How to Write NDIS Progress Notes: An AI-Assisted Audit Checklist

An audit-ready NDIS progress note records, soon after the support, the date and times, who delivered it, what was done, how the participant responded as an observation, how it links to a plan goal, and any follow-up. AI can help with voice-to-text and review, provided the worker checks every note and identifying details stay protected.

Most guides stop at what a progress note is. A provider facing audit needs more: a field-by-field checklist, good and weak examples side by side, a way to review a batch of notes with AI without exposing participants, and safeguards that keep a worker, not a tool, as the author of the record. The Commission sets outcomes rather than a template, so its own documents and your auditor have the final say on requirements; nothing here replaces them.

Follow me on Instagram@sagnikteaches

What auditors keep finding wrong

Provider guidance on audits tends to name the same weak spots, and they are worth keeping in mind as you read the checklist. Notes are vague ("had a good day"). Notes are copied from one shift to the next. Times in the note do not match the claim you submitted. The note describes what the worker did, but not how it connected to the participant's goals. Opinions are written as facts ("was being difficult"). And notes are written days later.

Connect on LinkedInSagnik Bhattacharya

Take care with where those claims come from: most come from software vendors and consultancies, not from a rulebook. The Commission's own documents set outcomes instead. The Practice Standards' information management outcome asks that each participant's information is "identifiable, accurately recorded, current and confidential", with an information management system that records it "in an accurate and timely manner". The Commission's Code of Conduct guidance for providers adds that records should be accurate, up to date and legible, detailed enough for continuity of support, free of demeaning or derogatory language, and made at the time of an event or as soon as possible afterwards. Neither prescribes headings, so the checklist below is a practical audit tool built on those outcomes, and you should test it against what your own auditor tells you. If you are choosing where to start with automation across the whole provider, the assessment checklist for which jobs to give AI first comes before this one.

Subscribe on YouTube@sagnikteaches

The field-by-field audit checklist

Use this as the standard every note is held to, whether the worker typed it, dictated it or had a tool tidy it. Each row says what a reviewer looks for and how to check it quickly.

Add a tenth check that is not a field: is it original? If two notes for the same participant on different days have identical sentences, one of them is probably not a record of that shift.

Good and weak notes, side by side

The examples below are invented. Names are left out on purpose, as they would be in a de-identified review sample. Read each pair and name the failing field before reading the verdict.

Example 1: the vague note

Weak:  "Support provided for community access. Participant was happy
        and engaged."

Better: "13:00-15:30. Took participant by bus to the library and then to
        the shops (goal: travel independently). Participant bought a
        bus pass at the kiosk using their own card, with a verbal
        prompt for the pin only. Chose the library over the park when
        offered both. Asked to practise the route home alone next week;
        agreed with participant to plan it on Tuesday."

The weak version fails on specific tasks, goal link and choices. The better version is not longer for its own sake: every sentence tells the next worker something.

Example 2: opinion written as fact

Weak:  "Participant was being difficult at dinner and refused to eat."
Better: "At 18:15 participant pushed the plate away and said 'I don't
        want that'. Offered two alternatives (soup, toast). Chose toast
        and ate half. No distress observed. Will mention food
        preferences at the team meeting."

"Difficult" is a judgement. The better note records what was said and done and what was offered. The Commission's provider guidance on the Code of Conduct specifically asks providers to keep demeaning or derogatory language out of records, and labels like "difficult" or "uncooperative" drift towards it. If a dispute ever arises, the second note is the one that protects both the participant and the worker.

Example 3: copy-paste

Mon: "Assisted with morning routine. Showered, dressed, breakfast.
       Participant in good spirits."
Tue: "Assisted with morning routine. Showered, dressed, breakfast.
       Participant in good spirits."

Identical text across shifts is a red flag, even if the shifts really did go the same way. The fix is to record what changed or what the participant did independently, for instance "Participant chose the blue shirt and buttoned it without help; needed a reminder to rinse hair."

Example 4: a note that hides an incident

Weak:  "Slight slip in the bathroom, all fine."
Better: "09:20 participant slipped on the bathroom floor (wet tiles),
        sat down, no visible injury, said 'I'm okay'. First aid check
        done, no pain reported. Manager rung at 09:35. Incident
        report started (ref in system). Floor dried; mat requested."

A fall must be recorded as an incident through your incident process, and the progress note should point to it. Never let the note be the only place something serious is mentioned.

Example 5: a note that cannot be matched to a claim

Weak:  "Support in the afternoon. Went out."
Claim: 4 hours at the community access rate.

A reviewer cannot confirm four hours from "afternoon". When billing and notes disagree, it is the claim that is in trouble. The tutorial on checking NDIS claims against shift notes with AI shows how to compare the two in bulk.

Voice-to-text: faster notes, with a safe workflow

Dictation helps workers who find typing slow at the end of a shift, and it tends to produce notes with more detail because speaking is easier than typing. The risk is where the audio goes. Phone dictation, a word-processor dictation feature and a care-software microphone button may each process speech differently, on the device or in a cloud service, and store it under different terms. Ask your software vendor, in writing, four questions before workers use any of them.

1. Where is the audio processed and is it kept, and for how long?
2. Is our data used to train any model by default, and can that be switched off?
3. Does the contract treat participant information as confidential and sensitive, and who are the sub-processors?
4. Can we export and delete notes and audio on request?

If the vendor cannot answer, do not use the feature for participant notes. Write the answers into your approved-tools list, with the date you asked, so an auditor can see the decision was made deliberately.

A dictation routine a worker can follow

1. Dictate inside the approved care-management tool, not a personal notes app.
2. Speak in the order of the checklist: times, what you did, the goal, how the participant responded, choices, concerns, follow-up.
3. Read the transcript before you save. Speech tools mishear names, medication names and place names, and "no" can become "know". Fix every error.
4. Remove opinion words, then add anything you forgot.
5. Save, and tick "reviewed by author" if your system has the option.

The fifth step matters for the audit. A note without a reviewed flag, or with a transcript that still reads like speech ("um, and then we went"), tells a reviewer the worker did not check.

Using AI to review notes without exposing participants

The safest and most valuable use of AI here is not writing the notes: it is reviewing a sample for weak spots. The manager exports a batch of notes, strips every identifier, and asks an AI tool to flag vague phrases, missing fields and copy-paste. A person then reads the flagged notes and talks to the worker. Run the review inside a tool your provider has approved, on a plan that does not train on your content by default, and keep the sample de-identified. The tutorial on anonymising client data before you paste it into AI explains the technique.

Preparing the sample

Take 20 to 40 notes across several workers and participants. Replace participant names with codes (P1, P2), remove addresses, phone numbers, plan numbers, other family or staff names, exact dates (use "Day 1" and "Day 2" if you need order), and any health detail not needed for the review. Keep the time of day and duration because you are checking those. If a note is too identifying to clean quickly, leave it out of the AI sample and review it yourself.

A review prompt, with illustrative output

You are a documentation reviewer for a disability support provider.
Below are 12 de-identified progress notes (N1-N12).
For each note, check these fields: specific support delivered, link to a
goal, participant response as an observation, choices recorded,
follow-up or handover, start/end times.
Return a table: note ID, fields missing, any opinion words or labels
(quote them), and whether the text looks copied from another note.
Do not rewrite the notes. Do not infer facts that are not in the text.
Notes:
[paste notes]

An illustrative fragment of the result:

N3  | goal link missing; follow-up missing | "was being difficult" | no
N7  | none missing                         | none                  | no
N9  | specific support missing ("assisted as needed") | "happy" | similar to N8
N11 | end time missing                     | "uncooperative"        | no

Now the safeguard step. Open N3, N9 and N11 yourself. Check that "similar to N8" is true; assistants sometimes see similarity where there is only a shared template line. Check that N7 really is fine, because tools can also miss a missing goal link when the text mentions a goal in passing. The tool tells you where to look first; the manager decides whether a note passes.

Four workers, 40 notes: one coordinator's monthly sample

Say a small provider has four support workers and one coordinator. The coordinator samples 40 notes from last month, ten per worker, and runs the review above in batches of ten. Illustrative results: 40 notes, 17 with no goal link, 12 with an opinion word, 9 with missing or rounded times, 6 that read as copies, and 14 that passed on every check. The coordinator then reads all 26 flagged notes in under two hours, agrees with the tool on 23 and overturns 3 where it was wrong: 88 per cent of flagged notes confirmed.

What follows is more important than the figures. The coordinator sees that 11 of the 17 missing goal links are on notes for the same two participants, written by three different workers, and finds the cause is not carelessness: those participants' new plans were never copied into the roster app, so the workers had no current goals to refer to. Updating the app fixes the pattern in a week. Running the same sample next month, the goal-link failures fall to 5 of 40. This is the real benefit of an AI-assisted review: it points to where the system, not just the person, is letting notes down.

Safeguards: keep the worker as the author

The record is the worker's account of what they saw and did, so no tool may become its author. Write these safeguards into your documentation policy and train workers on them.

• The worker reviews and owns every note. Not "AI generated". Not auto-saved.
• AI may tidy, never invent. A tool may fix grammar and structure what the worker said, but must not add facts, times, feelings or outcomes.
• Participant details stay in the approved system. Any general chat assistant gets de-identified text only, and the worker adds names and specifics afterwards in the records system.
• No AI for incident reports or clinical judgement. Use the proper incident process; a tool can help check wording only afterwards, and only with a person in charge.
• Keep an approved-tools list. Workers who use a personal chatbot on their phone are the biggest privacy gap, so name the permitted tools and say plainly what happens if someone breaks the rule.
• Be open with participants. If a participant or their representative asks how their notes are written, tell them what you use and that the worker checks every note. The tutorial on tailoring policies and procedures with AI helps you write this into your documents.

Treat the records as sensitive from the start: participants' notes describe health, behaviour and personal routines. Privacy law and the Commission's requirements apply to how you hold them, and your auditor can tell you what evidence they want. On retention, the Commission's Code of Conduct guidance for providers says to keep participant records for a minimum of seven years after the support was provided. Ask your auditor or adviser whether any other rule sets a longer period for particular records you hold, and make sure deletion of drafts, audio and AI chat history follows the same policy.

How the checklist changes with the kind of support

The same fields apply everywhere, but what a strong note looks like depends on the support. Three illustrative cases show where reviewers should lean.

Personal care and daily routines

Routine shifts are where copy-paste thrives, because the tasks genuinely repeat. The strong note records what the participant did for themselves and where help was needed, since that is what shows progress against a goal like "dress independently". "Needed help with the second sleeve only" tells a reviewer more than three lines of tasks. Choices matter here too: which clothes, what time, which meal. Dignity and consent are easiest to evidence in small choices.

Community access and transport

Here the weak point is usually time. Travel time, waiting and activity time can be claimed differently, so the note should separate them: "13:00-13:40 travel by bus, 13:40-15:00 library, 15:00-15:30 travel home". A reviewer then matches the note to the claim line by line. Add anything that went wrong, such as a cancelled bus, because it explains why the plan changed.

Support coordination

Notes here record calls, emails and plan actions rather than shifts: who was contacted, what was agreed, what happens next and by when. The common fault is a note that says "liaised with provider" and nothing else. A reviewer needs the outcome and the deadline, and AI review prompts should check for exactly those two things.

A note template that carries the checks

If your software lets you set prompts, use a short structure that makes the right fields hard to skip. This is a template to adapt, not a required format, since the Practice Standards set outcomes rather than a mandatory layout.

Date / start / end:
Worker (name, role):
Goal this supports:
What I did (specific tasks, in order):
What the participant did, said or chose (observations only):
Anything unusual or risky (and incident reference if any):
Follow-up for next shift or coordinator:
Reviewed by author: yes / no

A template like this does more for quality than any AI check, because it prevents the gaps instead of finding them afterwards. When you later run the AI review, the structured fields also make it far easier for the tool, and for you, to spot what is missing.

Train with the weak notes, not just the good ones

Take three weak notes from your own sample, de-identified, and ask workers to rewrite them in ten minutes. Compare the rewrites with the checklist together. This takes one team meeting and teaches the standard better than a policy document does. Repeat it each quarter with fresh examples so people see that the bar stays the same and the examples change.

A monthly note review that fits into three hours

Here is a schedule a small provider can keep up. The times are estimates for a four-worker team.

That is a little over three hours a month. Keep the monthly counts, such as missing goal links and rounded times, in a simple sheet. A downward trend is useful evidence of a working quality system when an auditor asks how you monitor documentation, and the audit evidence folder tutorial shows where to file it.

Give feedback that changes behaviour

Workers dislike being told their notes are weak. Two things help. Show them a good note from a colleague (de-identified) alongside their own, and explain which line made the difference. And start with the system fixes, such as goals in the roster app and a template with prompts for choices and follow-up, before you talk about performance. A short template reduces vague notes more reliably than reminders do.

When your software already has AI notes

Many care-management systems now offer dictation or AI summaries. They fit into the checklist the same way: the feature is only a drafting aid, and your four vendor questions plus the author-review rule still apply. Compare what you get with what you pay; the tutorial on NDIS software costs for small providers lists what to expect per user. Pilot the feature with two workers for a month, run your monthly review on their notes and on a control group's notes, and compare the counts of weak fields. If the AI-drafted notes are no better on goal links and choices, the feature saves time but not quality, which may still be worth having, as long as you know.

Questions providers ask about AI and progress notes

Can a support worker use ChatGPT to write progress notes?

Not with a participant's identifying details in a general chat tool. A safer route is a tool covered by your provider's contract and privacy review, or a method that keeps names and specifics out until the worker adds them in your records system. The worker remains responsible for what the note says.

Should the note say that AI helped?

Follow your own policy, and make sure you have one. A reasonable approach is to record in the policy which tools may be used and that every note is reviewed by the worker who delivered the support. If a participant asks, tell them plainly what you use.

How soon after a shift should the note be written?

The Commission's Code of Conduct guidance for providers says records should be created at the time of an event or action, or as soon as possible afterwards. Many providers make the same shift their internal standard. Late notes are a common weakness because memory fades and entries get vague.

How many notes should a manager review each month?

A sample is realistic for a small provider, for example five notes per worker, chosen across different participants and shifts, plus every note linked to an incident or complaint. Increase the sample for new workers and for anyone whose earlier notes showed problems.

Can AI check notes against invoices?

It can help you compare times and descriptions in a de-identified extract, then flag mismatches for a person to look at. A person decides what is wrong and what to claim. See the tutorial on checking claims against shift notes for the method.

Further reads

• How to Get Clients as an NDIS Provider: AI for Enquiries, Referrers — Handle new enquiries and referrers once your notes are sound.
• AI Accounting for NDIS Providers: Categorising Income and Claims — Categorise income and claims once notes support them.
• NDIS Occupational Therapy Report Template: A First Draft With AI — A first-draft report template for therapists.
• NDIS Plan Manager Invoices: Checking Claims Against Budgets With AI — Invoice checks from the plan manager's side.
• Can Therapists Use ChatGPT for Session Notes? — The privacy line for session notes in therapy work.
• NDIS Provider Website: Plain-English and Easy-Read Pages Using AI — Plain-English and Easy Read pages for participants.
• ABA Session Notes With AI: From Raw Session Data to a Draft Note — A practice-owner workflow for turning trial data into a draft ABA note: de-identify or use a BAA tool, a prompt, a worked example and sign-off.
• AI for Disability Services: Service Agreements and Incident Logs — Draft service agreements from your own template and structure incident logs with AI, while keeping participant details out of the prompt.
• 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.
• NDIS Service Agreement Template: Plain-English Drafting With AI — Draft a plain-English NDIS service agreement with AI, add an easy-read version, and know what an adviser and the participant must check.
• AI for NDIS Providers: What Is Safe Under the Privacy Principles? — Safe and unsafe AI uses for an NDIS provider: the Privacy Principles in plain words, the Commission's February 2026 position, and a rule to adopt.
• Can AI Write Care Plans? What a Home Care Provider Must Check First — Provider-level checks before AI drafts a home care plan: accuracy, consent, de-identification, manager sign-off and an audit trail.
• 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 Quality and Safeguards Commission, NDIS Practice Standards core module (information management and incident management outcomes) and The NDIS Code of Conduct: Guidance for NDIS Providers (April 2024), both checked 4 November 2026; confirm requirements with the Commission and your auditor.