AI Generated Complaints: Five Patterns and How to Respond
- Shiv Martin
- Jul 7
- 9 min read
A complaint handler I spoke to recently described her week as "managing one good complaint and seven that read like they were written by a chatbot pretending to be a lawyer." That description has stayed with me, because it captures something the policy conversation about AI in complaints has not quite caught up with yet.
AI-influenced complaints are not a single phenomenon. They are a cluster of distinct patterns, each of which calls for a different operational response. Treating them as one category, or worse, treating them as a generic "AI problem" to be managed by general policy, produces guidance that does not help the person actually working the file.
This article sets out the five patterns I see most often in complaint handling work, along with practical responses for each. It is written for the complaint handler at intake or assessment, not the policy lead. The point is to give you something you can apply to the next AI-influenced complaint that lands on your desk.
AI changes how complaints arrive, not what complaints are. The AI is context, not content.
Pattern 1: The over-expanded complaint
What it looks like. A complaint that should have been a paragraph arrives as twelve pages. Every email exchange is included. Every irrelevant procedural step is recounted. The substantive issue is buried somewhere around page seven, surrounded by material that does not affect the assessment. The complainant has used AI to expand a short story into a long document, on the implicit theory that more material means a stronger complaint.
Why it is tricky. Two reasons. First, the substantive complaint may be genuine and worth responding to, but the volume of material makes it hard to assess. The handler ends up reading twelve pages to find the half-page that matters. Second, an over-expanded complaint can read as more substantial than it is, which sometimes pulls handlers into giving it disproportionate weight. Both of those are operational risks.
How to respond. The technique that works most often is a clarification request rather than a refusal. A short email that says, in effect, "thank you for the detailed submission. To help us assess the matter efficiently, could you tell us in one or two paragraphs what specific outcome you are seeking and what the central issue is?" gives the complainant a structured way to summarise, and gives you the basis for a proportionate response. Most complainants respond to this request constructively; the small number who cannot or will not are a different problem (see Pattern 5).
Pattern 2: The legally-styled complaint
What it looks like. A complaint written in legal register, citing legislation, referencing case authority, using formal Latin terminology, structured like a statement of claim. The complainant is most often a self-represented individual who has used AI to translate their concern into something that looks like a legal document. The substance underneath is sometimes valid, sometimes not, but the form is doing a lot of work in either case.
Why it is tricky. The form can be intimidating, particularly for less experienced handlers, and the temptation is either to over-respond (treating it as if it were drafted by a lawyer) or to dismiss it (deciding the complainant is being difficult because the form is unusual). Neither response is right. The form is not the complaint. The form is one party's attempt, often with AI assistance, to be taken seriously.
How to respond. The most important move is to look past the form to the actual concern. A useful clarifying question is "I can see this matter has been carefully prepared. In your own words, what is the outcome you are hoping for?" This question does two things at once. It acknowledges the effort that has gone into the submission (which the complainant has often invested significantly in), and it asks for a re-articulation in the complainant's own voice. The answer usually reveals whether there is a substantive concern underneath, and what it is.
For verification of any cited authorities in this kind of submission, refer to the verification provisions in your complaint handling policy (or, if you have not yet updated yours, the practical approach is to ask the complainant to provide a source for any specific legal reference they are relying on, and to assess the matter on the verifiable material).
Pattern 3: The fabricated reference
What it looks like. A complaint that cites case authorities, statutory references, or quantitative claims that look real but cannot be verified. AI tools sometimes generate plausible-looking citations that do not exist. The complainant may not know this, particularly if they have asked the AI tool to "find precedent" or "support this with authority" and accepted what the tool produced.
Why it is tricky. Fabricated references are not always immediately obvious. A case name and citation in correct AustLII format, accompanied by a one-line summary of the supposed holding, can look entirely legitimate until you search for it. Two operational risks follow: handlers can spend significant time trying to locate references that do not exist, and a complaint relying on fabricated authority can affect the assessment if it is not caught.
How to respond. Where a complaint cites specific case authority or statutory reference, verify it. AustLII (austlii.edu.au) is the primary verification source for Australian case law; the Federal Register of Legislation (legislation.gov.au) is the primary source for Commonwealth Acts; relevant State and Territory equivalents are used for State and Territory references. Where a reference cannot be located, the right next step is usually a clarification request rather than a dismissal: "We were unable to locate the case you referenced at [paragraph X]. Could you please provide a copy of the decision or a link to where it can be located? If the reference cannot be substantiated, we will assess your complaint on the basis of the other material you have provided."
This is fair, defensible, and gives the complainant an opportunity to correct without being penalised for having relied on an AI tool that produced material they did not realise was fabricated. The Australian Federal Court has been clear that the responsibility for material relied upon in proceedings rests with the party putting it forward, not with the tool that produced it. The same principle applies in complaint handling.
Pattern 4: The vulnerability-driven submission
What it looks like. A complaint written by someone who could not have articulated it without AI assistance. The complainant has a disability, a language barrier, low literacy, mental health distress, or trauma that has made it difficult for them to put their concern into words. They have used AI to help. The result is a submission that is more articulate than the complainant could have produced unaided, and which masks the underlying vulnerability of the person making it.
Why it is tricky. This pattern is the most ethically loaded of the five. AI use in this context is, in many ways, exactly what AI should be used for: levelling access to formal processes for people who might otherwise be excluded. But it can also mean that the complaint handler does not see the vulnerability the system is supposed to recognise. The risk is not the AI use. The risk is missing the person.
How to respond. Look past the polish to the person. Specific signals that may indicate underlying vulnerability include inconsistencies between the formal register of the written submission and the complainant's communication style on the phone, references to circumstances that suggest distress or difficulty (medical issues, recent loss, financial hardship), and complaints that escalate quickly when a clarifying question is asked. None of these are diagnostic on their own. Taken together, they may signal a complainant who needs more careful handling than the polished submission suggests.
A useful first move is a phone call rather than a written response, where the complainant is willing. The phone reveals what writing can mask. If a phone call is not possible or appropriate, a written response that acknowledges the substance, offers process information in plain language, and signals openness to providing support, often produces a more honest second exchange.
This is not about being suspicious of AI assistance. It is about being curious about the person behind the submission, regardless of how the submission was prepared.
Pattern 5: The pattern complainant using AI
What it looks like. A complainant with a history of repeated or extensive complaints, who has now adopted AI as a force multiplier. The same patterns of behaviour that previously produced thirty-page handwritten submissions now produce hundred-page AI-amplified submissions, with the same core grievance restated across multiple channels and multiple matters. The complainant is using AI to scale their existing complaint behaviour, not to articulate a new concern.
Why it is tricky. This pattern intersects with established practice on unreasonable complainant conduct, but it is not the same thing. Most jurisdictions have policies and guidance for managing patterns of unreasonable conduct (the NSW Ombudsman Unreasonable Complainant Conduct Manual is the foundational text for most Australian Ombudsman offices and many regulators adapt it). AI does not change the underlying conduct, but it does change the operational scale and the documentation burden.
How to respond. The response is not new policy; it is the existing unreasonable complainant conduct framework applied with awareness that the volume of material may be AI-amplified. Three operational points are worth noting.
First, the volume of material does not itself constitute unreasonable conduct. The question is whether the conduct (not the volume) meets the threshold in your policy.
Second, AI-amplified submissions can take significant time to process even when the underlying conduct has been addressed. Reasonable limits on response length, frequency and channel may be appropriate, applied in the way your existing policy contemplates.
Third, where AI is being used to circumvent restrictions placed on a complainant (for example, to generate variations of the same submission across multiple email accounts), this is a behavioural issue rather than an AI issue, and the existing tools apply.
This is the pattern where the question "is this AI-influenced complaint behaviour, or is this unreasonable complainant conduct that happens to use AI" matters most. The answer affects both the response and the documentation. In most cases, the latter framing is more useful.
What ties the patterns together
A useful frame across all five patterns is this: AI changes how complaints arrive, not what complaints are. The substantive question is still whether there is a genuine concern to be assessed, what the appropriate response is, and how to handle the matter fairly. The AI is context, not content.
The complaint handlers who navigate AI-influenced complaints well share three habits.
They read past the form to the substance. Whatever the submission looks like, the question is whether there is a concern that needs to be addressed and what the right next step is.
They ask better clarifying questions earlier. A short, specific clarifying question at intake or early assessment resolves more AI-influenced complaints than any other single intervention.
They document AI-related decisions on the file. Where verification has been undertaken, where a vulnerability concern has been considered, where a clarification request has been issued: noting the work on the file supports both quality assurance and procedural fairness.
These are not new habits. They are the habits experienced complaint handlers already use, applied with awareness of what AI is now changing in the submissions that arrive.
Free guide: Responding to AI-Generated Complaints
A practical intake tool for complaint handlers, conciliators and intake officers working through AI-assisted correspondence. Focus on substance, ask better questions, and progress matters fairly. Download the guide here.
What experienced complaint handlers know is that the work has always involved making fair decisions on imperfect material, in a system that does not always make it easy. AI is the latest condition that complaint handlers are working within. It is not the first, and it will not be the last.
The five patterns above are what I see in the work right now. They will change. The next iteration of the work will involve patterns we cannot fully predict, including AI tools that respond in real time, voice-cloned phone submissions, and AI agents that file complaints on behalf of complainants. The patterns will keep evolving. The underlying skill, reading past the form to the substance, asking the better question at the right time, holding fair process steady, does not change.
If your team is working through AI-influenced complaints and the patterns above are showing up in your caseload, the conversation worth having is not "how do we stop this" but "how do we get good at working with it." That is what the consulting work is for.
Hi, I'm Shiv Martin. I'm a nationally accredited mediator, lawyer, conciliator, and conflict management specialist with over a decade of experience working across government, business, and community settings. I support teams to navigate complex and emotionally charged situations through mediation and conciliation, conflict skills training, facilitation, and practical advice on policies and processes. My approach is grounded in law, psychology, and real-world dispute resolution, with a strong focus on clarity, fairness, and workable outcomes.
If you'd like to talk about how I can help you or your organisation, you can get in touch here: 👉 Contact us
Comments