AI Is Here: Is Your Admissions Policy Ready?

A new report from Student Defense raises a question that colleges and universities will increasingly have to answer: Who is responsible for deciding how AI will be used in admissions?

The organization submitted public-records requests to 24 institutions in spring 2026, asking for policies governing AI use in undergraduate admissions and evidence of training for staff working across admissions, recruitment, and financial aid. Of the 20 institutions that responded, none produced a policy specifically governing AI use in undergraduate admission decisions, and only one provided AI training materials specific to enrollment management.

Those findings deserve attention, although perhaps not because higher ed institutions have somehow failed to produce enough policy documents.

• AI has moved into enrollment work unusually quickly, often through tools and capabilities embedded within systems institutions already use.
• Admissions office teams are experimenting while simultaneously trying to understand where automation is appropriate, what requires disclosure, how student data should be protected, and which decisions should remain firmly within the province of human judgment.
• Policies can establish boundaries around those questions, but governance requires higher ed institutions to develop ways of answering them repeatedly as technologies, practices, and circumstances change.

That distinction may become particularly important as institutions move from relatively straightforward uses of generative AI—drafting communications, summarizing information, or assisting with administrative work—toward predictive and analytical systems that can inform recruitment, financial aid, enrollment forecasting, and student success.

It also gives us an opportunity to think about governance across admissions rather than exclusively within undergraduate admissions. Graduate enrollment frequently distributes decision-making among central admissions offices, graduate schools, individual colleges, academic departments, and faculty committees.

Professional programs may operate under still another set of expectations, accreditation requirements, and review practices. The people using AI, the people interpreting its output, and the people ultimately accountable for a decision may therefore sit in different parts of the institution, which means a useful governance model must be capable of functioning across that organizational complexity.

Rather than trying to anticipate every possible use of AI through a growing list of permitted and prohibited technologies, institutions can establish a governance model around five practical areas: Purpose, Authority, Evidence, Judgment, and Review.

Together, they provide a structure that can apply across undergraduate, graduate, and professional admissions even when decision-making responsibilities differ considerably.

1. Purpose: Define What AI is Being Asked to Do

Institutions should document the specific purpose of each AI-enabled process and the enrollment decision or activity it is intended to support. Uses such as applicant communication, recruitment prioritization, enrollment forecasting, application review, and student support carry different levels of risk and should not automatically be governed in the same way.

A useful standard is to require teams to identify the problem being addressed, the intended outcome, and why AI is appropriate for that particular use before implementation.

1. Authority: Establish Who Can Use AI and Where

Governance should identify who is authorized to use AI-supported information, which activities it may inform, and who remains accountable for decisions. This is particularly important in graduate and professional admissions, where responsibility may be distributed among central enrollment teams, graduate schools, academic programs, and faculty committees.

Training should follow those responsibilities so that users understand not only how to use a system but also how to interpret its outputs and recognize its limitations.

1. Evidence: Make AI-supported Insights Understandable

Institutions should be able to explain what information informs an AI-supported process and how its outputs should be interpreted. Predictive systems, for example, should be connected to clearly defined institutional questions, appropriate data, and established measures of performance.

Liaison’s Othot analytics platform provides one example of this approach through its use of institution-specific “High Impact Questions,” which establish what leaders want analytics to help them understand before models are developed around institutional data. That process helps connect the technology to a defined enrollment objective rather than introducing analytics without a clear decision context.

1. Judgment: Define Where Human Responsibility Remains

Institutions should explicitly determine which activities AI may support and which decisions require human judgment. AI may appropriately organize information, identify patterns, summarize materials, or help teams prioritize outreach without being given authority to make consequential graduate school and college admissions decisions.

Liaison’s AI Application Summary illustrates one version of this boundary: It provides an informational overview of college application materials without scoring, ranking, or recommending applicants. Similar distinctions can help institutions establish practical limits around AI use without prohibiting technologies that can meaningfully support staff.

1. Review: Revisit AI-supported Processes Regularly

AI governance should include a regular review of performance, outcomes, user understanding, and emerging risks. This is especially important in enrollment management, where applicant behavior, institutional priorities, financial conditions, and public policy can change quickly enough to affect the usefulness of models and workflows.

For institutions using predictive analytics, ongoing support can become part of this review process. Othot combines institution-specific models with expertise that helps enrollment teams:

• Interpret results.
• Evaluate assumptions.
• Translate insights into strategy.
• Provide a structure for continued assessment rather than treating implementation as the end of the process.

Taken together, these five areas give institutions a practical way to govern AI usage without requiring a separate policy for every new capability. They also allow governance to operate across different admissions structures while preserving a consistent expectation that institutions can explain why AI is being used, who is responsible for its use, and how its role in decision-making will be evaluated over time.

Student Defense is right to call attention to the absence of formal policies and training. Institutions should know how AI is being used in graduate and college admissions, establish appropriate guardrails, protect applicant data, and make clear where human responsibility resides.

But the work cannot reasonably end when a policy is approved, because responsible AI use will increasingly depend upon the institutional capacity to ask good questions about technology: what problem it is solving, who has authority to use it, what evidence sits behind its conclusions, where human judgment enters the process, and how the institution will evaluate its performance over time.

Those questions apply whether a higher education institution enrolls 5,000 first-year undergraduates through a centralized admissions operation or 50 doctoral students through faculty-led committees distributed across academic departments. The technologies may be similar even when the organizational contexts surrounding them are profoundly different, which makes it difficult to imagine a universally applicable set of rules governing every possible use.

A more durable approach may be to build governance practices capable of traveling with the technology as it moves through different parts of the institution and takes on new functions. Institutions that develop that capacity can approach AI with greater confidence because they have established a way to determine what they are asking technology to do, how its outputs should be understood, who remains accountable for the decisions that follow, and how those arrangements will be reconsidered as the technology and the enrollment environment continue to change.