Your Next Applicant Already Asked ChatGPT About You

The enquiry form is not where the relationship starts

By the time a prospective student fills in your enquiry form, books an open day, or calls admissions, they have usually already asked ChatGPT or Perplexity about your school. UCAS's own research, based on a survey of 4,485 potential applicants, applicants and first-year undergraduates fielded in November 2025, found that 48% had used AI to explore their higher education options — most commonly to compare universities (61%), explore subject choices (59%), and check entry requirements (52%) (UCAS, published via Wonkhe, March 2026).

That is not a niche behaviour confined to tech-forward applicants. The Higher Education Policy Institute's 2026 Student Generative AI Survey, fielded by Savanta among 1,054 full-time undergraduates in December 2025, puts overall AI use among current students at 95%, with 94% using generative AI to help with assessed work (HEPI, March 2026). Today's applicants are not experimenting with AI for the first time when they research your school. They are extending a habit they already have for coursework, revision and everyday research into the one decision that determines where they spend the next three or four years — a pattern we tracked in more depth in our analysis of Gen Z's AI-first university search.

The practical consequence for admissions and marketing teams: the first "conversation" a prospect has about your institution is not with a member of staff. It is with a language model that has read whatever public information exists about you — your website, your UCAS listing, your press coverage, your Reddit mentions, your QAA and Office for Students records — and compressed it into a confident-sounding paragraph. Whether that paragraph is accurate is largely outside your control at the moment it's generated. What is inside your control is the information supply that produced it.

What ChatGPT already told them before they contacted you

AI answers about a school fall into three categories, and prospects rarely know which one they got.

Accurate. The model correctly states your entry requirements, your UCAS tariff points, your Clearing availability, your accreditation status. This happens when your public information is current, specific and consistently repeated across your own site and third-party sources such as UCAS, the QAA and the Office for Students register.

Outdated. The model cites last year's tariff points, a discontinued course, or fees from a previous academic year, because that is the version of your information that was most consistently published and cross-referenced when the training data was assembled. Nothing on your current site actively contradicts it clearly enough to override the pattern.

Fabricated. The model states something plausible but false — an accreditation you do not hold, a ranking position that does not exist, an intake date that was never scheduled. Large language models fill information gaps with statistically likely completions, not verified facts, and higher education has enough structural similarity between institutions that a fabricated detail can sound entirely credible.

The scale of the middle and third categories is not abstract. In the same UCAS research cited above, 73% of applicants who had used AI in their research said they had encountered or received incorrect information from it — the single biggest trust barrier the survey recorded. Applicants are not naively trusting whatever the model says. Most treat it as a first pass, then cross-check: only 13% named a chatbot as their first research resource, against 43% who still started with a university's own website. AI shapes the shortlist and the questions; your website remains where the shortlist gets tested.

Why this changes what "first contact" means

Traditionally, admissions teams have treated the enquiry form, the open day booking, or the Clearing hotline call as the first contact point — the moment they can start shaping a prospect's understanding of the institution. That framing assumes the prospect arrives with a blank or self-formed impression. It no longer holds.

By the time a prospect reaches out, they already carry an AI-generated mental model of your institution: rough entry requirements, an impression of your reputation relative to competitors, possibly a fee figure, possibly a course that was quietly retired two years ago. If that model is wrong, your admissions team's first job is not answering their question — it is correcting a wrong answer the prospect already believes, without making them feel foolish for having believed it. That is a materially harder conversation than answering a genuine unknown, and it costs staff time that a factually clean, machine-readable information base would have prevented.

This is the practical argument for treating your public information architecture — not your enquiry funnel — as the actual first contact point. Generative Engine Optimisation (GEO) is the discipline for this: structuring institutional information so AI systems can extract it accurately rather than approximate it. Our full GEO guide for UK schools covers the framework end to end; three actions matter most for the pre-contact problem specifically.

1. Close the gap between what's true and what's published

Every fact an AI model might cite — tariff points, fees, TEF rating, accreditation status, Clearing availability — should exist in exactly one current, unambiguous, dated form on your site, and should match your UCAS listing, your QAA entry and your OfS registration exactly. Contradictions between these sources are what push a model toward the "outdated" or "fabricated" outcome, because it has no reliable version to converge on.

2. Structure it so machines can extract it, not just read it

Schema.org's EducationalOrganization and EducationalOccupationalProgram types, implemented per Google Search Central's structured data guidance, give AI systems a machine-readable version of the same facts your prose already states. This does not replace good content — it removes the ambiguity that forces a model to guess. Perplexity's publisher documentation and OpenAI's documentation for how ChatGPT search retrieves and cites web content both confirm that structured, current, single-source-of-truth content is what these systems prefer to surface and cite.

3. Give models something current to find during high-volatility periods

Clearing is the sharpest version of this problem: availability changes hour by hour, and a prospect searching at 11pm during results week will ask ChatGPT before they call a hotline that's closed. A dedicated Clearing page with FAQPage markup, updated in real time, is one of the few places where keeping AI systems current has an immediate, measurable payoff. Our analysis of AI citation and accreditation signals covers how QAA, OfS and TEF status specifically feed into whether AI engines name your institution with confidence at all.

What this looks like across the applicant journey

The data protection dimension

There is a second, less obvious reason to take pre-contact AI answers seriously: UK GDPR still applies once a prospect moves from an AI conversation to an actual enquiry with you. If your own chatbot or enquiry tooling collects personal data, the Information Commissioner's Office expects clear, accessible transparency about what is collected and why — independent of whatever the prospect assumed based on an earlier AI conversation elsewhere. A prospect who arrives believing something incorrect about your fees or requirements, based on a third-party AI tool you do not control, is not a data protection issue in itself. But the moment they interact with your own systems, ICO transparency expectations apply to you, not to the AI tool that misinformed them — another reason the correction needs to happen cleanly and early, on your own properties.