Nearly half of US high-school students use AI to pick colleges. What should admissions pages answer?
EAB found 46% of US high-school students use AI in college search, and 18% dropped a college based on it. What program pages must answer, and how to check.
The short answer
Admissions and program pages should answer the questions students now put to AI: what it really costs after aid, what graduates earn and do, how the program ranks and who it suits. EAB found 46% of US high-school students used AI in college search in late 2025. Check what the engines say about you before fixing pages.
Key takeaways
- In EAB's survey of more than 5,000 US high-school students (October to November 2025), 46% used AI tools in their college search, up from 26% in spring 2025.
- In the same survey, 18% removed a college from consideration based on AI results, so an AI answer can end a relationship before the first visit.
- Cost after aid, graduate outcomes, rankings and fit are the questions a program page has to answer in plain sentences an engine can quote.
- Engines fill gaps with third-party data, so a page that hides net price or outcomes leaves the answer to someone else.
- Check what six engines say about your institution, on prompts students actually use, before rewriting anything.
In this article
How many students use AI to choose a college?
Nearly half of US high-school students now do. EAB, which sells enrolment marketing services to colleges, surveyed more than 5,000 US high-school students in October and November 2025. It found 46% used AI tools such as ChatGPT in their college search, up from 26% in spring 2025.1
The finding that should worry an admissions office is the next one. In the same survey, 18% of students said they had removed a college from consideration based on AI-generated search results.1 That is a decision taken before the student opened a viewbook, attended a fair or spoke to anyone on your team.
EAB also reported that 43% of students say AI will influence their career choice, and 39% are considering alternatives to college because of AI (October to November 2025).1 The question "is this degree worth it" is now part of the search, and AI is one of the places students ask it.
US college search, late 2025
AI is already in the shortlist
What do students ask AI about colleges?
EAB's release does not list the questions students ask, so be careful with anyone who claims to know the exact mix. Other research gives a direction. ICEF Monitor's January 2026 write-up of an INTO survey of newly enrolled international students covers how they used AI in their initial search and which questions they put to it, with rankings and reputation among them.2 Times Higher Education's October 2025 report on IDP's Emerging Futures research covers prospective international students using ChatGPT and other AI tools to choose a university.3
Both cover international students, not US high-school seniors, so they are context rather than proof. Our view: four question types decide a shortlist for most applicants, and they are the ones a program page has to answer.
- Cost: net price after typical aid, not sticker price, and what changes it.
- Outcomes: what graduates of this program earn and do, and how many finish.
- Rankings and reputation: where the program stands, and on whose list.
- Fit: who the program suits, entry requirements, class size, online or on campus.
Example prompts a US student might use: "Which colleges have a good nursing program and graduates who pass the licensing exam?", "Is a computer science degree at a state school worth it?", "Which small liberal arts colleges give generous aid to middle-income families?" None of them names your institution, and that is the point. The engine decides who appears.
Where the engines get their answers
Engines draw on whatever source answers the question most plainly. For US institutions that often means federal data, ranking publishers, review sites, forums and news coverage, alongside your own pages. The US Department of Education's College Scorecard publishes cost and outcome data by institution and field of study, and engines can read it.4
That creates a specific risk. If your program page says "affordable" and "strong outcomes" but gives no figures, the engine will quote someone else's figures, possibly out of date or for a different program. A student who asks about your nursing degree may get your institution-wide numbers, or a forum post from four years ago.
Our view: an admissions page that will not state net price and outcomes in a sentence is choosing to let a third party state them instead. The data is already public. The only choice is whose version the engine quotes.
What should a program page answer?
A program page should answer each of the four questions in the first lines of its own section, in sentences that stand alone when quoted. Answer first, then explain.
The same program page, two ways
Hard to quote
- "An affordable, world-class education"
- "Our graduates go on to great things"
- "Nationally recognised"
- Cost in a downloadable PDF
Easy to quote
- Net price after typical aid, by family income band, with the year
- Completion rate and graduate outcomes, with the source and cohort
- The named ranking, its year and the program's place
- "This program suits students who…" in one sentence
Four rules make the right-hand column work. Put the figure and its year in the same sentence. Use the program's own data where you have it, and say when you are using institution-wide data. Keep one page per program, so the engine is not left to guess which degree a number belongs to. Repeat the key facts in an FAQ on the page, in the words students use.
Consistency matters as much as content. If your catalogue, your program page and your profile on a ranking site give three different tuition figures, an engine has three answers to choose from. Our guide to fixing what ChatGPT gets wrong about you covers how to trace a wrong figure to its source and correct it where the engine reads.
How to check what AI says about your institution
Run the prompts students use, on several engines, on more than one day, and record who is named and what is cited. One run on one engine tells you very little, because answers vary from run to run.
- Write 30 to 50 prompts across the four question types, in the words a seventeen-year-old would use. Include prompts for your strongest programs and for the ones you most need to fill.
- Run each prompt on ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI, on three separate days.
- Record whether your institution is named, what the answer says about cost and outcomes, and which sources are cited.
- Mark every factual error and trace it to the page the engine quoted.
- Fix the pages in order of how many prompts they affect, then rerun the set a month later.
Report the results with a margin of error, not as a single percentage. Our buyer's spec for an AI visibility score explains how many runs you need before a change means anything.
This is the method our AI answer monitor uses: six engines, official APIs where they are offered and a licensed AI-answer data provider otherwise, and each prompt run on three separate days in an audit. Our view: start with the program you most need to fill, not the institution's brand. That is where an engine's answer costs you applicants.
