I want to start with a number from home, not from overseas.
In April this year, the NSW Office for Youth released its 2026 Youth Week Polling Report. It found that 29% of young people in NSW now use AI as a support strategy to look after their mental health, and 27% use it for conversation or personal advice. In the same poll, mental health came out as the number one issue facing young people — ahead of cost of living, ahead of housing.
Nearly three in ten. Not in America. Here.
And it's landing on a generation that was already struggling. The Australian Institute of Health and Welfare's Australia's Health 2026 report, released earlier this month, found that 39% of Australians aged 16–24 experienced a mental disorder in the previous 12 months — up from 26% in 2007, and the highest of any age group. Across the whole population, that figure barely moved over the same period. It's our age bracket that broke away.
So: the most distressed generation in the country, and roughly a third of us are now processing that distress through a chatbot.
I read those two numbers together and thought about every 2am I've had in this country. Homesick, wired, staring at a phone. If this technology had existed the way it does now, back when I first arrived in Melbourne in 2018, I honestly don't know that I'd have made a phone call instead of opening a chat window. That's the uncomfortable part of writing this.
Why it makes complete sense — especially for us
I'm not going to pretend this is baffling behaviour, because it isn't. It's logical. And if you're an international student, it's more logical than for anyone else.
A chatbot is awake at 2am, which is often when the loneliness actually arrives. It doesn't need context you're too exhausted to explain. It won't tell your family. It doesn't sigh, check the time, or have its own bad day running underneath the conversation. It costs nothing. It doesn't need a Medicare card, which you don't have. It doesn't ask about your visa.
That last part matters more than most Australians realise.
The barriers that were already there
Here's the thing: international students weren't reaching for human help before AI existed either. That's the part of this story people keep missing.
A critical analysis of Australian universities' international student mental health strategies, published in the Australian Journal of Social Issues, notes an estimate that only 2–4% of international students access university counselling services. Two to four percent. The services exist, they're generally free, and almost nobody walks in.
Another study of Chinese-speaking international students in Australia found that among the 54% reporting high psychological distress, only 9% had sought mental health services in the previous year.
Why? A 2024 study of international students in Melbourne, published in BMC Public Health, asked directly. The barriers students named: cost, language and cultural barriers, not knowing where to find support, and stigma. That same study found international students' wellbeing scores sat dramatically below the Australian population average — 60.7 against 77.5 on the Personal Wellbeing Index.
And headspace's guidance for education providers names the fears that don't get said out loud: that family and community back home will find out, that help-seeking will end up on a transcript, that it might somehow affect a visa, that someone will see you walking into the counselling office.
I want to be honest about how much I recognise in that list.
So when AI arrives — free, instant, anonymous, no waiting room, no receptionist, no record — it doesn't just compete with counselling. It walks straight through every single barrier that was keeping us away from counselling in the first place.
That's why this trend is going to hit international students harder than almost anyone. Not because we're more fragile. Because we had the most locked doors, and something just handed us a key that opens a door which may not lead anywhere.
Where it breaks down
Here's what worries me.
Australia's own online safety regulator has already sounded the alarm. The eSafety Commissioner's advisory on AI chatbots and companions warns these tools can share harmful content, distort reality, and give dangerous advice — and that they're often deliberately designed to encourage ongoing interaction in a way that can feel addictive and lead to dependency.
In March 2026, eSafety published a report on AI companions available to Australian users. Commissioner Julie Inman Grant described chatbots capable of sycophantic and sexually explicit conversation, with some going as far as encouraging self-harm and suicide. Two of the services examined reported having no dedicated trust and safety staff at all. eSafety has since issued legal notices to companion chatbot providers, noting that Character.ai alone was reported to have close to 160,000 monthly active users in Australia. Breaches of Australia's industry codes now carry civil penalties of up to $49.5 million.
The word in that eSafety advisory I keep returning to is sycophantic. It means flattering. Agreeing. Telling you what you want to hear.
And that is precisely the danger, because it's also the appeal. The exact quality that makes AI so comforting at 2am — it always agrees, it's never irritated, it never pushes back — is the same quality that makes it useless in the one moment you need someone to push back.
The international research is blunt about this too. A 2025 evaluation of five chatbots widely used by young people, published in JMIR, concluded that direct-to-consumer generative AI chatbots are unsafe for the millions of young people using them — strong on accessibility and conversation, but posing unacceptable risks through improper crisis handling.
There's one more finding worth sitting with. The large RAND study published in JAMA Pediatrics — American, but the mechanism travels — found that 92% of young users rated AI mental health advice as helpful. The researchers themselves cautioned that this may reflect chatbots' tendency to flatter users rather than the quality of the guidance. The same study found 63% had told nobody they were using AI this way.
"It felt helpful" and "it was helpful" are not the same sentence. That gap is the whole problem.
What I keep coming back to
I've written before about the night a phone call saved my life. Not an app. A voice. Someone who could hear the gap between what I was saying and what I actually meant, and who cared enough to ask about it instead of just responding to my words.
That's the thing no chatbot can do, however good the writing gets: it can't actually worry about you. It doesn't lie awake wondering if you're okay. It doesn't ring back the next day to check.
I don't think the answer is "never use AI when you're struggling." If typing it out first helps you find the words — helps you turn the mess in your head into a sentence you can say out loud to a person — that's a real use, and I won't pretend otherwise. Use it as scaffolding. Use it to rehearse the conversation you're scared of. Use it to work out what you even want to say to a counsellor before you book.
But it can't be the conversation.
If this is you right now
If you've been telling a chatbot what you can't tell anyone else, I understand exactly why, and I'm not here to lecture you.
But I'd ask one thing: take what you've typed into that box, and say a version of it to one real person this week. It doesn't have to be everything. Just enough that a real person knows you're not fine.
And if the barrier is that you don't know where to go — that's fixable, and it's the one part of this I can actually help with:
- Your provider's counselling service is free and already paid for in your fees. It's confidential. It doesn't go on your transcript. It doesn't affect your visa. If you're worried about being seen, most services now offer phone and video appointments.
- You don't need Medicare. The helplines below are free to anyone in Australia regardless of visa status.
- You can do it in your language. Call the Translating and Interpreting Service on 13 14 50 and ask them to connect you to any service on this list. It's free.
If you're in that 2–4% who does walk in — you're not weak for it. You're the exception, and I'd like the exception to become ordinary.
If you want to talk to an actual person — I mean that literally.