AI Could Help More People Be Heard. Institutions Still Have to Listen.

From navigating a complaint to questioning a political claim, AI could help people make their case. For neurodivergent users, that possibility deserves serious attention and evidence about what actually helps.

A complaint sitting unfinished in someone’s drafts cannot get an incorrect bill fixed. A folder full of evidence is little help if its owner cannot work out where to begin. Having a legitimate concern and being able to present it through a formal process are different things. Someone may remember every detail of what happened but struggle to put events in order. They may understand the problem perfectly while finding the forms, deadlines and correspondence exhausting.

This is one of the more useful possibilities for artificial intelligence: helping people get from knowing something is wrong to being able to explain it and ask for a response. My interest in this comes partly from working in IT. A system can function exactly as designed and still be difficult for the person who needs it. We should judge technology by whether it helps people complete the task in front of them. Used carefully, AI could make some of that work easier. The harder question is whether institutions will respond fairly once people make it through the paperwork.

Helping people make the case for change

Imagine a resident who wants a safer pedestrian crossing near a local school. They have photographs, notes about near misses and concerns shared by other parents. They know why something needs to change, but turning that information into a clear request to council feels overwhelming. An AI assistant could help organise their notes, outline a letter and explain what information might strengthen their case. The resident would still need to check every claim, protect other people’s privacy and make sure the final wording reflects what they want to say.

That assistance could help someone take their first step into political participation. A concern discussed at the school gate could become a specific request that councillors can consider and residents can follow up. There is research supporting AI’s usefulness for some writing work. In a 2023 experiment published in Science, Shakked Noy and Whitney Zhang assigned professional writing tasks to 453 university-educated participants. Access to ChatGPT reduced average completion time by 40 per cent and increased assessed output quality by 18 per cent. [1]

Those findings concern particular writing tasks under experimental conditions. They do not prove that an AI-assisted letter will persuade a councillor or change government policy. They do suggest that AI can help with some of the writing involved in making a case. The same possibility extends to preparing a parliamentary submission, asking an MP to explain a policy or proposing improvements to a public service. People should not need professional writing skills to contribute to decisions that affect their lives. For someone who has the ideas but struggles to get them onto the page, a usable first draft could help them join a debate they would otherwise sit out.

Neurodivergent people should shape the support

Neurodivergent people have different needs and preferences. Some may welcome help breaking a large task into smaller steps. Others may find another stream of suggestions distracting or intrusive. Useful assistance might involve turning rough notes into an outline, explaining an instruction differently, or preparing questions before a meeting. The user should control how much help they receive and what happens to their words.

A 2026 study by Zihao Zhu, Junnan Yu and Yuhan Luo explored these possibilities through co-design sessions with 20 university students diagnosed with ADHD, followed by interviews with five specialists. Participants and experts considered support for understanding tasks, reflecting on progress and managing emotions around task engagement. [2] This was an exploratory design study. It did not demonstrate lasting improvements in ADHD symptoms or establish that the proposed tools work for neurodivergent people generally. Its contribution is identifying support worth developing and testing with the people who might use it.

That involvement matters. A tool that continually rewrites someone’s direct language into corporate politeness may make their message less accurate. Removing anger from an account of mistreatment can remove part of what the person is trying to communicate. Users should be able to ask for a clearer structure without surrendering their tone, personality or judgement. They should also be able to reject the suggested rewrite.

Employers have work to do here. An AI subscription cannot resolve contradictory instructions, an inaccessible complaints process or a manager who refuses to explain a decision. Workplace support should include clearer communication and a willingness to discuss adjustments. Employees should not have to solve every organisational problem through better prompting.

Public information should be easier to question

The same practical assistance could help citizens scrutinise political claims. Suppose a minister says waiting times have fallen. The evidence may be spread across annual reports, budget papers and tables using different definitions. A reader needs to establish which service is being measured, which people are included and whether the reporting periods are comparable.

An AI tool with access to those documents could help locate relevant passages and draw up questions for checking. Has the measure changed? Does it describe the typical wait or an average affected by unusually long cases? Are people still waiting included?

These are questions a careful researcher would ask. AI may help a reader find where to start, provided its references lead back to the actual evidence. For a small publication such as thewokely, that is an appealing use: organising source material and identifying gaps that need further reporting. Responsibility for the resulting story remains with the publication. There is a democratic argument for making this work easier. People should be able to examine the evidence behind a public claim without having to become specialists in every reporting system they encounter.

If a discrepancy needs investigation

Asking a chatbot to prove that a politician is lying starts with a conclusion. A better question is: what do these documents support, what contradicts the claim, and what remains unclear?

A discrepancy may reflect an outdated figure, a changed definition, an honest mistake or deliberate deception. Establishing dishonesty requires evidence about what the person knew and how they represented it. The tool can also be wrong. The US National Institute of Standards and Technology identifies “confabulation”confidently generated false or erroneous material—as a risk of generative AI. Its guidance includes verifying sources and citations. [3]

That makes checking the original documents essential. A citation can look convincing while failing to support the sentence beside it. A summary can omit the qualification that changes a finding’s meaning. Ask for the page or table. Open it. Reproduce important calculations independently. Look for evidence that challenges the initial interpretation, and give the person or institution concerned a fair opportunity to explain a discrepancy before alleging misconduct. For readers using AI to prepare a complaint, the same discipline protects their case. An invented event, quotation or legal reference can undermine a valid concern. The final document needs to remain an account they understand and can stand behind.

Institutions need to check their own systems

Organisations could also test AI assistance for routine checks: finding missing attachments, comparing a report with its underlying records or identifying inconsistencies that warrant a closer look. These uses need evidence of accuracy and a clear route for correcting errors. A flagged mismatch should prompt examination. It should not become an automatic finding of fraud or a reason to dismiss a complaint.

We already have an Australian example of why governance deserves attention. In an audit published on 24 February 2025, the Australian National Audit Office found that the Australian Taxation Office’s arrangements for adopting AI were partly effective. It identified shortcomings in assurance, the integration of ethical and legal considerations, and monitoring. The ATO agreed to all seven recommendations. [4]

These are findings from that audit period, rather than proof that every weakness persists today. They nevertheless show why an organisation’s confidence in its technology needs independent examination. Before relying on a system, an institution should know how often it produces the wrong result, whether errors fall more heavily on particular groups, and who has authority to correct them. Those checks need to continue after deployment. The substance of a complaint must remain the focus. People should not receive a worse hearing because their writing is unconventional, emotional or less polished than the organisation’s response.

Robodebt shows why review must have teeth

Robodebt involved automated processes and income averaging. It was not a generative AI chatbot, and describing it as one would obscure what went wrong. Its relevance here is the danger of acting on an automated result without an adequate basis. [5]

The Royal Commission’s recommendations give us practical standards for scrutinising automation. Recommendation 17.1 called for consideration of a consistent legal framework and specified clear review pathways, plain-language explanations of automated processes, and access to business rules and algorithms for independent scrutiny. Recommendation 17.2 proposed considering a body with powers to monitor and audit automated decision-making. [6]

Those principles should inform the use of AI in public services too. A person affected by a decision should be able to find out what information was used, correct errors and reach someone with the authority to reconsider the outcome. Calling a process “human review” achieves little if the reviewer cannot inspect the evidence or challenge the system. Nor should a citizen need their own AI subscription to obtain a fair hearing. Public services must remain usable by people who cannot afford these tools, cannot use them or prefer to speak to someone.

Privacy is part of whether the tool is useful

From an IT perspective, the information going into the system deserves as much attention as the answer coming out. A complaint may contain addresses, financial records, health information and private details about other people. Uploading it can create a separate problem even if the resulting letter is excellent. Australia’s privacy regulator recommends, as best practice, that organisations avoid entering personal information, particularly sensitive information, into publicly available generative AI tools. Its guidance also calls for organisations to assess suitability, privacy risks, access to information and human oversight when selecting AI products. [7]

Paying for a subscription does not settle those questions. The product, contractual terms, settings and intended use all matter. For an initial draft, it may be possible to use placeholders and leave out unnecessary details. That is a precaution, not a guarantee of anonymity: a distinctive set of circumstances can still identify someone. Sensitive work may require a controlled environment or direct human assistance. People seeking help should not have to trade away more privacy than the task requires.

Measure whether people get heard

AI’s value in this setting should be judged through outcomes people can recognise. Could someone understand a letter they previously found confusing? Did they finish a task with less difficulty? Was an incorrect record corrected? Did an institution answer the question it had been avoiding?

Those measures would tell us more than the number of prompts submitted or documents produced. They would also expose cases where the tool simply creates more work: another draft to check, another false reference to remove, another explanation that misses the point. There is a constructive case for AI here. It could help more people organise their evidence, understand unfamiliar information and speak for themselves. For some neurodivergent users, that may make an everyday task considerably more manageable. But a clearer letter is only the beginning. The person receiving it still has to read it, examine the evidence and be willing to put a mistake right.

References

1. Noy, S. and Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science, 381, 187–192. DOI and publication; PubMed abstract.
2. Zhu, Z., Yu, J. and Luo, Y. (2026). Scaffolding Metacognition with GenAI: Exploring Design Opportunities to Support Task Management for University Students with ADHD. Authors’ paper on arXiv; the record identifies acceptance for CHI 2026. Paper and abstract.
3. National Institute of Standards and Technology (2024). Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, NIST AI 600-1. See section 2.2 and action MS-2.5-003. Report.
4. Australian National Audit Office (2025). Governance of Artificial Intelligence at the Australian Taxation Office, published 24 February 2025. See conclusions and agency response. Audit.
5. University of Melbourne, Pursuit (2023). The flawed algorithm at the heart of Robodebt. Analysis.
6. Royal Commission into the Robodebt Scheme (2023). Report, recommendations 17.1 and 17.2. Report and recommendations.
7. Office of the Australian Information Commissioner (2024). Guidance on privacy and the use of commercially available AI products. See the top five takeaways and product-selection guidance. Guidance.