AI in education has moved from experiment to habit in about two years. In the UK, 95% of university students now use AI in at least one way, and more than half of American students and teachers used it for school in 2025. The learning evidence is far less settled than the adoption numbers: the same tool can double learning gains in one study and lower exam scores in another, depending on how it is used.
This post collects the most reliable recent figures on who uses AI, what it is used for, what controlled trials show about learning, and where the risks sit. Every number comes with the organisation that published it, so you can judge the source as well as the figure.
Key Findings at a Glance
If you only need the headline numbers, these are the ones that hold up best:
- 95% of UK university students use AI in some way, and 94% use generative AI for assessed work (HEPI, 2026).
- 54% of US students and 53% of US English, math and science teachers used AI for school in 2025, up more than 15 points in one to two years (RAND).
- 5.9 hours a week is what US teachers who use AI weekly say it saves them, about six weeks over a school year (Gallup and the Walton Family Foundation, 2025).
- 17% lower exam scores for high school students who practised with an unrestricted GPT-4 chat, compared with students who had no AI (PNAS, 2025).
- More than double the median learning gains for college students taught by a carefully designed AI tutor, compared with an active learning class (Scientific Reports, 2025).
- Half of US students worry about being falsely accused of cheating with AI (RAND, 2025).
The pattern behind these numbers is consistent: use is close to universal, guidance trails behind it, and the effect on learning depends on design more than on the tool itself.
How Many Students Use AI in Education
Student adoption is the fastest moving number in this field, and it differs a lot by age and country.
University students
The most detailed picture comes from the UK. According to the 2026 HEPI survey, 95% of students report using AI in at least one way and 94% use generative AI to help with assessed work. Many of them are now encouraged or required to use it, as universities start teaching and assessing AI skills.
The share of students who paste AI-generated text directly into assessed work is still small but rising quickly: 12% in 2026, up from 8% in 2025 and 3% in 2024. Nearly two-thirds of students (65%) say assessment at their university has changed significantly because of AI.
Students are also using it beyond coursework. About 15% say they use AI for companionship, advice or to deal with loneliness, and the group is split almost evenly between those who say AI makes them lonelier (20%) and less lonely (21%).
School students
Younger students are catching up. The RAND survey panels found that 54% of US students used AI for school in 2025, with use higher in high school than in middle school. Pew Research Center tracked a similar rise for one tool: the share of US teens aged 13 to 17 who used ChatGPT for schoolwork doubled from 13% in 2023 to 26% in 2024.
The gap that keeps appearing is between use and instruction. More than 80% of US students told RAND that their teachers had not explicitly taught them how to use AI for schoolwork, and only 35% of district leaders said they offer students any AI training.
How Teachers Use AI in Education
Teacher adoption is slower than student adoption, but it is growing every year and the time savings are large for those who stick with it.
What teachers use it for
In the 2024-25 school year, six in ten US public school teachers used an AI tool for their work, according to a Gallup survey of 2,232 teachers run with the Walton Family Foundation. Only 32% used it at least weekly, and 40% did not use it at all.
The most common tasks were preparing to teach (37% of teachers at least monthly), making worksheets or activities (33%), adapting materials to student needs (28%) and administrative work (28%). Grading came much lower at 16%, and one-on-one instruction at 14%.
A second survey shows the trend continuing into 2026. Among 1,011 US math and science teachers surveyed for the AmplifyGAIN Center and fielded by RAND, 58% used generative AI for instructional tasks in 2026, up from 50% a year earlier, and frequent users rose from 21% to 31%.
The time teachers save
The headline figure from the Gallup poll is the "AI dividend". Teachers who use AI weekly estimate they save 5.9 hours a week, which adds up to about six weeks over a 37.4-week school year. Monthly users save roughly half as much, 2.9 hours a week.
Policy matters here too. Schools with an AI policy see a 26% larger time saving per teacher, yet only 19% of teachers say their school has one. Teachers report reinvesting the time in more detailed feedback, individualised lessons and messages to parents.
AI in Higher Education: Policy Lags Behind Use
Universities show the same gap between behaviour and guidance, just at a higher level of adoption. In the HEPI data, only 36% of students feel their institution encourages them to use AI, and only 38% say their institution provides AI tools.
Students also feel underprepared for what comes next. 68% believe AI skills are essential to thrive in today's world, but only 48% feel their teaching staff are helping them build those skills. Arts and humanities students feel the least supported.
The result is a split campus. About as many students agree that their university encourages AI use (37%) as disagree (36%), and their study habits split three ways: 33% lean toward traditional sources, 29% use an even mix, and 37% favour AI sources. The OECD published a 2026 report on policies for responsible generative AI adoption in higher education, a sign that governments now treat this as a system-level question rather than a matter for individual lecturers.
What Research Says About Learning Outcomes
Adoption surveys tell you how often AI is used. Controlled trials tell you whether it helps anyone learn, and here the results point in two directions at once.
When an AI tutor helps
A randomized trial published in Scientific Reports compared college students who learned from a custom AI tutor with students in an in-class active learning lesson. The tutor was built on the same teaching practices as the class.
Students in the AI group showed median learning gains more than double those of the in-class group, and they did it in less time: a median of 49 minutes, compared with a 60-minute lesson. They also reported feeling more engaged and more motivated. The key word is design: this was a tutor shaped by teachers, not an open chat window.
When AI hurts learning
A large field experiment with nearly 1,000 high school math students in Turkey tested two GPT-4 tools. One was a standard chat interface similar to ChatGPT. The other was a tutor with teacher-designed prompts that gave hints instead of answers.
During practice, both helped: grades rose 48% with the standard chat and 127% with the tutor. Then access was removed for the exam. Students who had practised with the standard chat scored 17% lower than students who never had AI, while the tutor group performed about the same as the control group.
The researchers traced the drop to students using the chat as a crutch, asking for answers rather than working through problems. The standard chat was also wrong often, giving a correct answer only 51% of the time. Students in both AI groups thought they had learned more than they had.
The wider evidence base
A November 2026 review from Stanford's SCALE initiative pulled together 52 causal studies of AI in education. Its central finding fits both trials above: AI tends to work better for learning when students use it for learning support than when they use it to produce work, and neither use consistently improves results.
The review adds a warning that is easy to miss. AI often improves how students feel about learning even when test results do not move. A tool that feels helpful is not the same as a tool that helps.
Benefits of AI in Education
Put together, the research points to a handful of benefits that show up repeatedly in the data.
Time. For teachers, the time saving is the most consistent finding, with weekly users reporting almost six hours a week back. For students, 49% in the HEPI survey say AI has improved their experience, mostly by saving time, improving understanding and giving instant support.
Personal pacing. In the Scientific Reports trial, students set their own pace, and time spent did not predict their scores. 70% of the AI group finished in under an hour and 30% took longer, with no link between time spent and score.
Instant feedback. Stanford's review reports that AI can increase the quantity and quality of feedback teachers give and help them encourage students to complete assignments.
Narrowing gaps during practice. In the PNAS study, both AI tools reduced the spread of grades during practice, narrowing the gap between stronger and weaker students. That effect did not survive into the unassisted exam, which is why design and guardrails keep coming up.
Pros and Cons of AI in Education: The Risks
The cons are less about the technology failing and more about how it gets used without guidance.
Skill erosion. 61% of US parents, 55% of high school students and 48% of middle school students agree that more AI use will harm students' critical thinking, according to RAND. Only 22% of district leaders agree. The PNAS result gives the worry some weight: unrestricted help during practice led to weaker unassisted performance.
False accusations. Half of US students say they worry about being falsely accused of cheating with AI, and HEPI reports similar anxiety in UK universities. Detection tools add to the problem, and our guide to how AI detectors work explains why they misfire on genuine student writing.
Uneven policy. Only 45% of US principals reported having AI guidance, and only 34% of teachers had a policy on AI and academic integrity. Teachers are also divided on the outcome: in the 2026 AmplifyGAIN survey, 29% rated AI's impact on student learning positively and 32% negatively.
Side effects on motivation. One study in Stanford's review found that giving teachers in Türkiye a broad AI teaching tool reduced student motivation and confidence, with no significant effect on test scores.
How to Use AI So It Supports Learning
The research gives a fairly clear rule: use AI to make you think, not to think for you. In practice that means a few habits.
Ask for hints, questions and explanations instead of finished answers. The tutor that gave hints in the PNAS study avoided the exam drop that the answer-giving chat caused.
Test yourself without the tool. Since students consistently overestimate how much they learned with AI, the only honest check is recalling the material on your own. Our guide to active recall techniques covers ways to do that.
Use AI on your own material. Turning a lecture or reading into practice questions keeps you working with the content you will be tested on. Primo Notes does this from a recording, a photographed page, a PDF or a YouTube link, turning the note into a quiz, flashcards or a mind map you then work through yourself.
Explain it back. Explaining material out loud exposes the gaps that rereading hides. Primo Notes includes a Feynman mode, a guided spoken session where you explain a topic and the app works through what you missed, which puts the effort on you rather than on the AI.
Conclusion
AI in education is now part of daily study for most university students and a growing share of school students and teachers. Adoption has outrun guidance at almost every level, from schools without policies to students who were never shown how to use the tools well.
The learning evidence is mixed but not random. AI built to guide, hint and question can help students learn more in less time. AI used to produce answers can raise practice scores while lowering what students can do on their own. Teachers who use it regularly gain real time, and students gain speed and support, as long as the work of thinking stays with them.



