AI use among college students

About 57% of U.S. college students use AI in their coursework at least weekly, and about 21% use it daily, according to the 2026 Lumina Foundation–Gallup State of Higher Education study. Instructure's research puts occasional classroom AI use even higher, at 90% of students. Campus AI policies and faculty training still lag behind how students already use these tools.

How often college students use AI

The Lumina Foundation–Gallup 2026 State of Higher Education findings show AI is already a weekly habit for a majority of U.S. associate and bachelor’s students (Lumina Foundation–Gallup State of Higher Education 2026):

• 57% use AI in coursework at least weekly; about one in five (21%) say they use it daily.
• Daily use is higher among men (27%) than women (17%).
• Business, technology, and engineering majors report the most frequent use; associate and bachelor’s rates are similar overall.
• Top weekly uses include help with material they do not understand (64%), checking answers (60%), editing writing (54%), summarizing notes (54%), and generating ideas (49%).

Instructure AI 2.0: use, trust, optimism, and training

Instructure’s AI in education research (AI 2.0) surveyed higher-ed students, educators, and K–12 parents/guardians in the U.S. Higher-ed highlights (Instructure AI in education research (AI 2.0)):

• 90% of higher-ed students use AI in class at least occasionally; 61% of higher-ed educators and 68% of K–12 educators report the same; 73% of parents say their child uses AI at least occasionally.
• Trust is mixed: students report the highest trust; educators sit just above the midpoint. Across audiences, people worry AI can sound confident when wrong—65% of higher-ed educators and students, 69% of parents, 60% of K–12 educators.
• Top shared concerns: overreliance on AI; loss of critical thinking and deep learning; academic integrity and plagiarism.
• Optimism remains high: 94% of higher-ed students and 82% of higher-ed educators expressed optimism (vs 88% of parents and 77% of K–12 educators).
• Why higher-ed students are optimistic: accessibility for diverse learning needs (41%), learning gains (33%), time savings (32%), personalized learning (30%), AI-enhanced tutoring (29%). Educators more often cite time savings (37% higher ed; 44% K–12). Parents most often cite accessibility (51%).
• Accepted uses vs. red lines: AI is more accepted for support tasks (explaining concepts, brainstorming, finding resources) and least accepted for final grades or academic decisions.
• Restrict use is a minority view in higher ed: 39% of educators and 28% of students favored restricting student AI use. Teaching responsible AI use and critical AI literacy ranked first or second among expectations for schools.
• Educator training lags use: among higher-ed educators who answered—41% no formal training, 35% some, 11% comprehensive. Average AI confidence sits well below the midpoint of a 1–5 scale in both higher ed and K–12. Educators want hands-on tutorials, clear institutional policy, and ongoing professional development.
• 70% of parents said research evidence on generative AI’s impact on teaching and learning is very or extremely important to them.

Prefer the primary Instructure brief over secondary trade headlines when citing the 90% / 61% adoption pair—the full brief also covers trust, optimism, boundaries, and training depth.

Campus rules vs. student behavior

Gallup–Lumina reports a clear policy gap: more than half of enrolled students say their school discourages AI (42%) or prohibits it (11%), while about four in ten say AI is encouraged freely (7%) or with limits (35%). Technology, vocational, and business programs are more likely to encourage use.

Among monthly-or-more users, nearly nine in ten say understanding complex material is an important reason to use AI; saving time and grades follow closely. Non-users most often cite ethics and school policy.

That mismatch—majority weekly use plus majority discouragement/prohibition—lines up with Instructure’s finding that audiences want literacy and guardrails more than blanket bans.

Cheating, access gaps, and grades

UC Berkeley’s Center for Studies in Higher Education led the largest published undergraduate GenAI survey to date—tens of thousands of students across research-intensive public universities in the SERU Consortium (findings in Science, 2026; SERU Consortium / Science (Chirikov et al.)). Coverage: Berkeley News.

• About two-thirds (~67%) of respondents used generative AI; almost 40% used it monthly or more.
• At least 9% of students who used AI reported using it to cheat—higher among daily users (26%) than monthly users (7%).
• Low-income, racially underrepresented, and female students reported less AI use—raising equity concerns as paid models become more capable.
• Authors argue blanket bans are a weak fix; discipline-specific assessment redesign matters more.

Related CSHE working paper on grade inflation (Chirikov, 2026): in courses with more AI-exposed tasks (writing, coding), the share of A grades rose about 13 percentage points after ChatGPT’s release (~30% relative to the 2022 baseline)—consistent with AI substituting for student work on homework-weighted courses (CSHE working paper).

Experimental work on skill formation also cautions that heavy AI delegation can impair conceptual learning even when output looks polished (Shen & Tamkin). USC’s Center for Generative AI and Society similarly distinguishes “executive help” (shortcuts) from “instrumental help” (learning support)—and finds faculty guidance strongly shapes which path students take (USC Today).

What faculty report

College Board surveyed thousands of U.S. faculty:

• 74% say students use AI to write essays or papers; 67% cite paraphrasing/rewriting.
• More than 84% agree AI reduces critical thinking, originality, and deep engagement; 92% worry about AI-facilitated dishonesty.
• Only 21% feel very confident guiding classroom AI use; about half report a formal classroom policy.
• Sentiment is more negative at selective institutions and in writing-intensive fields; STEM and business faculty report more personal AI use and somewhat more positive views.

A Frontiers in Psychology mini-review of student and academic well-being effects of AI in higher education is available via PMC11830699.

Nature’s careers feature (Sept 2026) documents how professors worldwide are redesigning assessment—oral exams, in-class writing, AI critique assignments, and process logs—rather than relying on detectors alone.

Degrees, enrollment, and the AI talent pipeline (Stanford AI Index 2026)

The Stanford HAI 2026 AI Index — Education chapter places student tool use inside a wider supply-and-skills story:

• ~4 in 5 U.S. high school and college students now use AI for schoolwork (cross-level synthesis; school policies lag—especially in K–12, where only about half of middle/high schools have AI policies and 6% of teachers say policies are clear).
• CS enrollment at U.S. four-year universities fell 11% from 2024 to 2025, while master’s graduates in AI software-related fields rose 17% from 2023 to 2024—specialization demand holding up as broader CS cools.
• The U.S. remains a global leader in ICT graduates, but Turkey, Brazil, and Mexico have increased ICT graduate output faster in recent years.
• New AI PhDs in the U.S. and Canada rose 22% from 2022 to 2024; growth went to academia while the industry share stayed flat—reversing a decade-long industry tilt.
• Outside formal degrees, people are advertising AI skills on resumes; AI literacy has grown faster than engineering-oriented AI skills in most countries.
• More than 90% of countries now offer computer science in primary or secondary schools; formal AI education is slower, though China and the UAE mandated AI education for 2025–26.

Majors and career paths under AI pressure

Students are not only using AI for homework—they are revisiting what to study:

• CNBC (July 2026): 31% of students/workers enrolled in school changed or considered changing their target industry due to AI (39% among students not employed); nearly 40% reconsidered major/field/coursework (44% among non-employed students).
• Gallup–Lumina reporting cited in the same coverage: about 42% of bachelor’s and 56% of associate students gave at least “a fair amount” of thought to changing majors over AI concerns.
• Broader State of Higher Education 2026 framing—value remains high while affordability confidence falls—lives on the Gallup–Lumina research hub (including “Valued but Out of Reach”).

Commentary such as The Atlantic on integrity norms describes how take-home and fully online formats face different enforcement limits than proctored, in-person assessment.

Policy and research framing: UNESCO and IES REL Northwest

UNESCO’s artificial intelligence in education work stresses a human-centred approach: AI may help address access and SDG 4 goals, but risks have outpaced regulation. Flagship products include Artificial intelligence and education: Guidance for policy-makers (Beijing Consensus framework), AI competency frameworks for students and teachers, and the Recommendation on the Ethics of Artificial Intelligence. Useful for campus leaders writing literacy policies—not a U.S. student-usage survey.

The U.S. Department of Education’s REL Northwest AI in education resource sheet (Jan 2025) synthesizes empirical literature into five common use categories and four implementation considerations:

• Use categories: AI-driven instruction/tutoring; personalized learning; assessment; predictive and learning analytics; administrative/logistical tasks.
• Considerations: address ethics, bias, and privacy; treat AI apps as continuous improvement cycles; prepare people to work with AI; use AI as a tool, not a human substitute.
• Evidence snapshots cited in the sheet: classic intelligent tutoring (e.g. PUMP/PAT algebra) showed ~15% gains on standardized tests in the 1990s evaluation literature; hybrid human–AI personalized math tutoring studies report roughly double the standardized-score gains vs. comparison groups—illustrating “AI + human” rather than AI-alone models.

Universities and colleges with AI company deals

Campus–vendor partnerships are growing. They usually buy licensed access (often “Edu” or Enterprise tiers), privacy terms for prompts, and faculty pilots—not a single shared academic-integrity policy. Always verify current contract status on the institution’s site.

What this means if you are choosing an AI degree

• Ask about assessment design—oral defenses, in-class work, and project process matter more when take-home writing is AI-saturated.
• Ask how the school teaches responsible AI use in your major—not only whether ChatGPT Edu is available (Instructure: literacy > bans; faculty training is still sparse).
• Watch the pipeline, not just the chatbot: AI Index shows CS cooling and AI master’s/PhD growth—credentials still matter for specialization.
• Build skills AI does not replace easily: problem framing, verification, domain judgment, and security/privacy literacy.
• Separate teen habits from college evidence when citing stats—see our teens page for Pew and College Board high-school figures.

Sources & methods

Technical details for journalists and fact-checkers—kept out of the findings above.

• Gallup–Lumina: web survey Oct. 2–31, 2025; U.S. associate and bachelor’s students. Gallup release.
• Instructure AI 2.0: three parallel U.S. surveys June 12–15, 2026; total n=1,125 (410 educators: 296 K–12 / 114 higher ed; 312 current/recent higher-ed students graduated after Jan. 2023; 403 K–12 parents/guardians). Educator AI confidence means: 2.60/5 higher ed, 2.53/5 K–12. Instructure brief.
• Berkeley SERU / Science: >95,000 undergraduates at 20 research-intensive public universities; spring 2024 fielding; published May 2026. DOI 10.1126/science.aec5115.
• College Board faculty: 3,000+ U.S. faculty; summer 2025 fielding; brief Feb. 2026.
• CNBC: July 2026 survey (fielded via SurveyMonkey) of students/workers enrolled in school—career/major reconsideration items.
• Stanford AI Index 2026: Education chapter synthesis—not a primary campus survey.
• UNESCO / IES REL Northwest: policy guidance and evidence synthesis—not U.S. usage polls.

Related guides & rankings

Frequently asked questions