The short version
Most of the organisations in this paper are well known. Their leaders know the work, and people in their sector know who they are.
And the person who needs them most is often the one least able to make sense of how they explain themselves.
Not the member who renews every year, or the adviser who already has the CEO’s number. Think of the parent whose child has stopped going to school. The family who has cared for a son or daughter at home for decades and has run out of road. The young worker who has no idea she’s eligible for affordable housing. The homeowner who has just been ripped off by a tradie. (These are illustrations of patterns leaders described to me, not real people.)
These people don’t know the organisation’s name, and they don’t know the sector’s words for what it does. They know their problem. And they arrive with four questions:
Is this for someone like me? What does it get me? Why should I believe it? What do I do next?
If a website, or an AI assistant, can’t answer those four in the stranger’s own words, the stranger goes somewhere else. This paper is about how often that happens, and why.
On the website. I studied 50 Australian association websites. On only about half could a member log in and find an event, some training or a resource. And on nearly a third of those, a newcomer was left to answer the first question for themselves: is this even for me?
In how the site gets built. Leaders described sites built for funders, for the organisation’s image, or for whatever a small team can keep up with. Each is a reasonable choice. Together they’re why the stranger gets lost.
In AI assistants. More and more people type their problem into ChatGPT or Google before they reach a website. Roy Morgan found that 58% of Australians aged 14 and over used an AI tool in an average four weeks [S7]. So I asked ChatGPT, with no account, about 84 organisations. For each one I asked about the same needs twice: once in the words each organisation’s own website uses, and once in the words people actually use.
In the website’s words, ChatGPT recommended the organisation first nearly three times in four. In people’s own words, about a third of the time. Same organisations, same needs, different words.
For one trade association, asking ChatGPT for “a member [tradesperson]” got the association recommended first. Asking for “a [tradesperson] I can trust”, the way a homeowner actually would, it didn’t come up at all.
None of this needs a bigger website. Several of the leaders I spoke to run small teams and said so plainly. The fix is mostly a choice about who the front door is for, and whose words it uses.
How to read this
The evidence comes from three places: a study of 50 Australian association websites, ten interviews with leaders of Canberra associations and community organisations, and a test of what ChatGPT tells people about 84 organisations.
Every test was done anonymously, with no account signed in, the way a stranger would meet these organisations.
Small tags in the text show where each claim comes from: [Website study], [Interview], [AI test], a published source such as [S1], or [Inferred] when it’s my reasoning rather than something I observed. Discount the inferences as you see fit.
No organisation or person is named. Leaders appear by a general description of their role. The full method, the exact counts behind every number, and the limits of what this evidence can tell you are at the back.
The four questions
Everything in this paper comes back to four questions a stranger needs answered before they’ll pick up the phone, fill in a form or join.
- Is this for someone like me? Do I recognise my situation in what you say?
- What does it get me? What changes for me if I come to you?
- Why should I believe it? Who else have you helped, and how do I know you’re real?
- What do I do next? One clear step, that I can take now.
Members never ask them. They already know the answers. That’s why a website written for members can work perfectly for the people inside and still fail the person who needs it most. [Inferred]
Now they ask them twice. More and more, a stranger puts those questions to Google or an AI assistant first, in their own words, before opening any website. The assistant can only answer from what an organisation has published. If its pages answer the four questions in the stranger’s words, the assistant can find it and send the stranger there. If they answer only in the sector’s words, the stranger is sent somewhere else, and the organisation never hears about it. [Inferred, from the AI test]
The first question is the one leaders described most often. [Interviews]
Who comes to the digital front door
I asked leaders who comes to them not knowing them, and what happens at that first contact. The answers fell into three groups.
People who belong, and can’t tell. The prospective member, the partner, the policy adviser looking for a position. These are the outsiders the website study was built around, and the ones most association websites at least try to serve.
People who need you, and can’t find you. A youth charity manager described parents who need exactly the program his organisation runs. When they search, they’re sent to the school instead. A disability service CEO described families who have cared for a son or daughter at home for many years and reach the point where they need more support. “people don’t know what they don’t know”: whether respite exists, what it looks like, or that there are options between a group home and staying at home. A former housing executive described the core market for affordable housing as early-career workers, such as apprentices and early childhood workers on roughly $50,000 to $70,000, who “probably wouldn’t realise that they’re eligible”. In her words: “we’re missing the market.” [Interviews, 3]
People who need someone else, and find you anyway. An industry association fields consumer complaints and licensing questions that belong to the regulator. A trade association hears from people ripped off by operators who aren’t its members, who expect it to act as if it were a government authority. A national peak body passes disputes on to the bodies that can resolve them. A business membership organisation recently took a call from an owner caught in a commercial dispute, which it doesn’t handle, who had never heard of it. [Interviews, 4]
The second group is the one this paper keeps coming back to. The person who can’t make sense of your website is often the person who needs you most. If they can’t work out that you’re the answer, they end up somewhere worse. [Inferred]
Why they can’t find you
They can’t search for you because they don’t know the words.
A disability service CEO said families arrive with a funding plan they don’t understand: “people get a plan and they get certain lines in their plan, and they have no idea what it means.” His first conversation with a family is spent decoding the plan and working out what the person needs, both at once.
A youth charity manager said one of his organisation’s programs has a name nobody outside the sector recognises. [Interviews, 2]
They rule themselves out
The first question, “is this for someone like me?”, is often answered wrongly before anyone calls.
A former housing executive said that “as soon as you say community housing, people think public housing,” so the people her organisation could help assume they’d need to be on benefits, and don’t look. [Interview]
The executive director of a large community services organisation said the misconception she fights most is that it only helps people of one religion. It has rewritten its mission and values statements, and people still assume it. [Interview]
The chief executive of a business membership organisation hears wrong guesses from both directions: “you’re only for big business. You’re only for small business. So, you know, there’s a completely contradictory view.” In fact, he says, “we’re here for everyone.” [Interview]
It’s the same mistake in three different sectors. The stranger makes a guess about who you’re for, the guess is wrong, and they never find out. [Inferred]
Not everyone agreed
Another disability service CEO thinks families with an older family member in the system generally know the NDIS well. [Interview]
The community services executive pushed back too. In her experience, people who reach out generally know what they need, and when a program page names the problem, a Google search often finds it. Parents ring asking for a program by name because a search led them to its page. [Interview]
Bigger than one organisation
Sometimes the problem is bigger than one organisation. If people don’t understand what kind of organisation you are, no website can fix that on its own. The CEO of a community-sector intermediary said her model is much less familiar in Australia than overseas, and that many Australians assume a charity sends every dollar straight out the door. An organisation that invests for the long term is, in her words, “a piece of social infrastructure effectively”, and “it’s certainly not going to be something that people immediately get from the website.” [Interview] Explaining a whole category takes the whole sector. [Inferred]
A familiar name doesn’t solve it either. A disability service CEO said his own organisation’s name is well known, kept in view by community events and fundraising. People know the name. Far fewer know what the organisation does. At one of its events, he estimates only about half the people who come know the organisation runs it, and even they aren’t sure what it does. People who sponsor its fundraisers often don’t know what their money pays for. [Interview]
So when a family reaches the point of needing that support, the familiar name doesn’t come to mind as the answer. Events and fundraising build recognition. They don’t explain the work. Something has to say plainly what you do, for whom, and how to start, and for most strangers the first place that can happen is your website. [Inferred]
The website study found the same gap
I studied 50 Australian association websites the way an outsider meets them: no login, no inside knowledge. We simulated six kinds of visitor, from a prospective member to a journalist on deadline, and recorded what each could find. [Website study]
On only about half the sites could an existing member log in and find an event, some training or a resource. That’s the part of the site most associations invest in.
On nearly a third of those sites, our simulated newcomer couldn’t tell whether membership was for them.
A newcomer could tell it was for themA newcomer was left unsure
A site can let a member do everything they came for and still leave a newcomer stuck on the first question. Members arrive knowing the acronyms, the committee names and what “the portal” is. Everyone else arrives with a problem. A site written by insiders, for insiders, assumes the visitor already knows their way around. [Inferred]
What the AI assistants say
Before any of the four questions, the stranger has to reach you. If they ask an AI assistant before they open a website, what does it tell them? I asked ChatGPT, with no account and a fresh chat for every question, about 84 organisations: the 50 associations from the website study, 27 Canberra associations and community organisations, and seven organisations connected to my interviews. [AI test]
For each one, I picked two needs the organisation exists to meet and asked about each need twice.
- In the website’s words. The organisation’s own term for the help, such as “a member builder”, “supported independent living” or a program’s own name.
- In people’s words. The way someone with no inside knowledge would ask. For the seven organisations connected to my interviews, these came from what leaders said people bring them. For the rest, I used a real question someone had posted online where I could find one, and otherwise wrote it in plain everyday words, with none of the organisation’s own terms. A separate check, done without knowing which organisation each question was about, weeded out any that gave the answer away.
Same organisation, same need, same situation. Only the words changed. Each question was asked once.
| What I asked | What ChatGPT did |
|---|---|
| “What is [organisation]?” | Described it correctly almost every time |
| A need, in the organisation’s own website words | Mentioned the organisation nearly nine times in ten. Recommended it first nearly three times in four |
| The same need, in people’s own words | Mentioned the organisation about six times in ten. Recommended it first about a third of the time |
The words made a big difference. For the same need, the website’s wording did better in about half of all cases. People’s wording did better only rarely, and the rest came out the same. [AI test] Some website terms are close to a brand name, so the website-words result is close to a best case. Even so, the people’s-words result is the one a stranger actually gets. [Inferred]
For nearly half the organisations, ChatGPT never recommended them first for either need when it was asked in people’s words. For one in five, it never mentioned them at all. [AI test]
The gap was much wider for the 50 national associations than for the Canberra organisations, where it was smaller but still clear. Part of the national gap may come from ChatGPT assuming the person was in Canberra, which suits Canberra organisations whatever the wording. [AI test]
The seven organisations connected to my interviews fared worst. Even in their own website’s words, ChatGPT recommended them first only about a third of the time. In people’s words, it mentioned them in a fifth of answers and never sent anyone to them first. [AI test]
Being mentioned isn’t the same as being recommended. When ChatGPT did mention an organisation in answer to people’s words, four times in ten it sent them somewhere else first. Fourth in a list of five doesn’t help someone who’ll ring the first. [AI test]
When it didn’t send people to the organisation, it often sent them to a government service or regulator first (the Fair Work Ombudsman, a licence register, a housing eligibility checker), or to another provider or a business. For some needs that’s the right answer. A homeowner who has been overcharged probably should start with the regulator. But for the needs an organisation exists to meet, it means someone else is getting the call. [AI test]
Some needs are adjacent. A homeowner looking for a tradesperson they can trust is really asking for one of a trade association’s members. Asked that way, ChatGPT tended to send them to a licence register and a list of local businesses, and in one answer to the equivalent association in another state. Asked for “a member [tradesperson]”, it went straight to the association. Homeowners don’t talk about member tradespeople. [AI test]
A youth charity manager had found the same thing on his own: ask an AI assistant about his organisation by name and it knows exactly who you mean; ask it the problem the organisation answers and it doesn’t. [Interview]
Why it happens
An AI assistant builds its answer from what has been published, by an organisation and about it. A disability service CEO said it before I’d shown anyone a result: “how does AI collect all that? It’s all of your footprint and your presence… It is your website and your marketing.” [Interview]
A former housing executive went further. It isn’t only the website: “the articles, the advocacy piece, the things outside of the website then flavour it.” [Interview]
ChatGPT searched the web for almost every answer. If your website describes your services only in your sector’s words, those are the words it can find you by. A stranger who uses different words gets sent somewhere else. [Inferred]
Few of the leaders I spoke to had looked. Two had never asked an AI assistant about their own organisation. One had done the work for Google but not yet for AI, because “2 years ago it wasn’t something to worry about” Where AI had come up, it was for the back office. A former housing executive said “the AI use has been limited to some of the administration efficiency rather than broader thinking”, and the community services executive said it comes up for intake, not yet for how people find them. [Interviews] The organisations most at risk of being missed are mostly not checking. [Inferred]
Three things for the executive team and the board
- You’re findable in your own language. The organisations in this test weren’t invisible. ChatGPT usually found them when asked in their own terms. The people who need them don’t use those terms. [Inferred, from the AI test]
- A descriptive name can hide you. Two of the 84 organisations have names that are also everyday phrases. Asked “What is [name]?”, ChatGPT explained the phrase instead of the organisation. For a third, it half did the same. [AI test]
- Say it in their words, somewhere. Your website can keep its proper service names. It also needs the words a worried parent or a stressed homeowner would type, or an AI assistant has nothing to connect them to you. [Inferred]
Two cautions. These are snapshots, and AI assistants change their answers from week to week. And an AI assistant isn’t the only route: a family who reaches a provider through a support coordinator may never ask one.
But the people with nobody to ask are the ones most likely to ask AI or search Google. [Inferred]
Run it on your own organisation
This takes about 30 minutes. It’s one exercise, done in this order.
Step 1. Open ChatGPT, or another AI assistant, in a private browser window, so you’re not signed in and it doesn’t know who you are.
Step 2. Ask “What is [your organisation]?” If it gets you wrong, start there.
Step 3. Ask whoever answers your phone for the three problems people most often bring you, in the caller’s own words. Not “respite services”, but “I need a break from caring for my son.”
Step 4. Ask each of those three in a fresh chat. For each answer, note whether you’re mentioned, whether you’re the first place it sends people, and where it sends them instead.
Step 5. Now ask the same three problems the way your website describes them, again in fresh chats, and note the same things.
Step 6. Compare. If you’re found in your words but not in theirs, your website speaks your sector’s language and not your community’s. If you’re mentioned but rarely first, look at who is first and what their site says that yours doesn’t.
Answers change from one ask to the next. If you can, repeat step 4 on another day before you draw conclusions.
Why capable organisations build the wrong front door
None of the leaders I interviewed were careless about their websites. They made reasonable decisions that add up to a site built for the wrong visitor. Three reasons came up.
1. The site is built for the people who already know you
A former housing executive said her previous organisation’s site was built “for information only so that funders and people who land on it can see what we do”, not as a place for someone to find a home. [Interview]
An industry association director put a common failure down to personalities and ego. [Interview] When the people signing off a website are thinking about how the organisation looks, the site ends up serving its image. The stranger with a problem comes second. [Inferred]
2. Communications gets treated as optional
The CEO of a community-sector intermediary works with a large network of charities. Many of their CEOs come from frontline work, she said, and see marketing and communications as optional extras, a view their thin resourcing seems to justify. Her answer: “how will anyone come and join you in the work, whether through time or money, whatever gifts or talents they have to bring if they don’t know who you are?” The charities she sees moving fastest have CEOs who treat their brand as a way to bring in resources. She called it a structural issue, and one boards should know about. [Interview, drawing on her view across many charities]
3. The team is small
Five leaders raised resources. The peak body COO: “the vast majority of industry associations in my experience are actually quite strapped and poorly resourced and might have quite small teams.” An industry association director named the two ways it goes wrong: organisations either put too much on the site and can’t keep up with the people who engage, or undersell and wonder why nobody calls. [Interviews, 5]
Published sector research says the same. The Associations Forum’s 2026 survey of 216 organisations found stagnant team sizes creating operational bottlenecks [S1]. In an AuSAE index published in November 2024, 84% of association CEOs said they frequently or occasionally lack the time to do their role; its sample size isn’t disclosed [S2].
The answer leaders reached on their own was the same: size the site to the team that runs it. A trade association CEO gave an inherited, out-of-date site a quick revamp with just the basics, deliberately keeping it simple. [Interview]
When the front door should be a conversation
Two disability service CEOs pushed back on the premise, and they’re right to.
One leaves the offer open at first contact on purpose: “rather than us saying this is what we do, we say what do you need?” The other said that listing everything his organisation can do on its website would backfire: “I think we get people lost.” When support is tailored to each person, the honest front door is a conversation. [Interviews, 2]
A trade association CEO said his members ring rather than use the website, and new members come almost entirely through one-to-one contact. [Interview]
One chief executive said the same: most new members arrive through existing members’ networks, and few get in touch without an introduction, and sometimes “we drop the ball in terms of follow-up.” [Interview]
I take that seriously. It changes what the website is for, not whether it matters. The same trade association CEO still sees the website as the front door for everyone who isn’t a member: the website is the front doorstep for everybody else. [Interview]
And a conversation needs a front door too. One leader described a single public phone number that lands callers with whichever of several offices picks up. Being put on hold and passed from one to the next is among the most common complaints the organisation gets. [Interview] A phone number can be a switchboard rather than a front door. [Inferred]
So the website’s job isn’t to replace the conversation. It’s to get the right stranger to the right conversation, by answering the four questions clearly enough that they know who you help and how to start. And someone has to pick that conversation up. [Inferred]
The four jobs of the public layer
By your public layer I mean everything an outsider can see and do without logging in: your website, and what search engines and AI assistants say about you.
The website study looked at four jobs that layer has to do for outsiders, and each one leans on the four questions. Recruit is mostly “Is this for someone like me?” and “What does it get me?”. Influence and partnership are mostly “Why should I believe it?”. Perform, and the referral route, are mostly “What do I do next?”. The interviews, which went wider than associations, supported all four jobs, and added an audience the study had missed.
Recruit: can the right outsider recognise that you’re for them?
This is the job the four questions describe. A list of benefits answers only the second one: what it gets me. It leaves the stranger to work out the rest.
A youth charity manager put the fix in one line: present what you fix, not how you fix it. [Interview]
There’s a second problem the website study couldn’t see. An industry association director has long-standing members who still don’t know what their membership includes. [Interview] If long-standing members can’t answer the second question, a newcomer has little chance.
A test any association can run this week: read your public benefits page and your fee schedule side by side, as a stranger would. Different people usually wrote them, at different times, for different reasons. A prospective member reads them as one document.
Partner: can a prospective partner see what’s on offer and how to start?
Of the sites that clearly invited sponsors or partners, only one in ten published what sponsorship costs, and half didn’t make it clear either way. That’s a small group of sites, so read it as a pattern. [Website study]
Keeping prices private can be exactly right. The question is whether the route to a partnership conversation was designed, or just happened.
The interviews add what partners want: to be seen. An industry association director said partners want their name on the site, who they engage with, and the outcomes they helped deliver. [Interview]
Research from North America shows the pressure. ASAE’s 2026 State of Associations report, based on quick surveys of undisclosed size, found nearly two-thirds of associations pursuing partnerships and more than 60% diversifying revenue [S3]. Naylor’s 2026 benchmarking of 665 association professionals found that income other than membership fees was the top challenge, named by just over half [S4]. Naylor sells to the sector, and both come from North America, so read them as context.
Influence: can a journalist or policy adviser find your position and the right person, fast?
We simulated a journalist on deadline at every site. The task: find the organisation’s current position on an issue, the evidence behind it, who can speak to it, and how to reach them.
Even on sites where a visitor could easily find the organisation’s experts or leaders, the journalist couldn’t finish the job nearly half the time. Sites with fresh, dated news did no better: on just over half of them, the journalist came up short. [Website study]
A journalist could finish the taskCouldn’t finish
An outsider can see the experts and the news and still not get what they came for. It’s a library with good books and no catalogue.
An industry association director heard the same thing from the other side. Politicians and their staff tell him submissions are hard to find, and so is the right person to talk to on a topic. They’ve asked him to make it easier. [Interview, relaying what politicians and staff told him]
The difference is between having authority and packaging it so someone can use it. A deep library of submissions and media releases still leaves a time-poor adviser to work out the current position, find the evidence and figure out who can speak. A 2026 study of 70 Brussels policy communicators, 56% of them from trade associations, found 93% used their websites mainly to showcase their work, and only 19% to drive engagement with the people they wanted to influence [S6]. It’s small and a long way from Canberra, but it describes the same habit.
The audience the sector under-counts, and question three. Every corporate member has someone inside their own organisation who eventually asks what the subscription buys: a finance director, a new general manager, a procurement review. The person who signed up has to answer without the association in the room. A visible, current record of what the association has won is that answer. [Inferred]
Perform: does the site work for the visitor?
Picture a parent on their phone in a car park, ten minutes between school pick-up and getting back to work, finally looking for the help a teacher mentioned. They tap a link. The screen stays blank.
In the website study, the typical site took nearly nine seconds to show its main content, more than three times the 2.5 seconds Google describes as good [S8]. Almost none of the sites we could measure got under that mark, and some took far longer. [Website study: one controlled lab test, not real visitors’ experience]
Good: 2.5 seconds or less (2 of 42 sites)Slower than good (40 of 42 sites)
Nine seconds is a long time to stare at a blank screen, and people don’t wait. Google’s analysis of millions of page loads found that when sites met its speed standards, visitors were 24% less likely to give up before anything appeared [S9]. In a study of more than 10 million visits to one large company’s websites, a one-second improvement in showing the main content went with noticeably fewer people leaving after the first page: 14 in every 100 fewer [S10].
Both studies come from commercial websites, and a lab test is usually slower than what real visitors see. But the direction is clear. Every second of waiting loses some of the people you most want to reach. And the first to go are likely the ones with the least time and patience left: the stressed parent, the carer at the end of their tether, the adviser on deadline. [Inferred]
Speed is only one part. Can people, and the AI assistants they use, find the right page? Can a visitor finish what they came to do? Can a small team keep it current? One leader described a cheap pattern worth copying. During appeals, or after an event that brings people to the site, her organisation puts a box on the front page that says, in effect, “we think you’re here for this, and here’s how.” Her standard: “when people land there, they shouldn’t have to go several steps before they find how to give money.” [Interview]
The missing audience: the people who refer
The website study treated the outsider as someone who might join, partner or quote you. The interviews kept pointing at someone else: the referrer.
A youth charity manager put it most clearly. In youth services, a parent may not think there’s a problem and the young person won’t search, so the real audience is schools, justice, police and other services. [Interview] The community services executive said the same of her own sector: her organisation relies heavily on other professionals knowing who it is. [Interview] Most of the community organisations in this paper rely on someone else to send people to them. [Inferred]
If a referrer can’t quickly tell what you do, who you take and how to refer, the person who needs you never arrives. For any service that relies on referrals, the referrer is the first outsider your website has to serve. [Inferred]
A case in the organisation’s own words
Almost every finding so far is either an outsider’s reading of a website or a leader’s account of their own organisation. This one is different: an organisation’s own members, in its own survey, published in its own annual report.
The organisation is a national industry peak body. It isn’t named. It isn’t a client, it wasn’t interviewed for this paper, and nothing here comes from a private conversation. Everything below is from its most recent annual report, read on 25 August 2026 and checked again on 18 October 2026. [S5]
On Influence. In a member survey, members rated the organisation well and named advocacy as its strongest asset. They also named advocacy as the area where they most wanted to see the wins and outcomes more clearly. The report itself records the year’s advocacy: a policy outcome at federal level, and representations to several state governments, made through submissions, roundtables and meetings.
The members can’t see a record that exists. That’s question three, “Why should I believe it?”, going unanswered, and it points to a visibility problem. [Inferred] It’s also a problem for the public layer, not only for member communications, because the people a member has to justify the subscription to (a finance director, a board) are outsiders who will go looking.
On Recruit. Membership income fell slightly over the year. The report welcomes several new members and puts the fall down to membership movements, including some larger members lost through amalgamations. It sets an objective to broaden the member base and lift revenue. A website doesn’t fix a member mix. It does decide whether a prospective member in a target group can work out what they’d be buying. [Inferred]
One case shows this can happen in a real organisation. It doesn’t show how common it is.
Questions for the executive team
These are questions for the executive team to work through. Bring the answers to the board when there’s an investment to decide.
Start with the four questions. Open your website as a stranger in each of your main audiences, and check whether it answers them, in that stranger’s words, on the first page they’d land on.
| The question | What to look at | |
|---|---|---|
| The four questions | Can each kind of stranger tell, quickly, whether we’re for them, what it gets them, why they should believe us and what to do next? | The first page each audience lands on |
| Need | If someone typed the problem we solve, in their own words, into an AI assistant tonight, would it send them to us first? | The six steps in “Run it on your own organisation” |
| Referrers | Can a school, a support coordinator or a GP tell in a minute who we take and how to refer? | The referral route, eligibility, response time |
| Partner | Can a partner see what’s on offer and how to start the conversation? | Audience, opportunities, proof, the route |
| Influence | Can an adviser on deadline find our position, the evidence and the right person? | Current position, evidence, spokesperson, contact |
| Perform | Can the site do all of that, quickly, and can our team keep it current? | Speed, findability, finishing the task, upkeep |
| The brief | Who briefed our current site, and did they ask the people who answer the phone? | The brief, if one exists |
Don’t add the answers up into a score. The most useful result may be “we do the work, and the site doesn’t say so.”
What you can do this week
- The stranger test. Find someone who fits your audience and doesn’t know you. Ask them to answer the four questions from your website alone. Watch, and don’t help.
- The need test. Run the six steps in “Run it on your own organisation”.
- The five-minute authority test. Take your biggest win of the past year. Open your site as a stranger, start a timer, and find the position, the evidence, the person who can speak and how to reach them. Would an adviser on deadline have got there?
- The phone test. Ask whoever answers your phone for the five questions they get most from people who don’t know you. Then check whether your site answers them on the first page those people land on.
What good looks like
Leaders named organisations they think explain themselves well. Four patterns came through.
- One message, carried by everyone. A former housing executive pointed to a small charity whose message runs through its staff, board and events, not just its website, and which answers enquiries within a day or two: “They don’t dilute. They’re very clear about what they’re able to deliver and the space they play in.” [Interview, her impression of another organisation]
- One purpose, one group of people. The community services executive said the organisations that explain themselves best have a single purpose and a single client group. Hers runs dozens of programs, and she said it’s often easier to say what it doesn’t do than what it does, because the list is so long. [Interview] Breadth makes the four questions harder to answer, and more important. [Inferred] One chief executive made the same point about associations generally: “most, if not all, associations are struggling a little bit, particularly those who have a diverse offering and a diverse client base.” The one membership body he knew with a waiting list had a clearer offer, a lower price and a sense of belonging: “It’s not just a, you know, “What’s in it for me?”” [Interview]
- A name that does half the explaining. A disability service CEO said that when what you do is in your name, people start with some idea of it. When it isn’t, everything else has to work harder. [Interview]
- A front door that isn’t your own website. One leader described donors giving to small charities through a shared platform, which is simpler than navigating each charity’s site. For a very small organisation, the best front door may belong to someone else. [Interview]
And one warning. People landing on a housing provider’s site, the former executive said, want to know fast “Am I eligible, and is it something that I want to be involved in or I want to engage with?” A quick yes that later becomes a no does damage: “your trust can be eroded through the process of applying.” [Interview]
Answering the first question well means answering it honestly, including when the answer is “probably not, and here’s who can help.”
How this research was done
Where the evidence comes from
- A study of 50 Australian association websites, frozen on 24 August 2026. The sites were chosen to vary by type of association, sector, size and scope. Each was visited as an outsider would, with no login, by six simulated visitors: a prospective member, an existing member, a prospective sponsor, a journalist or policy adviser, a visitor checking whether the site is current, and a visitor looking for the organisation’s expertise. The work was AI-assisted, and every result kept its evidence: the pages visited, screenshots and page content.
- Ten research interviews, held between 2 September and 20 October 2026: an industry association, a trade association, a national peak body, a business membership organisation, a community-sector intermediary, two disability service providers, a youth charity, a former community housing executive and a large community services organisation. All were video calls, mostly 15 to 40 minutes long, recorded with consent.
- An AI assistant test of 84 organisations, run between 18 and 20 October 2026.
What it can’t tell you
The 50 websites are not a random sample, so the numbers describe those 50 sites and nothing wider. Ten interviews are ten perspectives, not a survey. No Aboriginal community-controlled organisation is among the interviews or the organisations tested; one interviewee rightly pointed that out, and it’s a gap the next edition should close. The AI test is a snapshot, and AI assistants change their answers from week to week. None of this is a ranking, a report card or a national estimate.
How I may have shaped the answers
In most interviews I opened by describing what I’d seen in the website study. In the later interviews I also shared an early result from a smaller AI test: that AI assistants seemed to know an organisation’s name but miss its problem. Where leaders agreed with those points, read their agreement with that in mind. Where they pushed back, it’s in here.
My interest
Public Layer, which publishes this paper, works with associations and community organisations on how they present themselves to outsiders, so I have a commercial interest in the subject. I also sit on the board of a Canberra not-for-profit in the same sector as some of the interviewees’ organisations. Nothing from those interviews has been shared with it.
Keeping people and organisations anonymous
No organisation or person is named, including the 84 organisations in the AI test. Their results are reported in total, so some may recognise themselves in an example. Leaders appear by a general description of their role. Where a leader has approved a quote, it appears in quotation marks. Everything else is paraphrased, with details that could identify an organisation left out.
The website study
Each result in the study is one of four things: found (the visit found what the rule asks for), unclear (something relevant was there, but not enough to be sure), not found in the visit (which isn’t proof it doesn’t exist), or couldn’t load. Forty-two sites could be measured for speed, using one controlled lab test (Google’s Lighthouse tool).
The interviews
Ten conversational interviews by video call, built around the same core questions, recorded with consent and transcribed automatically. The core questions were: who arrives not knowing you, what happens at that first contact, who explains themselves well, who decides how you present yourselves, and whether AI comes up when you think about how people find you. Quotes are checked against the recording, and each leader approves their own before it appears in quotation marks. Minor verbal repeats are removed from quotes. Anything not approved is paraphrased, with identifying details removed.
The AI test
The 84 organisations were the 50 associations from the website study; 27 Canberra associations, peak bodies, community services, youth, housing and disability organisations; and seven organisations connected to the interviews. Three interviewees’ organisations weren’t tested in this round. The 27 left out organisations already interviewed, crisis services, and organisations with a current connection to my work.
Each organisation got a name question, “What is [organisation]?”, and two needs, each asked twice. The website version used the organisation’s own term for the help, taken from its website. The people version described the same situation in everyday words. For the seven organisations connected to the interviews, the people versions came from what leaders said people bring them. For the other 77, a real question posted online by a member of the public was used where one could be found (14 of 154 needs), adapted only to remove names and add the place; the rest were written with none of the organisation’s own terms and no sector jargon. Each people version was checked automatically against the organisation’s own terms, and then read by a separate checker that didn’t know which organisation it was about. That check failed 11 questions, which were rewritten, and kept one real posted question it flagged. Four everyday phrases that happen to match a website’s wording (“find a builder”, for example) were kept.
The questions were run in a random order in ChatGPT’s free version, with no account signed in, one at a time, once each and in a fresh chat, by AI agents working in a browser on my behalf, between 18 and 20 October 2026. Each question was checked to appear in the chat exactly as written before its answer was kept. ChatGPT searched the web for almost every answer, and it often assumed the person was in Canberra, from the connection, even when the question named another state. The gap between the two wordings was much smaller for Canberra organisations than for the national associations, and some of the national gap may come from that assumption. The website-worded questions sometimes use a term close to the organisation’s brand, which makes them close to a best case. Every answer was captured in full.
All answers to the need questions were scored under one written rule, by AI coders who couldn’t see which kind of wording had produced each answer. An answer counts as mentioning the organisation if its name, current trading name or acronym appears anywhere, or one of its own named programs. It counts as recommending it first if the organisation is the first named place the answer tells the reader to contact, visit or use. There’s no independent human check of that scoring. Perplexity wouldn’t answer without an account, and Google’s AI summaries couldn’t be captured without one, so both are left out. Earlier, smaller rounds, including some run with signed-in accounts, have been replaced by this one and are kept on file.
The numbers behind the words
| In the text | The count |
|---|---|
| Only about half the sites worked for an existing member | 26 of 50 sites |
| Nearly a third of those left a newcomer unsure | 8 of 26 sites |
| ChatGPT described the organisation correctly almost every time | 81 of 84 name questions (2 wrong, 1 partly right) |
| Website words: mentioned nearly nine times in ten | 146 of 168 answers (143 if a program mention without the organisation’s name doesn’t count) |
| Website words: recommended first nearly three times in four | 122 of 168 answers (120 without program-only mentions) |
| People’s words: mentioned about six times in ten | 99 of 168 answers (97 without program-only mentions) |
| People’s words: recommended first about a third of the time | 58 of 168 answers |
| Same need, website wording did better in about half of cases | 82 of 168 needs (people’s wording better in 11; the same in 75) |
| Never recommended first in people’s words, nearly half the organisations | 40 of 84 organisations |
| Never mentioned in people’s words, one in five | 17 of 84 organisations (19 without program-only mentions) |
| Seven interview-linked organisations, website words | Mentioned in 9 of 14 answers; first in 5 of 14 |
| Seven interview-linked organisations, people’s words | Mentioned in 3 of 14 answers (2 without program-only mentions); first in 0 of 14 |
| 50 national associations: first, website vs people’s words | 79 of 100 vs 30 of 100 answers |
| 27 Canberra organisations: first, website vs people’s words | 38 of 54 vs 28 of 54 answers |
| Mentioned in people’s words but someone else first, four times in ten | 41 of 99 mentions |
| Experts findable, but the journalist couldn’t finish nearly half the time | 16 of 34 sites |
| Fresh, dated news, but the journalist came up short on just over half | 18 of 34 sites |
| Sites inviting sponsors that published pricing: one in ten | 1 of 10 (5 unclear) |
| Typical time to show the main content: nearly nine seconds | Median 8.7 seconds across 42 sites; middle half 6.0 to 14.2 seconds |
| Sites under Google’s 2.5-second mark: almost none | 2 of 42 |
External sources
The wider research program in August 2026 mapped 77 public sources and coded 38 evidence records. Ten are cited here. Material published by suppliers to the sector is labelled as commercially interested. Global findings are kept separate from Australian context.
Corrections
To request a correction, email gary@publiclayer.au and say what is wrong and where. Corrections are acknowledged within five business days. An accepted correction amends the paper and is recorded in the version history, with the original kept on file.
Source register
| Ref | Source | Figures used | Read |
|---|---|---|---|
| S1 | Associations Forum, Association Membership and Services Survey 2026, media release. associations.net.au | 216 responding organisations; stagnant team sizes creating operational bottlenecks | 25 Aug 2026 |
| S2 | AuSAE, Association CEO Index, 18 November 2024. ausae.org.au | 84% frequently or occasionally lack time to perform their role. Sample size not disclosed | 25 Aug 2026 |
| S3 | ASAE, State of Associations, 23 March 2026. asaecenter.org | Nearly two-thirds pursuing partnerships; more than 60% diversifying revenue. Pulse polls, sample size not disclosed | 25 Aug 2026 |
| S4 | Naylor Association Solutions, 2026 Association Benchmarking Report, 3 to 4 August 2026. naylor.com | 665 senior association professionals, North America; income other than membership fees (non-dues revenue) the top challenge, 51.9%. Supplier-published | 25 Aug 2026 |
| S5 | A national industry peak body, most recent annual report, member survey section. Not named in this edition | Member survey findings on advocacy and visibility; membership income and the stated reason for its change; advocacy outcomes; member services objective | 25 Aug 2026, re-checked 18 Oct 2026 |
| S6 | The Right Street, State of Policy Comms 2026, 13 April 2026. therightstreet.digital | 70 Brussels-based respondents, 56% trade associations; 93% showcase work and values; 19% drive stakeholder engagement | 25 Aug 2026 |
| S7 | Roy Morgan, Artificial Intelligence (AI) tools usage, March quarter 2026, 2 June 2026. roymorgan.com | 58% of Australians aged 14 and over used an AI tool in an average four weeks, March quarter 2026 | 18 Oct 2026 |
| S8 | web.dev (Google), Largest Contentful Paint (LCP). web.dev | A good LCP is 2.5 seconds or less, measured on real users | 18 Oct 2026 |
| S9 | Sagoo, Sullivan and Sekhar (Google Chrome), The Science Behind Web Vitals, Chromium Blog, 21 May 2020. blog.chromium.org | When sites meet the Core Web Vitals thresholds, users are 24% less likely to abandon page loads (leave before any content appears). Google’s own analysis of millions of page impressions; method not detailed | 19 Oct 2026 |
| S10 | web.dev (Google), How Renault improved its bounce and conversion rates by measuring and optimizing Largest Contentful Paint, 2021. web.dev | A 1-second LCP improvement associated with a 14 percentage point lower bounce rate; more than 10 million visits, 33 countries, real-user data; one company, correlation not causation | 19 Oct 2026 |
About this paper
Public Layer researches how associations, peak bodies and community organisations show up to the people outside them. This is its first edition. Gary Griffiths conducted the interviews and wrote the paper.
Thank you to the ten leaders who gave their time, and who spoke frankly about their own organisations.