: Can AI Draft It? What a Manager Checks

Yes, with limits. A support worker or manager can use AI to tidy the wording of a factual incident account, but only with participant details removed or inside a system your organisation has approved. AI must never decide whether an incident is reportable; a manager checks the category and the 24-hour and 5-business-day clocks.

The reason is the clock. Under the incident-management rules, the most serious incidents must reach the NDIS Commission within 24 hours of the provider becoming aware of them, and a tool that softens a description or guesses a category can push a report into the wrong lane. The rules on the Federal Register of Legislation were still the 2018 version in November 2026, but the Commission's guidance and reporting portal change more often, so check them where flagged below.

Follow me on Instagram@sagnikteaches

The template: fields that every report needs

An incident report has one job: to let a person who was not there understand what happened, what was done and what comes next. Use the same fields every time so a manager can scan them.

Connect on LinkedInSagnik Bhattacharya

The rules say what a 24-hour notification to the regulator contains, and the fields above map to it: the provider's name and contact details, a description of the incident, the impact on or harm to the participant, the immediate actions taken (including whether police or another body were told), the notifier's details, the time, date and place if known, and the names and contact details of those involved. Witnesses and further proposed actions follow within five business days. Use the Commission's own form or portal for the notification itself; the template is your internal record that makes it easy to fill in.

Subscribe on YouTube@sagnikteaches

Factual tone: the rule that makes a report usable

Most weak incident reports fail on tone, not on missing fields. Three habits fix most of it.

• Observation, not opinion. "[Participant] was shouting and threw a cup at the wall" is an observation. "[Participant] became aggressive and was being difficult" is a judgement. Write what you saw and heard.
• No diagnosis or cause. "There was a red mark about 3 cm long on the left forearm" rather than "she was injured by the other resident".
• Sequence and times. Use the clock: "At about 2.15pm", "Two minutes later", "At 2.30pm the manager was phoned".

An illustrative before and after. Before: "[first name] got upset in the kitchen and was aggressive towards staff, so we had to restrain him and he got hurt." After: "At about 2.15pm, in the kitchen, [participant] raised his voice and said 'leave me alone'. A worker moved the knife block out of reach. At 2.20pm [participant] pushed past the worker and fell against the bench. [Participant] had a red mark on his right elbow. First aid was given at 2.25pm. The manager was phoned at 2.30pm." The second version is longer, and it lets a manager decide what category it is. Notice that it also avoids the word "restrain", which would need a very different response, unless a restraint actually happened.

Where AI helps in drafting

AI is good at three things in this job, all about language. It can turn a rambling voice-note transcript into the template's headings. It can flag opinion words and suggest observation wording. And it can check that every field has something in it, listing gaps such as "no time given for when the manager was called". None of this needs the participant's name.

A routine that respects privacy: the worker writes or dictates the account with initials or role labels ("Participant A", "Worker 1"). Before anything leaves the system, the worker removes any real names, addresses, dates of birth and funding numbers. The AI sees only the cleaned text. The manager adds real identifiers back in the incident management system after review. If your organisation has a tool inside a contracted system with a proper data agreement, the same workflow can run there with fewer redactions, but that is a procurement decision, not a worker's choice.

Below is a worker's account of an incident, with names removed.
Rewrite it into the following fields: Date and time; Location; People
involved (roles only); What happened in time order; Observed harm;
Immediate actions; Witnesses. Rules: use only facts in the account;
do not add causes, diagnoses or opinions; keep any direct quotes in
quotation marks; if a field has no information write [MISSING];
list every opinion word you removed at the end.

Account: [paste cleaned account]

A realistic sample output (illustrative) is a tidy set of fields with two [MISSING] flags, say the time first aid was given and whether the participant's family was told, and a closing list: "Removed: 'aggressive', 'refused to cooperate'." The manager now has gaps to chase with the worker, which is the whole point. The fix a manager makes to the AI output is usually to restore a quote the model paraphrased. A paraphrased quote is no longer a quote, so check each one against the worker's original.

Where AI must not go

The same tool that tidies wording can make a report worse, and some uses are off limits.

• Reportability. Do not ask AI whether an incident is reportable. The categories come from the rules, they turn on facts and on definitions your manager has to apply, and the notification duty sits with the provider. Using AI's answer as the decision is a risk; so is letting it nudge the manager towards "not reportable".
• Filling gaps. A model asked to "complete" a report will invent a plausible time or a reason. The "write [MISSING]" instruction is not optional.
• Softening. Models tend to polish. "Hit" becomes "made contact". A report that understates harm is a safeguarding failure, not a style choice.
• Identifying details in public tools. A participant's name, diagnosis, address or funding details do not belong in a general chatbot.
• Replacing the worker's account. The original statement, in the worker's words, is evidence. Keep it unedited.

The tutorial on daycare incident reports with AI covers the same factual-narrative method in another setting, and whether AI should write incident reports at all sets out the accuracy limits more generally.

The manager's check: is it reportable, and when did the clock start?

This is the part no template does for you. The incident-management rules distinguish incidents that must be recorded and managed from those that are "reportable incidents". A manager works through the same questions each time.

1. Which category? The rules, as made in 2018, list reportable incidents including the death of a person with disability, serious injury, abuse or neglect, unlawful sexual or physical contact or assault, sexual misconduct (including grooming), and the unauthorised use of a restrictive practice. Unlawful physical contact whose contact with, and impact on, the person is negligible is excluded. Read the definitions in the current rules and Commission guidance, not a summary, and note that alleged incidents are included.
2. Which clock? Death, serious injury, abuse or neglect, unlawful sexual or physical contact or assault, and sexual misconduct must be notified within 24 hours of the provider becoming aware, by telephone or in writing, with any remaining information in writing within five business days. Other reportable incidents, including an unauthorised use of a restrictive practice, are notified in writing within five business days. If the same event also amounts to serious injury, abuse or assault, treat it under the 24-hour clock. The Commission's guidance has its own wording on restrictive practices that cause harm, so check it before relying on five days.
3. When did we become aware? Record the time. If you are unsure, take the earlier time and ask the Commission or an adviser.
4. What do we send? Use the Commission's approved form or portal. A 24-hour notice may be given by telephone or in writing, but the follow-up information due within five business days must be in writing. The Commission can also require a final report within 60 business days (or a longer period it sets), covering any investigation, its findings and the corrective action taken.
5. Who else is told? Police where a crime is suspected, family or guardians, the participant's support coordinator, and any other body your policies name.
6. How long do we keep the record? The rules require records of reportable incidents to be kept for seven years from the day of the notification in the cases they specify, and other laws may add requirements.

Even incidents that are not reportable must still be recorded and managed under your incident management system, which the rules say must cover incidents that have, or could have, caused harm to a participant. So "not reportable" is not "ignore". Record the decision, the reason and the follow-up.

A timeline example the manager can copy

An illustrative case with invented times. At 2.15pm on a Tuesday a worker sees a participant fall against a bench and sustain a cut to the elbow. First aid is given at 2.25pm and the manager is phoned at 2.30pm. The manager arrives at 3.10pm, sees that the cut needs medical attention and arranges a doctor's visit. At 3.20pm the manager decides the injury might meet the "serious injury" category and records that the provider became aware of a possible serious injury at 2.30pm.

The point of the table is that the clock is a fact you can read off, so write it into the report as it develops. If the manager decides it is not a serious injury after medical advice, the decision, the time and the reason are still recorded, and an internal review follows. If in doubt, many providers report and explain rather than hold back, but that is a judgement for your key personnel and your adviser.

Three more drafting examples

Medication incident. A worker finds that a participant's evening medication was given at the wrong time. The AI-tidied account lists times, who gave what, and what was done. It must not state the clinical consequence; the manager seeks advice from the participant's health practitioner and records that. The categorisation (is it reportable?) is a manager's decision.

Allegation of neglect from a family member. The report records the family member's words as a quote, when and how they were received, and who was told. AI may tidy the layout, but must not rephrase the allegation in softer words, because an allegation is itself within the rules' scope.

Near miss. A participant almost left the house unsupervised, and a worker intervened. No harm occurred, so it may not be reportable, but it falls under incident management because it could have caused harm. A good report here is shorter, and the follow-up is the valuable part: what changed so it will not recur. AI can help draft the review questions, with the names removed.

Cleaning an account before AI sees it: a filled-in example

Redaction is where most privacy slips happen, so practise it. Here is an illustrative raw account as a worker might dictate it, then the cleaned version.

Raw: "I was at [participant's first name]'s flat on [street name] with her sister [sister's first name] at about quarter past two. [Participant's first name] got upset about her lunch and threw her cup. It hit the wall and she cut her hand on the broken bit. Her NDIS number is on the file. I rang [manager's first name], my manager, at half past two." (In a real raw account each bracket would be an actual name, which is exactly what the cleaning step removes.)

Cleaned: "I was at Participant A's home with a family member at about 2.15pm. Participant A became upset about lunch and threw a cup. It hit the wall. Participant A cut a hand on the broken pieces. I phoned the manager at 2.30pm."

What was removed: the first name, the street, the sister's name, the funding number and the manager's name. What was kept: times, actions and roles. Note one trap: "became upset" is a mild judgement, and the manager may ask the worker what they actually saw, such as raised voice or crying. The cleaned text goes to the AI; the raw text stays in the incident management system with its access controls.

Teach new staff with three practice accounts like this one, and ask them to clean each before touching a tool. Anyone who leaves a name in during practice is better caught in training than in a live report.

Using the template to spot patterns, not only to file reports

Once reports share fields, a monthly review becomes possible, and the rules expect providers to use incident management to identify systemic issues and improve supports. A manager can tally incidents by type, time of day, location, worker roster and participant, with identifiers kept inside the secure system. Patterns often appear: four falls in the same bathroom, three medication delays on the shift handover, several behavioural incidents on days the routine changed.

AI can help here, if you keep it to de-identified, aggregated lines such as "Falls: 4, all between 6pm and 8pm, three involving the same stairs". Ask it for questions to investigate, not conclusions: "What other facts should we collect to understand why falls cluster in the evening?" The manager and the team decide what the pattern means and what to change. Record the change and check next month whether the numbers moved. An illustrative result: after adding a handrail and moving the evening meal earlier, evening falls drop from four to one. That is the kind of evidence a quality review, an audit or a family meeting finds persuasive, and it comes from consistent reports rather than clever software.

Drafting slips that creep into AI-tidied incident reports

• The overnight softening. A model changes "hit" to "touched". Fix: compare the AI version with the original line by line for the first month, and keep the instruction "do not soften".
• The invented time. A report gains a "2.45pm" that nobody said. Fix: [MISSING] flags and a manager who asks the worker.
• The awareness gap. The report records when the incident happened but not when the manager knew. Fix: make the awareness field compulsory in the template.
• The unreviewed report. A report is filed because it looked complete. Fix: no sign-off, no filing; the manager's initials and time are a field.
• Using a personal account. A worker pastes an account into their own chatbot at home. Fix: a clear policy and an approved alternative, because banning without an alternative just hides the practice.

Setting a simple policy for your team

Write half a page that every worker can follow. It names the template, says who may use which AI tool and for what, bans identifying details in public tools, requires the worker's original account to be saved unedited, makes the manager the only person who decides reportability, and sets a training date. Test it with three made-up incidents before relying on it, and review it after the first real month. The wider picture, including service agreements, is in AI for disability services: service agreements and incident logs, and a review-first approach to policies is in home care agency policies and procedures drafted with AI.

Four signs the reporting routine is getting better

Look at four things after a month. Reports arrive with every template field filled or marked missing. Opinion words fall. The time between incident and manager review shortens. Every reportability decision has a name, a time and a reason. If AI made reports faster but any of those got worse, stop using it for that step.

This is practical guidance, not legal advice. For your obligations as a provider, rely on the current rules and the Commission's guidance. If you want to map your incident routine with someone, my AI implementation consultation covers that. Otherwise, copy the template and try the drafting prompt on one invented incident today.

Questions providers ask about incident reports and AI

Can a support worker use ChatGPT to write an incident report?

Only if your organisation's policy allows it and no participant details are entered. A safer routine is for the worker to write the facts, remove names, and let a manager decide whether AI tidies the wording inside an approved system. Never put identifying details into a public chatbot.

Who decides whether an incident is reportable?

A person with authority in the provider, usually the manager or key personnel, decides using the reportable-incident categories in the rules. AI must not make or influence that decision. Record who decided, when and why, including when you decide it is not reportable.

Does the 24-hour clock start when the incident happens?

The rules start it when the provider becomes aware the incident occurred, not when it happened. Write the time you became aware in the report, and check the Commission's current guidance for how awareness is defined for key personnel.

Should the AI draft be kept in the file?

Keep the final approved report and the original worker account. If an AI tool changed the wording, note that in your process and keep the source facts, so the record shows what people saw and said rather than what a tool rephrased.

Further reads

• Home Care Agency AI: A 90-Day Plan for the Owner — A 90-day plan for a care provider's owner.
• Home Care Agency Employee Handbook: Draft With AI, Then Review — Drafting an employee handbook, then reviewing it.
• Home Care Agency Welcome Packet: Build the Admission Pack With AI — An admission pack built with AI.
• Adult Day Care Software for a Small Centre: Billing and AI Notes Checklist — Attendance, billing and AI notes in day-care software.
• AI Home Care Services: Turning Visit Notes Into Family Updates — Turning visit notes into family updates.
• How Can AI Help in Care Homes? Ten Back-Office Uses — Ten back-office uses for a care provider.
• 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.
• NDIS Software for Small Providers: Per-User Costs and What You Get — ShiftCare's AUD per-user tiers, Brevity's per-client rates, worked monthly totals for a 12-person provider, and which AI features come with each plan.
• NDIS Audit Checklist: Build the Evidence Folder With AI — Map each practice standard to the document that proves it, list the gaps, and keep a version log, with AI doing the sorting and you doing the judging.
• 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.
• Can AI Write a Care Home Incident Report? Template and Checks — A factual care home incident report template, prompts that tidy staff notes without adding facts, and the manager checks AI must never replace.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: National Disability Insurance Scheme (Incident Management and Reportable Incidents) Rules 2018, current version on the Federal Register of Legislation (checked November 2026). The NDIS Commission's website could not be reached for this check, so confirm its current guidance and portal steps there.