How do LLMs represent public services to the public they serve?

How do I navigate aged care in Australia? How easy are government services to deal with? How well supported are people with disabilities? These are big questions about life in Australia, and many people now put them to an AI tool before they visit a website, pick up the phone or speak to anyone. They rarely mention a government agency by name, yet the answer they get still shapes what they believe about the services behind it: whether the system works, who is responsible and whose information they can trust. That is reputation taking shape, in a conversation no communications team is part of.

To see how that plays out, we used Your Prompts, a new feature in Isentia’s AI View, to track how leading AI models answer these questions. Here’s what we found, whose information the models used, how each model goes about sourcing its answers, and what it all means for public-sector communicators.

Starting with the question

Most AI visibility tracking starts with an organisation’s name. Your Prompts lets you set your own questions, which gives public-sector teams a much wider view of their reputation. People rarely ask an AI tool about an agency by name; they ask for help with whatever problem is in front of them, whether that’s getting a parent into aged care or working out what support is available for a carer. Or they ask something bigger, such as how well the country looks after older people.

We tracked six sets of questions across leading AI models, covering the things people deal with at different stages of life: finding their way around government services, getting older, living with disability and caring for someone, raising a family, staying healthy, and making ends meet. For this article we focused on three of them: access and navigation, aged care, and disability and carers. These are the moments when people depend most directly on the government, which means a wrong or incomplete answer can have real consequences for someone trying to get help, and can colour how they feel about the system as a whole.

We compared answers from ChatGPT, Claude, Gemini, Perplexity and Google AI Overview, looking at what each model said, which sources it cited and which organisations it named. The questions were set up for tracking rather than drawn from a survey of what the public actually asks, and the findings reflect a single point in time, using the latest data from 27 October 2026. Even so, the patterns in how AI finds and presents information about public services were clear and consistent.

How easy are government services to deal with in Australia?

We asked this alongside a second question: how well government support reaches the people who need it. Most models landed on a similar view: digital services work well for simple tasks, while complex cases, phone contact and people in vulnerable situations are where things get harder. The more interesting part is where that view comes from. The models formed it largely through oversight bodies, with the national audit office’s findings carrying particular weight.

Government’s own voice only came through when a model cited official pages, as when one model described digital services as a major strength, using a government website as the source. In other words, when AI assesses how the government is performing, it largely takes the word of the ‘referees’. Owned content can earn you a line in the answer, but auditors, commissions and advocates write most of the scorecard. For reputation, that gives an audit finding a second life. Well after the news cycle moves on, it can still be part of the answer people get when they ask how the government is doing. That brings us back to a question we raised in our article on AI visibility for government agencies: is your organisation still seen as the main authority on its own services?

How do I navigate aged care in Australia?

Almost every model named government services as the place to go for financial assessments, payments and in-person help, and the tone was neutral to positive. One model went as far as recommending that people complete a means assessment early so they don’t end up paying more than they need to.

The catch is that the agencies people were being sent to were rarely the source of that information, which mostly came from other government portals and guides written by aged care providers and advocacy groups. In one answer, even an agency’s own contact details were taken from a seniors’ advocacy group’s website. Being named in an answer, then, is quite different from being its source. It builds recognition, but someone else is setting people’s expectations of what they’ll find when they get there, and if that guide is out of date, it’s the service they feel let down by.

Zoom out to the bigger question of how well Australia looks after older people, and the agencies that were the first port of call in the practical answers didn’t appear at all. These answers were framed by the Royal Commission into Aged Care Quality and Safety, which concluded in 2021, with most models opening on its findings and some giving them more weight than the reforms that followed. This is the reputational point that’s easiest to miss: the practical question tells people where to go, while the big question tells them what kind of system they’re dealing with. As we covered in why content freshness matters for AI visibility, old stories have a habit of shaping AI answers long after the facts have moved on.

How well supported are people with disabilities in Australia?

The models discussed income support payments without naming the agency that delivers them, focusing instead on the NDIS and the gaps in support. That said, not every absence is a problem. Some questions genuinely aren’t about a particular organisation, so it makes sense for it to sit them out. The useful distinction for communicators is between a relevant absence, where the question simply isn’t about you, and a missed opportunity, where your work is being discussed and you aren’t getting the credit. Repeated across enough answers, a missed opportunity leaves people with a picture of support that doesn’t include the people delivering it.

Questions about caring for someone played out differently. Every model gave almost the same answer and pointed people to the same national carer support service. When official information is clear and consistent across agencies, peak bodies and support services, AI repeats it faithfully. There was one notable exception, where a model explained the government’s carer payments using a blog from a private company that sells personal medical alarms. The information was mostly accurate, but it shows how far an explanation can travel from its original source. The good news is that consistency across your wider network really does pay off: when the organisations around you describe your services the way you do, AI is more likely to get them right, and that shared story becomes part of your reputation.

Why each model tells the story differently

Across all three categories, each model showed clear habits in the sources it reached for. We can see what each one cites, even if we can’t see exactly how it decides, and the patterns are consistent enough to act on. In practice, your reputation can look different depending on the tool someone uses. These habits also build on what we found when we looked at how different LLMs represent your organisation.

One question, several versions

Put together, the same question can produce an audit-style answer, an advocacy-informed analysis, a research summary or a current snapshot, depending on which tool someone happens to open. Two people asking the same question can come away with quite different impressions of the same service. There’s no single AI view of an organisation to manage, because each model has its own sources and each needs a slightly different approach.

Zooming out: how these answers add up to reputation

Look at any one of these answers and it can seem like a visibility issue. Put them together and it’s a reputation issue. For public-sector organisations, reputation has always been about more than what’s said about you by name. It rests on whether people believe the system works, know where to turn and trust what they’re told, and those impressions are now forming in AI answers, one question at a time.

Across these questions, four things shaped the impression an answer left:

• Who gets named. Whether you appear when your work is being discussed, and get the credit for it.
• Who does the explaining. Whether people hear how your services work from you, or from a provider, an advocate or a company blog.
• Who acts as referee. Whose assessment of your performance gets repeated as the verdict.
• Which story sets the frame. Whether the answer opens on where things stand today, or on a finding from years ago.

None of this shows up if you only ask AI about your organisation by name. You see it when you ask the questions your audience is asking. And it matters, because AI is now one of the main places where people go for information about the things that matter to them.

What this means for public-sector communicators

Government organisations can’t control what AI says, but they can do a lot to influence the information it draws on, and with it the reputation that forms around them. Five practical steps came out of this research.

• Track the big questions, not just your name. The most useful insights came from questions about services and real-life situations rather than about any one organisation, so start with what your audience is trying to do or understand, and check what AI is telling them.
• Work with the bodies that assess your performance. Auditors, commissions and advocates shape how AI judges your work, so respond to their findings publicly and make sure your progress is recorded in the sources they, and AI, rely on. That way an old finding doesn’t have to stand alone.
• Support the organisations that explain your services. Sister agencies, peak bodies, providers and advocacy groups often explain your services for you, so give them clear, current information to work from. As we found in our analysis of Tier 1 and Tier 2 media in AI answers, specialist and niche sources often matter more to AI than their profile suggests.
• Keep practical information clear and up to date. Plain-language pages on eligibility, processes and how to get help are what AI repeats when people ask how to do something. Organisations that deliver services directly rely on this more than most, because outdated information can lead someone to miss out on support they’re entitled to.
• Check answers across several models regularly. Each model uses different sources, so one model’s answer is a poor guide to the others, and because answers shift as new content is published, regular tracking shows whether your updates are reaching the sources AI uses.

Whose information is AI using?

AI is already answering big questions about life in Australia, whether or not the public-sector organisations involved are part of the conversation. This research shows that an organisation can be mentioned often and still have surprisingly little say in how its services are described or how its performance is judged, with other organisations doing much of the explaining and each AI model weighing those voices differently.

For communicators, the job is to understand that wider network of sources and make sure accurate, current information runs through all of it. It’s worth asking what AI says about your organisation. It’s just as worthwhile asking what AI says when nobody mentions you, and whose information it’s using to say it. That’s where much of your reputation is now taking shape.

Want to see what AI tells people about the questions that matter to your audience? Get in touch about AI View.

The post How do LLMs represent public services to the public they serve? appeared first on Isentia.