AI in Higher-Ed Recruitment: Where It Helps, Where It Doesn't

Somewhere above you, a president or provost has read that AI is transforming enrollment and has asked what you are doing about it. The honest answer is more interesting than either yes or no, but it depends on a distinction most of the coverage skips.

I worked in university recruitment in 2015. There was no AI in the funnel. What there was, in enormous quantity, was manual work a machine should obviously have been doing: chasing missing transcripts, re-explaining the same six deadlines, hand-checking whether a program page still listed the right credit hours.

A decade later, I work on the other side of the problem, on how search and answer engines read institutional content. The pattern connecting both jobs is that AI has always been good at the parts of recruitment nobody wanted to do by hand, and bad at the parts people were hired for.

What follows is a working map of that distinction: where the evidence says AI earns its place in recruitment, where the jury is out, and where it creates risk the institution ends up carrying.

How Widely Are Universities Actually Using AI in Recruitment?

Most marketing and enrollment teams are already using AI, and adoption is climbing quickly. In EducationDynamics' 2025 Marketing and Enrollment Management AI Readiness Report, 65% of respondents reported actively using AI in their marketing and enrollment efforts, up from 40% in 2024. The wins are real but modest: 69% cite improved workflow efficiency, and 48% believe AI has positively affected their enrollment funnel. Efficiency is the consistent gain. Enrollment impact is believed more often than it is measured.

The caution is equally documented, and in Ellucian’s third annual higher education AI survey, published in March 2026 from 779 respondents across more than 300 institutions, data security and privacy remained the top barrier at both the individual level (61%) and the institutional level (56%).

Given the exposure that general purpose AI tools create for records protected under FERPA (the Family Educational Rights and Privacy Act), that caution reads less like hesitation and more like foresight.

Where Do Prospective Students Now Start Their College Search?

Increasingly, but not yet mostly. Search engines still lead, and AI assistants have become a routine second stop rather than a replacement. Carnegie’s 2025 research, based on May 2025 responses from more than 3,400 prospective students and parents, found AI use in the college search among graduating seniors rising from 4% in 2023 to 10% in 2024 to 23% in 2025. Rising students, asked about intent rather than behavior, came in slightly lower at 20%.

These tools are answer engines: systems that return one synthesized answer with citations rather than a list of links. ChatGPT, Perplexity, Gemini, and Google’s AI Overviews all work this way, and what happens inside their answers matters as much as who uses them.

In UPCEA and Search Influence's 2025 AI Search in Higher Education study, fielded in March 2025 among 760 adult prospects for professional and continuing education, 79% said they read Google’s AI Overview when one appears. Of those who read it, 56% said a cited source becomes more trustworthy, though 39% said it makes no difference to their trust at all.

Note the cohort. Carnegie surveyed graduating high school seniors; the UPCEA panel was working adults. The behavior shows up in both, but they are different audiences with different journeys.

One finding from the same study cuts against the panic. Asked which platforms they use the way they would use a search engine, 84% named search engines, 61% YouTube, and 50% AI tools. AI is now a mainstream research channel, but it has not displaced the others.

And AI Overviews are not the traffic apocalypse they are described as: among prospects who read one, 11% click through to the sources every time, 40% most of the time, and 43% occasionally.

Image

Still, the direction is clear enough to plan around. When a student asks an assistant to compare nursing programs, your site is competing to be the source the model cites rather than for a blue link. Before AI does anything clever on your website, it has already decided, on someone else’s platform, whether your programs enter the conversation at all.

Where Does AI Clearly Help University Recruitment Today?

AI helps when it organizes, checks, or translates content your team already owns and governs. These are bounded answer tasks, meaning each has one correct answer a human has already defined, and they map directly onto the sprawl that makes university websites hard to run.

1. How Does AI Help a Large Program Catalog Get Found?

By tagging it consistently. A big university publishes hundreds of programs across dozens of subsites, and prospective students rarely land where you would like. AI assisted tagging and structured content keep a 200 program catalog searchable by both people and answer engines, which is now part of getting recommended at all. The operational detail on Drupal is a separate discussion, but the principle holds on any platform.

2. Can AI Personalize a Program Finder?

Yes, because it is a matching problem. Guiding a student from "I like biology, and I want to stay near home" to three real programs has defined inputs and defined outputs. In the same 2025 AI readiness report, 35% of respondents named personalized marketing as the single greatest area of opportunity for AI at their institution, and program discovery is where that pays off first.

3. Can AI Handle Accessibility QA at Scale?

It can do the first pass, not the sign off. AI is genuinely useful for drafting alt text, flagging contrast failures, and generating plain language summaries across thousands of pages. This matters now, because WCAG 2.1 Level AA (the Web Content Accessibility Guidelines conformance level) is the enforceable standard for public universities under the ADA Title II rule.

On April 20, 2026, the DOJ issued an interim final rule extending the compliance date for entities serving populations of 50,000 or more to April 26, 2027. Building to WCAG 2.2 AA is the practical target rather than 2.1, since 2.2 is backward compatible and avoids remediating twice.

The limit here comes from the federal government itself. Among its reasons for the extension, the DOJ cited slower-than-expected technological solutions, including the limits of generative AI for remediation at scale.

AI can draft the alt text. It cannot own the compliance. That distinction between drafting and owning applies to every AI use case in recruitment.

4. Is AI Translation Good Enough for International Recruitment?

For a first experience, yes. In our own university projects, machine translation is now reliable enough to give an international prospect a credible entry point in their own language across a large site, without a full localization project per market. Keep a native reviewer on program names, fees, and eligibility, where an error is a compliance problem rather than a style one. It is a reach multiplier on a defined, reviewable task.

Which AI Recruitment Tools Are Promising but Still Unproven?

Some AI applications are plausible and worth piloting, but do not yet have the evidence to justify betting your funnel on them. Treat these as experiments with guardrails, not decisions.

1. Should You Build an Open-Ended Admissions Chatbot?

Not before a bounded one. The most-cited randomized controlled trial in higher-ed AI recruitment tested a narrow bot. In 2016, Georgia State University pointed a text-message assistant named Pounce at summer melt: the students who accept an offer and then never arrive in the fall. Lindsay Page and Hunter Gehlbach measured a 21.4% reduction in melt and a 3.3 percentage point increase in on-time enrollment across more than 7,000 admitted students.

It worked because fewer than 1% of the 50,000-plus messages it received ever needed a Georgia State staff member, and because every question it answered (what is the FAFSA deadline, did my transcript arrive) had exactly one correct answer already sitting in the university's systems.

The caveat matters as much as the result. When comparable text-nudge campaigns were later run at national scale, Bird and colleagues found the effects largely did not replicate. Narrow and locally owned worked. Broad and scaled did not.

What the evidence covers, then, is the bounded bot. A general-purpose assistant that answers anything a prospect asks is a riskier proposition, because the moment it improvises about deadlines, aid, or eligibility, a wrong answer becomes an institutional statement.

The Pounce result gets retold as "chatbots work." What it actually shows is narrower: automate the questions that already have exactly one right answer, and leave the rest alone.

2. Does AI-Generated Content at Scale Actually Work?

It works for speed, not for persuasion. Producing program pages and campaign copy quickly is easy. Producing content students trust is not. Research on consumer response to AI-generated marketing describes a measurable trust penalty: audiences engage less once they sense a message was machine-written, and labeling it as AI-generated satisfies disclosure without removing the effect.

3. Is Predictive Engagement Scoring Reliable Yet?

It is useful but unsettled. Models that score which admitted students are most likely to enroll can sharpen yield spend, and many teams already use them. The open questions are fairness and feedback loops. A model trained on who enrolled last year can quietly encode who the institution has historically favored, then route attention away from the students least like them. Pilot it with a bias audit and a human review step, not as a standing allocation rule.

Where Is AI Overhyped or Genuinely Risky in Recruitment?

AI is overhyped or risky wherever the task is to generate trust rather than organize it, and wherever it touches identifiable student data without governance in front of it.

1. Can AI Replace Human Admissions Counselors?

No. The counselor’s job is judgment and reassurance during one of the largest financial decisions a family makes, which is the opposite of a bounded-answer task. Pounce, the recruitment bot with the strongest trial evidence behind it, succeeded by not attempting this and by routing human-judgment cases to actual humans.

2. Should AI Write Faculty Bios?

No. A faculty bio is a credibility signal, and credibility cannot be generated. The same trust penalty that suppresses engagement with AI-written marketing applies with more force to the page meant to prove a person is real, and Google’s guidance on helpful content weighs demonstrated first-hand experience and expertise. Synthesizing the one artifact meant to establish a human’s credentials is a bad trade.

3. Can You Use AI on Student Data Without Governance?

No, and this is the sharpest risk of any AI use case in recruitment. General-purpose AI tools can expose FERPA-protected records to third parties or absorb them into model training.

Current institutional guidance is consistent: use contracted enterprise tiers with explicit data-processing terms and no training on student data, or redact at the source so no education record leaves your control. The question is not whether the tool is impressive. It is whether a defensible data path exists before it is switched on.

Which AI Recruitment Use Cases Should Universities Fund?

Sorted into three tiers: fund now (bounded tasks with evidence), guarded pilot (plausible but unproven), and do not (the task is to produce trust, or the data path does not exist).

Our View: AI Should Organize Trust, Not Manufacture It

The line between AI that helps recruitment and AI that hurts it is not about how advanced the tool is. It is about whether you are asking AI to organize trust the institution already has, or to manufacture trust it does not.

The data says the trust is already in the website. In the same UPCEA and Search Influence study, 77% of prospects rated university and college websites extremely or very trustworthy when researching a program, ahead of traditional search engines at 66% and well ahead of social media. Your site is the most credible thing in the search. The job is to make it retrievable, not to replace what makes it credible.

An AI faculty bio, a scaled content mill, and an autonomous counselor all ask the model to produce the credibility signal itself, and that is the one thing recruitment cannot fake. On a 40-subsite university estate where content is already drifting, pointing AI at generation multiplies the drift. Pointing it at organization pulls the sprawl back into order.

This is why the strongest AI opportunity for most universities right now is boring on purpose. Build a large, governed, accessible content base that answer engines can confidently cite, rather than generating more content nobody vouches for.

Image

How Should You Decide Which AI to Fund? Three Questions

Before funding any AI recruitment project, run it through three questions. If it fails the first, do not automate it. If it fails the third, do not start until governance exists.

The universities getting real value from AI in recruitment are not the ones using it most. They are the ones who drew this line early and stayed on the right side of it.