Report · Research Center

AI in Education Report 2026: Students Are Ahead of Their Institutions

88% of students use AI; 77% of faculty do. But 57% say their assessments carry inadequate AI guidance and only 29% believe their instructors can advise them. The adoption is settled — the policy is not.

AI in Education Updated 2026-08-04 788 words · about 4 min read

The debate about whether AI belongs in education is over. Students settled it without waiting.

88% of students now use AI in their learning. Among university students, usage rose from 66% in 2024 to 92% in 2025. 88% use generative AI specifically for assessments, up from 53% the year before.

The open question is no longer adoption. It is that institutions have not caught up with a change their students completed two years ago.

The numbers#

Measure2026
Students using AI in learning88%
University students using AI (2025)92% — from 66% in 2024
Students using generative AI for assessments88% — from 53%
Faculty using AI in teaching77% — up 16 points on 2025
K-12 teachers using generative AI63%
Higher-education instructors49%
High school teachers69%
Elementary teachers42%
Pre-K teachers33%
Students aged 13–17 using ChatGPT for assignments26% — from 13% in 2023

The gap that matters#

Adoption is near-universal on both sides. Support is not.

Measure
Students saying assessments carry inadequate AI guidance57%
Students who believe instructors are equipped to guide them on AI29%
Faculty who agree their institution involves them in shaping AI policy31%
Students concerned AI makes learning too shallow65%

Read those together and a specific picture emerges: students are using AI on assessments without clear rules, taught by faculty who mostly use it themselves but are not confident advising on it, inside institutions whose policies were written without either group.

That is not a technology problem. It is a governance vacuum, and everyone in it knows.

The most interesting finding#

65% of students are themselves worried that AI makes learning too shallow.

This is not the finding the debate expects. The population using AI most heavily is also the population most concerned about what it costs them. Students are not seeking to avoid learning — they are using an available tool under time pressure, while suspecting it may be undermining the thing they are there for.

That suggests policy framed as restriction misreads the room. Students appear open to guidance about when AI helps and when it hollows out the work. Almost nobody is giving it: only 29% think their instructors could.

What the evidence says about outcomes#

One controlled result stands out: a 2025 Harvard physics study found students using AI tutors learned more than twice as much in less time than those in traditional active-learning classrooms.

Handle this carefully. It is a single study, in one subject, with a structured tutor rather than a general chatbot, and active-learning classrooms are already a strong baseline. It does not generalise to "AI improves education". What it does show is that well-designed AI tutoring can outperform good conventional teaching on measurable learning — which is a meaningful existence proof, and quite different from students using a chatbot to draft an essay.

The distinction between those two uses is exactly what institutional guidance should be making, and mostly is not.

How it is actually used#

ApplicationShare
Research and content gathering44%
Creating lesson plans38%
Summarising information38%
Generating classroom materials37%

Faculty use clusters on preparation, not assessment or instruction. That is the low-risk, high-tedium quadrant — the same pattern every other sector shows.

What we would do#

For institutions: write assessment-level guidance, not blanket policy. "AI is permitted for X, not for Y, and here is why" is usable. A prohibition nobody enforces teaches students that the rules are decorative.

Involve faculty in writing it. 31% currently feel involved. A policy the teaching staff did not shape will not be taught, and only 29% of students think their instructors could advise anyway.

Redesign assessment rather than police it. If a task can be completed by a chatbot in thirty seconds, it was measuring the wrong thing before AI existed.

Take the shallowness concern seriously. 65% of students raised it unprompted. That is an opening for a conversation about when to use the tool, which is more durable than detection.

Method and limitations#

Synthesis of published 2026 surveys of students and faculty. Not primary research.

Self-reported usage surveys in education have particular biases: respondents may under-report use they believe is discouraged, and survey populations skew toward institutions willing to be surveyed. Figures vary between sources — student adoption appears as 88% and 92% in different studies with different definitions and periods; we have shown both. The Harvard result is a single study and should not be generalised.

Published 2026-08-04.

Sources#

What else is coming for AI in Education

Report Ready

The findings, with sources.

Data Not yet

The underlying figures.

Method Not yet

Where each number came from.

Updates Not yet

What changed since publication.