Why some AI note apps put NDIS support workers at legal risk

This post explains what that means, why it matters, and how to check any app before you trust it with your documentation.

What fabrication means in an AI note context

When a support worker finishes a shift, they sit down to write their notes. The note is only as good as what the worker brings to it. More detail, more context, more accurate observation, all of it produces better documentation. But regardless of how much or how little a worker provides, the question remains the same: where does the content in the final note come from?

There are two approaches.

The first approach fills the gaps. If the worker wrote “took participant to the shops and helped with lunch,” the app might add that the participant “appeared engaged and in good spirits,” that they “demonstrated independence selecting items,” and that the session “aligned with goals around community participation and daily living skills.” None of that was in the worker’s input. The app invented it. It reads well. It is longer. It will fail an audit.

The second approach flags the gaps. If the worker wrote “took participant to the shops and helped with lunch,” the app identifies what is missing, how the participant presented, any relevant observations, goal relevance, and asks the worker to fill those gaps before the note can be signed. The output is shorter. It is accurate. It reflects what actually happened.

The first approach is fabrication. The second is documentation.

That is what Clio Care does. Every note generated in Clio identifies what is missing from the worker’s input, flags the gaps, and asks the worker to fill them before the note can be signed. The worker stays the author. The AI stays the tool. That is the Note Integrity Standard — and it is the only approach that holds up when an auditor reads the file.

Why fabrication fails NDIS audits

NDIS auditors are not checking whether your notes are long. They are checking whether your notes are accurate.

The NDIS Practice Standards and the National Disability Insurance Scheme Act 2013 are explicit: support workers are responsible for the accuracy of the documentation they sign. The software that helped produce the note is not a party to that obligation. You are.

When an auditor finds a note that describes a participant as “engaged and in good spirits” during a session where incident records show a behavioural episode, or finds consistent goal alignment language across every note regardless of what actually occurred in each session, two things happen.

First, the note is considered unreliable. All notes from that worker may then be scrutinised.

Second, the worker, not the software company, is held accountable for the inaccuracy. The app’s terms of service will not protect you. They almost never do.

A fabricated note is not a compliant note. A longer note is not a safer note.

What note integrity means and why it is not a default setting

Note integrity means the note contains only what the worker observed, did, or was told. Nothing added by the software to fill gaps or improve readability.

This is not a default behaviour in AI systems. The underlying models used by most note apps are optimised to produce fluent and complete sounding text. Left unconstrained, they will fill gaps with plausible content. That is what they are trained to do.

Producing a note with integrity requires explicit constraints on the model. Rules that prohibit adding detail not present in the worker’s input, that identify gaps rather than fill them, and that are tested consistently to confirm they hold.

These constraints have a cost. Notes are shorter when the worker provides less. But a shorter accurate note is worth infinitely more than a longer fabricated one. The worker can sign it knowing every word reflects what actually happened.

A word on enhance and polish functions

Some tools offer a one tap function that rewrites or substantially expands your note from a template, regardless of what you actually documented. This is not AI assistance. This is template generation with your name on it.

The NDIS Code of Conduct, specifically the obligation to act with integrity and transparency, and the NDIS Practice Standards Quality Indicator 1.5, which requires that records are accurate and reflect the support provided, are unambiguous: documentation must reflect what actually occurred. A note generated from a template and signed by a worker is a misrepresentation of that worker’s record keeping, regardless of how the software describes the function.

If an app offers a button that substantially rewrites or expands your note beyond what you provided, ask what legislation that function complies with. A credible answer will reference the Practice Standards directly. A marketing answer will not.

How to test any AI note app before you trust it

This test takes two minutes and will tell you more than any marketing claim.

If the note describes the participant’s mood, their engagement level, their progress toward goals, their communication during the session, or anything else not present in your two sentences, the app fabricated that content. You would be signing a note containing things that may not have happened.

If the note produces a shorter output and identifies the missing information as gaps for you to complete, the app is handling your input honestly.

Run this test on every app you are considering. The results will vary significantly.

Also watch for apps that allow you to generate a complete shift note from a template title alone, with no session description required. If an app can produce a full note from the words “community access” or “personal care” without any input from you about what actually happened, every word in that note is fabricated. It may read well. It will not survive an audit.

The current market

Some tools now market the ability to expand minimal notes into longer documentation as a compliance benefit. The argument appears to be that longer notes are less likely to attract scrutiny. This is incorrect. Auditors are trained to identify notes that are inconsistent with the complexity of the support provided, that contain language unlikely to reflect the worker’s own observations, or that show uniform structure across all sessions regardless of what occurred.

Independent support workers carry significant personal liability for the documentation they sign. The choice of tool is a risk management decision, not just a workflow decision.

How Clio Care approaches note integrity

Most AI note apps were built to produce impressive output. Clio was built to produce accurate output. Those are different goals and they produce different products.

Before Clio generated its first real note, we wrote a formal standard for what compliant NDIS documentation requires. That standard has not changed. Every decision about how Clio handles worker input flows from it.

Clio’s test suite includes exhaustive anti-fabrication testing across a range of input types, from minimal to detailed. Every scenario must confirm that the output contains only what the worker provided. Anti-fabrication testing is a condition of every update to how Clio generates notes. If any test fails, the update does not go live.

Notes with missing information cannot be signed off. The worker is shown what is missing and completes it before signing. This is not a feature. It is a constraint, one we consider absolute for any tool that touches NDIS documentation.

Clio is the only NDIS note app with a publicly available note integrity standard. Read it at cliocare.com.au/note-integrity-standard, then ask any other app you are considering whether they can show you the same.

Questions worth asking any AI note app

Before you commit to a tool, consider asking: