AI Complaint Handling Policy: Practical Processes for Australian Teams
Updated: 17 hours ago
A lot of complaints policy work in Australia is being done using policy guidance that was written before AI mattered. The NSW Government model policy, the ACNC template, the Victorian Ombudsman's model complaint handling policy. These are great documents to guide complaints management. Many organisations have adapted them well. None of them address AI.
What I have been hearing from complaints officers and dispute resolution teams over the past year is variations of the same question. Their existing policy is fine in principle but silent on practice: silent on AI-drafted submissions, silent on AI use disclosure, silent on verifying AI-generated case references, silent on what to do when a vulnerable complainant has used AI to articulate something they could not have written on their own. The policy was not written for the situation the team is now in. The teams are now in a situation where they are finding themselves increasingly buried in AI generated correspondence.
This page covers the practical updates a complaint handling policy needs to make sense in the AI era. In my consultancy I work with clients to develop policies in-house that address these gaps in current frameworks. We work together on creating practical, usable AI Complaint Handling Policies that public sector, Ombudsman, tribunal and regulatory teams can take, adapt and run as their own. These policies are tailored to your organisations requirements and demand.
The existing policy is fine in principle but silent on practice. The policy was not written for the situation the team is now in. That is, buried in AI generated correspondence.
What does a complaint handling policy need to say about AI?
A complaint handling policy that addresses AI does not need to be a separate document. In most cases, the better approach is to update the existing complaints policy with AI-specific provisions in the right places, rather than create a parallel AI policy that sits alongside. It is important not to treat a complaint or complainant differently for the only reason that they used AI. That is likely to cause risks for an organisation.
If you are considering updates to your complaints management policies, there are seven provisions worth thinking about. Not every organisation will need all seven, and not every organisation will need them at the same level of detail.
The seven provisions are:
Scope and definitions: what the policy means by AI, and what kinds of AI use are covered.
AI use disclosure: whether and how complainants are asked to disclose AI assistance, and what the team does with that disclosure. Is it needed at all?
Verification of AI-generated material: how the team handles AI-drafted citations, references and quantitative claims.
Procedural fairness in AI-influenced matters: how AI use by one party affects the team's obligations to all parties.
Privacy and confidentiality of AI-influenced complaint information: how AI use changes (and does not change) the team's privacy obligations.
Staff use of AI in complaint handling: what the team's own staff are permitted, encouraged or restricted from using AI for.
Training and review: how the policy stays current as AI tools and stakeholder behaviour change.
The remainder of this article walks through each provision at a practical level. If you'd like to discuss how we can tailor the actual policy wording I'd love to have a chat, you can email me at contact@shivmartin.com or call me directly on 0433 904 303.
How does a policy address AI use disclosure?
This is the question that comes up first in almost every policy conversation. Should we ask complainants to tell us if they have used AI to prepare their submission?
In most contexts, the answer is yes, but the framing matters. A disclosure question that reads "Have you used artificial intelligence to prepare any part of this complaint?" produces almost no useful information. Many people will not know how to answer it. They used ChatGPT to summarise their emails but typed the rest themselves, or they used Google Translate, which is also AI. Others will simply not answer at the level the question is asking about.
A more useful disclosure question is one that names specific assistance the team needs to know about.
The point of disclosure is not to disadvantage complainants who used AI. It is to give the team useful context so they can verify specific material if needed. The policy provision should make that purpose explicit.
Shiv Martin talks about the challenge of AI generated correspondence in complaints processes.
How does a policy address verification of AI-generated material?
The most documented AI risk in complaint handling is fabricated case citations. AI tools sometimes generate plausible-looking legal authorities, statutory references or precedent cases that do not exist. This is a well-known problem in the legal profession and it has appeared in Australian tribunal matters often enough to be a practical concern.
The policy provision worth having is not a ban on AI-generated material. It is a verification requirement: any factual or legal claim that the complaint or response relies on must be verifiable. If a citation cannot be located, the team is entitled to ask the party to provide the source or withdraw the claim.
How does a policy address procedural fairness in AI-influenced matters?
This is where the deeper conceptual work happens, and it is the section of the template that varies most across organisations.
The core question is whether AI use by one party affects the team's obligations to all parties. In most cases, the answer is that the underlying obligations do not change. Every party is still entitled to be heard, to know the case against them, and to have their material considered without bias. What changes is how those obligations are operationalised.
If one party arrives with an AI-drafted submission that runs to forty pages and the other party arrives with a handwritten letter, the team's job is not to police AI use. It is to ensure both parties have a fair opportunity to put their case. That may sometimes mean offering the second party more time, more clarification questions, or more support in articulating their position. The policy provision should make explicit that the team takes this responsibility seriously and how it does so in practice.
A separate article on this site covers procedural fairness in AI-assisted complaints in much more depth. The policy provision is the operational version of those principles. The article is the thinking behind them.
How does a policy address privacy and confidentiality of AI-influenced complaint information?
This is the question Privacy Officers ask. It does not always come up in the first conversation, but it comes up before anything is signed.
AI use by complainants raises two practical privacy questions. First, if a complainant has pasted complaint information into a public AI tool to draft their submission, that information may have been retained by the AI provider. The team cannot control what has already happened, but it can advise complainants about safer practices for any further submissions. Second, if staff use AI tools to assist with case work, the team needs clear guidance on what complaint information can and cannot be entered into those tools.
The policy provision worth having addresses both questions: it acknowledges that complainants make their own choices about AI use and the team does not police those choices, but it sets clear rules for what staff can put into AI tools when working on cases.
For organisations with statutory confidentiality obligations, Ombudsman offices and tribunals in particular, the staff-use provision should be tightened to reflect the relevant Act. When working together on these policies it's important to highlight organisation-specific statutory references because this is an area where one-size-fits-all wording could actively cause problems.
How does a policy address staff use of AI in complaint handling?
This is the provision that has changed the most in the last twelve months and is likely to keep changing. A year ago, most organisations were either banning staff AI use entirely or staying silent. Today, most are working out a middle ground.
The policy worth writing is one that recognises both ends of the spectrum. Staff use of AI for general productivity tasks (drafting correspondence, summarising long submissions, suggesting clarification questions) is increasingly common and, in most contexts, defensible. Staff use of AI to draft a substantive finding, recommendation or decision is a different question, and most organisations are not yet comfortable with that level of delegation.
When working with clients on developing these policies I provide a tiered structure: AI use for productivity tasks is permitted with team awareness, AI use to draft decision-making content is restricted, and the line between the two is explicit enough that staff can apply it day-to-day. The structure can be tightened or loosened depending on the organisation's appetite. What it should not be is silent.
How does a policy stay current as AI changes?
The honest answer is that any AI complaint handling policy written today will need revisiting within twelve months. The tools are changing, stakeholder behaviour is changing, and the legal and regulatory environment is changing.
The policy provision worth including is a review cycle: a named owner, a defined review date, and a trigger condition that brings the policy back to the table earlier if a significant development warrants it. The template provides wording for this with a default twelve-month cycle and a list of trigger conditions (significant case law, new statutory guidance, material change in AI tool capability, sustained pattern of new complaint behaviour).
The review provision is not a procedural afterthought. It is the provision that keeps the policy from going stale in the way that the existing pre-AI policies have.
A working AI complaint handling policy is not the destination.
It is the operational scaffolding that lets the team handle the situation they are actually in, with the confidence that they are doing it consistently and defensibly.
We support client's to develop policies in-house that are real and usable. Our AI Complaint Handling Policies are designed for Australian public sector, Ombudsman, tribunal and regulatory teams. They are not fill-in-the-blanks documents. They are policies we would draft for an organisation in the first meeting, refined through actual policy advisory work with complaints teams across the country.
Please email us at contact@shivmartin.com or find out more about how we support our clients with Policy and Process Development and Support here.
.png)