ChatGPT
Claude
Perplexity
Gemini
GrokThe real first contact isn't the open day
By the time a Year 12 student emails your admissions team, or a parent calls to ask about ATAR cut-offs, they have usually already asked ChatGPT or Perplexity the same question. What the AI told them — correct, outdated, or invented — is already shaping how they read your reply.
This changes the sequence institutions have planned around for decades. The open day, the course guide download, the first phone call: these were designed as the start of the relationship. In 2026, they are increasingly the second or third touchpoint. The first is a chat window the institution never sees, cannot correct in real time, and often does not know happened at all.
For Australian institutions specifically, this matters more than the general trend suggests. Admissions here run through five separate state-based centres — UAC in NSW/ACT, VTAC in Victoria, QTAC in Queensland, SATAC in SA/NT, TISC in WA — instead of one national portal. ATAR cut-offs, CSP availability, and pathway options vary by state and by year. That fragmentation is exactly the kind of detail a language model is likely to blur, average, or simply get wrong when a prospect asks "what ATAR do I need for [course] at [your university]?"
What AI engines actually know about your institution
ChatGPT and Perplexity do not "look up" your institution the way a prospect browsing your website does. They draw on two different sources, and the mix determines what a prospect hears.
The first is the model's training data: a snapshot of the public web, with an inevitable lag behind your current course guide, CSP allocations, and intake dates. The second, for engines with live retrieval — Perplexity and Gemini with Search are the clearest examples — is a real-time search pass layered on top. Even then, the model still has to decide which pages to trust and how to summarise them, and it favours pages with clear structured data, named entities, and consistent figures across sources.
This produces three recurring failure modes for Australian institutions:
- Stale fee and ATAR figures. A model trained months earlier may cite last year's Commonwealth Supported Place fees or ATAR cut-off, presented with the same confidence as current data.
- Go8 default bias. Ask a generic question like "best business school in Australia" and the model reaches for Group of Eight names first, because they dominate the English-language training corpus. A well-regarded regional or specialist provider can be omitted entirely, not because it is worse, but because it is less represented in what the model has read.
- CRICOS and accreditation gaps. International applicants specifically ask AI tools to confirm a provider's legitimacy — CRICOS registration, TEQSA status, professional accreditation (CPA Australia, Engineers Australia, and equivalents). If that information is not stated clearly and consistently on your own site, the model has nothing reliable to repeat, and may hedge, omit it, or guess.
The evidence: AI is already the first stop, not a curiosity
There is no published Australia-specific survey yet that isolates AI use during university search the way US researchers have measured it — a gap worth naming rather than papering over with an invented figure. The clearest quantified picture comes from the United States, where the scale of measurement is furthest along.
46% of US high school students used AI tools such as ChatGPT during their college search between October and November 2025, up from 26% just months earlier in spring 2025 (Source: EAB survey of 5,000+ high school students, released February 2026). The same survey found that 18% of respondents removed a college from consideration based on information an AI tool surfaced about it — meaning the AI's answer, right or wrong, directly cost some institutions a prospect they never knew they lost.
That figure is US-scoped and should not be read as an Australian number. But the direction is not in doubt. UAC, which runs the country's largest annual survey of Year 12 school leavers, added artificial intelligence to the topics it tracks alongside course and university selection starting with its 2024 Student Lifestyle and Learning Report — a sign Australia's own admissions bodies already treat this as worth measuring every year, even without a published headline percentage yet. The qualitative direction — a first-contact channel institutions do not control — already applies here.
Book a demoWhat this means for a Group of Eight vs a regional or specialist provider
The visibility gap is not evenly distributed. A Go8 university's name, accreditations, and general reputation are well represented in most language models by default — prospects rarely need to ask "does this university exist" about Melbourne or UNSW. A newer, regional, or subject-specialist provider carries none of that inherited weight. Every fact an AI engine states about it has to come from somewhere identifiable and current.
| Applicant question | What a Go8 name gets by default | What a regional/specialist provider needs to supply explicitly |
|---|---|---|
| "Is this a real, TEQSA-registered university?" | Assumed, rarely questioned | TEQSA registration and CRICOS provider code stated on-page |
| "What ATAR do I need?" | Often cited, sometimes stale | Current ATAR/entry pathway per course, updated each intake |
| "Is the degree accredited?" | Assumed for well-known faculties | Named professional body (e.g. Engineers Australia, CPA Australia) per programme |
| "What do graduates earn / go on to do?" | Sometimes cited from rankings | QILT Graduate Outcomes data, sourced and dated |
| "Is it one of the 'best' universities?" | Defaults to yes, via Go8 membership | Needs Good Universities Guide or QS/THE placement, named and dated |
The practical implication: a smaller or newer provider is not competing with the Go8 on marketing budget in this channel — it is competing on how legible and current its own public information is to a model that has no institutional memory of the brand.
What Australian institutions should structure differently
Three changes reduce the odds of a prospect getting the wrong answer before they ever contact you.
Make the state-specific admissions detail explicit, on-page, and dated. Because Australia runs ATAR cut-offs and CSP places through five separate state centres rather than one national body, a page that says "check UAC for entry requirements" without stating the actual current ATAR band gives a model nothing concrete to repeat. State the number, the intake year, and the pathway alternatives (foundation year, diploma, adjustment factors) directly.
Name your accreditations and registration status as entities, not adjectives. "Fully accredited" is not a fact a model can verify or repeat with confidence. "TEQSA registered, CRICOS provider code [X], accredited by [named professional body]" is. Google's structured data guidance and schema.org's EducationalOrganization type both exist for exactly this reason: a named entity in machine-readable markup beats a descriptive adjective in prose. The difference determines whether an international applicant researching legitimacy from overseas gets a confident, correct answer or a vague, hedged one — see our detailed breakdown of what makes an accreditation citable by ChatGPT and Perplexity.
Treat your own chatbot as the correction layer, not just a lead-capture tool. A widget that answers "what ATAR do I need for nursing?" or "is my degree CRICOS-registered?" accurately and instantly gives prospects who arrive with a wrong or outdated AI answer a chance to be corrected on the spot — before they act on bad information or, worse, rule your institution out. This is the same underlying discipline covered in our guide to GEO for schools: how to appear in AI answers, applied to the moment a prospect finally does reach out.
Why this matters more for international applicants
Australia's higher education sector depends on international enrolments in a way few other national systems do, and international applicants are, structurally, the segment most likely to research entirely via search and AI tools before any human contact — often from a time zone where a phone call to your admissions office is not a realistic first step.
An applicant in Delhi, Jakarta, or Ho Chi Minh City asking ChatGPT "is [your institution] a legitimate place to study in Australia?" needs a confident, sourced answer about CRICOS registration and ESOS Act compliance, not a vague one. Study Australia, the government's own international promotion channel, and TEQSA's public register are both sources AI engines can and do draw on — but only if your own site states the same facts consistently, so the model has no reason to hedge.
What to check this month
Start narrow. Ask ChatGPT and Perplexity, as an incoming Year 12 student would: "What ATAR do I need for [your flagship course]?", "Is [your institution] TEQSA registered?", and "What are graduate outcomes like at [your institution]?" Read the answers as a prospect would — not as someone who already knows the right answer. Where the response is wrong, outdated, or vague, that is the exact page on your site that needs the fact stated more explicitly, with a date attached.
This single-page audit approach is expanded in our guide on how Gen Z now starts a school search on AI tools, which walks through the broader shift in where prospects begin their research and what it means for the pages that used to be the entry point.
