The Australian Public Service (APS) is increasingly using artificial intelligence (AI) to support communication efficiency.
The impact of this shift on communication quality in a system defined by ministerial accountability, cabinet sensitivity and public scrutiny is still emerging.
AI is doing exactly what it is designed to do, producing fast, polished and highly readable content. However, readability is not the benchmark that matters in government. Communication must be accurate, aligned with policy intent and appropriate to the ministerial and public context in which it sits.
Source: Gaurav Uttamchandani
How is AI Being Used?
Across APS communications teams, AI is being used to draft content that is then reviewed quickly and progressed because it appears clear and well structured.
It is efficient, but it introduces risk. Content can be accepted because it reads well rather than being properly tested for accuracy, nuance and alignment.
A draft that sounds credible is less likely to be challenged. A summary that appears clear is not always checked against source material. A message that flows well is not always tested against ministerial intent or policy sensitivity. Over time, judgement is applied less consistently and begins to be assumed rather than exercised.
Source: Kirill Chabanov
The Issue with Quick Review
We have already seen examples in Australia and internationally where AI-assisted government-related outputs have included fabricated references, incorrect attribution and misinterpretation of source material. These issues rarely arise from carelessness; they come from accepting outputs that appear credible without properly testing them.
AI does not usually produce obviously flawed content. It produces content that is convincing enough to pass an initial review. That is the pressure point.
You need sufficient experience to identify where errors sit because they are no longer obvious.
This matters more in the APS than in most environments. There is no tolerance for communication that is good enough. Messages must be precise, defensible and fully aligned with policy intent. That standard depends on judgement.
Source: Martin Koch & Nurlan Suleymanov
AI has a clear role in supporting communication. It can assist with drafting, summarising complex material, translating technical language into plain English and repurposing content across channels. These efficiencies matter in high-volume environments. But they sit within production, not decision-making. AI should not determine final messaging, shape tone in sensitive contexts, or replace human review.
The real issue is not AI capability. It is discipline.
When 'good enough' is not good enough
Most APS agencies have access to AI tools. What varies is how consistently outputs are reviewed, challenged and approved. Without rigorous scrutiny, communication systems begin to prioritise speed and volume. Work moves faster, but the space for deeper interrogation of narrative consistency, stakeholder alignment and policy nuance is compressed.
This is reinforced by a practical constraint. Communications teams are already operating at or near capacity, managing high volumes of taskings, ministerial priorities and complex stakeholder environments. At the same time, expectations continue to rise for faster turnaround, clearer messaging, greater accessibility and stronger strategic alignment.
The result is not a lack of effort, but pressure on judgement. When time is limited and volume is high, there is a tendency to accept work that appears “good enough”.
That is the real shift.
Source: Denis Matusovskiy
DON'T OUTSOURCE JUDGEMENT
The challenge is no longer producing communication efficiently. It is ensuring it holds up under public, political and administrative scrutiny.
There is no margin for “good enough”; it must be right.
AI will continue to play a growing role in government APS communications. What must not change is the standard, judgement and “pub test” applied to its outputs. In government, accuracy is not optional and accountability cannot be delegated.
AI can support communication development, but it cannot be responsible for what is published.
Use the tool. But don’t outsource judgement.
