The Impact of AI on Work in Higher Education

Synthesis: EDUCAUSE's 2026 national survey of 1,960 higher education employees finds that work-related AI use has become ordinary while the institutional scaffolding around it has not caught up. Nearly all respondents (94%) used AI tools for work in the past six months, yet only 54% were aware of policies and guidelines meant to guide that use, and 56% used tools their institution does not provide. Institutions report strategies at scale — 92% have a work-related AI strategy — but only 13% measure return on investment, and fewer than half of respondents feel their policies give them clarity. The report pairs this governance gap with a stable emotional profile: 81% feel enthusiastic or a mix of caution and enthusiasm, and majorities simultaneously rate many risks as urgent and many opportunities as promising. Read as an evaluation of institutional readiness rather than of learning outcomes, the study locates the near-term problem in governance, communication, and AI literacy rather than in tool availability, and it shows that institutional AI infrastructure decisions about approved tools, data handling, and privacy are being made in practice by employees acting without clear guidance.

Key Findings

1. Work-related AI use is near-universal and mostly voluntary. 94% of respondents reported using AI tools for work in the past six months; of those, 73% use them daily or weekly. Only 11% said they are required to use AI tools, while 64% said they are not required and do not expect to be.
2. Policy awareness lags far behind adoption. Just 54% of respondents were aware of policies and guidelines for work-related AI use, leaving 46% unaware. Even 38% of executive leaders, 43% of managers and directors, 35% of technology professionals, and 30% of cybersecurity and privacy professionals reported no awareness.
3. Strategies are widespread but measurement is thin. 92% of respondents said their institution has a work-related AI strategy, and only 5% said it has none. Yet just 13% said their institution measures the return on investment for work-related AI tools, and most of those respondents said they were not sure how it was done.
4. Use spans a wide range of tasks, not a single workflow. More than half of respondents (54%) reported using AI for eight or more of 30 listed task types. The most common were brainstorming (63%), drafting emails (62%), summarizing long documents or meetings (61%), proofreading or copyediting (56%), and creating presentations (47%).
5. Risk concern and opportunity optimism coexist at similar breadth. 67% of respondents rated six or more risks as urgent, and 67% rated five or more opportunities as most promising. The leading risks were increased misinformation (55%), use of data without consent (52%), loss of fundamental skills requiring independent thought (51%), and insufficient data protection (51%); the leading opportunities were automating repetitive processes (70%), offloading administrative burdens and mundane tasks (65%), and analyzing large datasets (60%).
6. Institutions favor upskilling over new hires. Among respondents whose institution's strategy includes increasing AI-related workforce skills, 69% said the institution is doing so primarily by upskilling or reskilling existing staff and faculty. Within that group, 80% said the institution encourages employees to develop skills on their own and 71% cited in-house professional learning.
7. Access gaps and unapproved tools shape the real environment. 51% of respondents said their institution provides some but not all of the AI tools they want, 39% said it provides all of them plus some, and 10% said it provides none. Nearly a quarter of faculty respondents (23%) said their institution provides none of the tools they want, and 56% of all respondents reported using tools not provided by their institution.

Adoption outruns policy

The report's central tension is not resistance but unaccompanied practice. Respondents are not waiting for permission: 94% have used AI for work, and the reasons non-users give are lack of interest (58%) or lack of need (50%) rather than prohibition. Only 15 respondents who had not used AI for work cited uncertainty about institutional or unit-level policies as the reason, which the authors read as evidence that missing or unknown guidance does not deter use. That interpretation matters for governance: a policy that exists only in a handbook, without socialization, produces the same operational exposure as a policy that was never written.

The awareness breakdown sharpens the point. Because 38% of executive leaders and 35% of technology professionals report no awareness of guiding policies, the authors argue that many institutions simply do not have such policies rather than that communication has failed. Decision authority is distributed — institution-wide leadership (59%), IT (53%), and an AI governance unit or committee (38%) were the most-cited owners — which helps explain why coherent guidance is hard to produce and easy to leave unfinished.

Where policies do exist, they lean permissive: 47% of respondents with policies characterized them as somewhat or extremely permissive and 30% as neutral. Permissiveness has not translated into confidence. Only 34% said they feel confident and 22% very confident that current policies give them clarity for work-related AI use.

What staff and faculty actually do with AI

The task list shows a practice built around text and preparation rather than judgment-heavy work. Brainstorming, email drafting, summarizing, proofreading, and presentation building lead; writing policies or procedures (40%), taking notes (39%), and creating training materials (38%) follow. The least-selected uses include scheduling meetings (9%), writing blog posts (9%), onboarding new employees (7%), and application screening (6%). The authors offer several explanations without settling the question: a preference for professional over personal communication, avoidance of high-stakes activities, and perceived poor fit of available tools.

One role difference is substantial enough to report: 63% of faculty said they had used AI to create learning activities or assessments, compared with 32% of staff. That split matters for AI literacy programming, because the faculty use case touches assessment design and therefore academic integrity adjacent decisions, while staff use concentrates on operational writing and analysis.

Frequency is also uneven in a way that complicates simple adoption claims. Among recent users, 38% use AI daily and 34% weekly, while the remainder use it less often. Breadth of task use (54% at eight or more types) therefore describes a large experienced segment rather than every respondent.

Risk, opportunity, and the capacity gap

Respondents hold both sides at once, and at similar intensity. The risk ranking is dominated by epistemic and data concerns: misinformation (55%), data use without consent (52%), skill loss (51%), and insufficient data protection (51%). Copyright and intellectual property violations and student AI use outpacing faculty and staff skills both appear at 47%. Job replacement, notably, sits at 21%, well below the information-integrity concerns.

The opportunity ranking is operational: automating repetitive processes (70%), offloading administrative burdens (65%), analyzing large datasets (60%), generating insights for data-informed decision-making (53%), and real-time data analytics (51%). Job transformation (39%) also outranks job replacement (5%) on the opportunity list.

What blocks progress is capacity, not conviction. The top challenges were AI's pace of change (60%), lack of AI expertise (55%), lack of best practices (48%), lack of time to learn new skills (46%), and the number of risks associated with AI (41%). Insufficient institutional governance (36%) and lack of professional development support (35%) round out a picture in which the constraint is organized support rather than willingness. For administrators, the implication is that privacy and data-protection risk is being managed by individuals making local choices, including the 56% who use tools the institution has not provided and therefore has not evaluated for data handling, accuracy, accessibility, or copyright.

From strategy to governance

The strategy data suggest a field that has produced documents faster than it has produced measurement. 64% of respondents selected five or more elements in their institution's strategy, and the most common elements were piloting tools (65%), evaluating opportunities and risks (60%), encouraging use (59%), and creating policies and guidelines (54%). Only 5% said their institution discourages or prohibits work-related AI use. Yet the ROI finding — 13%, with many of those respondents unsure of method — indicates that most institutions cannot say whether the tools are worth what they cost.

The authors' recommendations follow directly: communicate strategy and include staff and faculty in planning; provide formal training rather than relying on self-directed learning; write AI responsibilities into job descriptions; measure ROI and explain how tools are evaluated; create and socialize policies; and ask employees which tools they want before procuring them. These are governance and workforce actions, and they assume that institutional AI infrastructure includes the human and policy layer, not only procurement and networks.

What this means for practice

• Administrators. Treat policy awareness as the metric to move, not policy publication: 46% of respondents were unaware of guidance, including 38% of executive leaders, so measure awareness by role and unit rather than assuming a campus-wide memo landed.
• Faculty developers. Build on the tasks people already do — brainstorming (63%), summarizing (61%), and proofreading (56%) — and address assessment design explicitly, since 63% of faculty have used AI to create learning activities or assessments.
• IT and governance staff. Close the unapproved-tool gap deliberately: 56% of respondents use tools the institution does not provide, so approved alternatives and clear data-handling guidance reach further than prohibition.
• Institutional researchers. Pair adoption and attitude items with ROI and outcome measures; at 13% measurement, the field currently has ample self-report on use and very little on value.

Limitations

• All measures are self-reported by a single respondent per institution, so reported policy awareness, use frequency, and institutional strategy reflect individual perception rather than verified institutional practice.
• The survey is a voluntary, non-probability sample with a constrained sample frame and standard response biases; the authors note this limitation and do not claim a representative census of U.S. higher education employees.
• Inclusion criteria required respondents to be currently employed at a higher education institution and to answer four specific items, including attitude toward AI and policy awareness, which excludes otherwise eligible employees and may bias the analyzed pool toward those with formed views.
• The report was sponsored by Zoom as a 2026 EDUCAUSE Mission Partner; the report states that the sponsor had no editorial influence, but sponsorship remains a funding relationship that readers should weigh.
• The design is cross-sectional, capturing attitudes and use from September 29 to October 13, 2025, so it cannot establish trends, causal effects of AI use, or whether policies changed behavior over time.
• Several findings rest on small subgroups — faculty were 12% of respondents, and only 13% reported ROI measurement — so role-specific percentages such as the 63% faculty assessment figure and the 23% faculty access figure carry wider uncertainty than campus-wide totals.

Citation

Robert, J. (2026). The Impact of AI on Work in Higher Education. EDUCAUSE.