95% of UK undergraduates now use AI in at least one way, up from 66% in 2024 and 92% in 2025, with 94% using it to support assessed work. Students arriving since 2022 have written their whole degree alongside it.
Stephenson & Armstrong 2026; Freeman 2025, cited in Hossain et al., 2026
Corroborated86% of 3,800 university students surveyed across 16 countries already used AI for their studies in 2024.
Digital Education Council survey, cited in Rotter et al., 2026
57% of the world's 50 top-ranked universities had issued guidance on generative AI, mostly telling students to acknowledge and cite it rather than how to use it well.
Moorhouse et al., cited in Micallef et al., 2026
Fourteen AI-text detectors tested scored below 80% accuracy overall; about a fifth of AI-written passages passed as human, and about half escaped detection once edited.
Weber-Wulff et al. 2023, cited in Topaz et al., 2026
Faster work, tutoring that reaches further, and capability that now fits in a pocket.
In a staggered rollout to 5,179 customer-support agents, an AI assistant raised issues resolved per hour by 14% on average, with the largest gains for newer, less experienced workers.
Brynjolfsson, Li & Raymond (NBER), 2023
Peer-reviewedRori, an AI maths tutor on WhatsApp used twice a week for 8 months by 1,000 secondary students, raised test scores by 0.36 standard deviations alongside normal lessons.
Rori evaluation, cited in Boateng et al., 2026
Single-sourceAcross AP Physics exams, frontier models' mean scores sat in a narrow band from 86.6% (ChatGPT) to 91.7% (Claude): exam-level help is now available to any student with a phone.
Paul et al., 2026
Single-sourceKwame for Science, a free AI question-answering tool for West African secondary students, was rated 87.2% helpful across 1,500 questions from 750 users over eight months.
Kwame deployment data, cited in Boateng et al., 2026
Company-reportedAccording to a 2023 World Economic Forum survey, over 60% of educators in advanced economies now use some form of AI in their classrooms.
WEF 2023, cited in arXiv review of AI ethics in higher education, 2024
Single-sourceIn a survey of European schools, 66% of students said access to AI is significant for their success at school; 48% named ChatGPT as their main tool.
Vodafone Stiftung 2025, cited in arXiv review of teacher AI competencies, 2025
Company-reportedIn an OECD survey, 80% of workers who use AI on the job reported higher performance; 57% of finance and 63% of manufacturing employers said it raised productivity.
OECD 2023, cited in Brynjolfsson, Li & Raymond (NBER), 2023
Corroborated74% of European students surveyed believe AI will play an important role in their future careers.
Vodafone Stiftung 2025, cited in arXiv review of teacher AI competencies, 2025
Company-reportedNo verified figure exists for how much of the earnings premium comes from schooling itself rather than from who gets schooled. The estimates we found rest on instruments their own authors call weak.
We could not find a defensible global figure for adult retraining completion. Reporting is voluntary, definitions differ by provider and no cross-national register exists.
Detectors that fail unevenly, writing habits in flux, and institutions a step behind their students.
How often detectors caught AI text depended heavily on which tool wrote it: 76.9% for Bard, 23.9% for GPT-4 and 17.7% for Claude 2.
Perkins et al., 2024
Single-sourceAcross 805 samples, detectors' mean accuracy was only 39.5% on unedited AI text and fell to 17.4% when simple evasion techniques were applied.
Perkins et al., 2024
Single-sourceMost reported student use of ChatGPT is writing work, at a time when students give academic writing about eight hours a week, and half say over-reliance hurts them.
Mogavi et al. 2024, cited in Hossain et al., 2026; Digital Education Council, cited in Rotter et al., 2026
Single-sourceAsked how an AI tutor should help, 52% of physics students wanted a mix, 39.6% wanted direct answers and only 7.6% wanted guided questions.
Sirnoorkar et al., 2026
Single-sourceOnly 35% of students said their institution actively supports developing their AI skills, though 68% called those skills crucial.
Gabelmann et al., 2026
Single-sourceNo verified statistic exists for the effect of generative AI on measured attainment at scale. The figures in circulation are vendor-reported, on cohorts under a year old, without control groups.
We could not establish how many teachers have had paid training time on AI in the last twelve months. Programmes are reported; hours per teacher are not, on any comparable basis.
False accusations, missing teachers, untrained staff and productivity that only feels faster.
Seven AI-text detectors flagged more than half of 91 TOEFL essays by non-native English speakers as AI-written, a 61.3% average false-positive rate, while scoring native eighth-graders' essays nearly perfectly.
Liang et al. 2023, cited in Topaz et al., 2026
Peer-reviewedAn additional 15 million teachers must be recruited by 2030 to achieve universal primary and secondary education: the gap AI tutors are increasingly asked to fill.
UNESCO, cited in Boateng et al., 2026
Corroborated68% of students surveyed believe whether a teacher is competent with AI depends on chance rather than systematic preparation.
Vodafone Stiftung 2025, cited in arXiv review of teacher AI competencies, 2025
Company-reportedIn a 2025 randomised trial, experienced developers were about 19% slower on mature codebases with AI assistance, yet believed it had sped them up.
METR 2025, cited in Alenezi, 2026
Single-sourceAs of 2022 only 69% of primary teachers in sub-Saharan Africa met the minimum required qualifications, down from 75% in 2010.
UNESCO, cited in Boateng et al., 2026
Corroborated63% of organisations expect AI to increase their overall productivity within three years. A curriculum built around today's tools is already behind the workplaces it prepares students for.
McKinsey 2023, cited in Academy for Educational Studies, 2025
CorroboratedOne large-university study found the share of A grades rose by about 13 percentage points in the courses most susceptible to generative AI, with no change in C or D/F grades.
Chirikov 2026, cited in Dumlao et al., 2026
Single-source75% of students surveyed believe using ChatGPT to cheat is wrong but still do it, and nearly 30% think their professors are unaware.
arXiv AI education policy framework, 2023
Company-reportedSelf-report surveys found 54.7% of students admitted cheating in online exams during the pandemic, up from 29.9% before it, likely an undercount.
Newton & Essex 2024, cited in Topaz et al., 2026
Corroborated97.1% of graduate medical trainees report no formal AI instruction, yet 85.5% already use AI tools for clinical decisions, writing and research.
Benton et al., cited in Tsim et al., 2024
Single-sourceScores that swing with the exam, detectors that barely beat chance and policies that skip the main risk.
The same model scored anywhere from 71.9% (2017 exam) to 100% (2021 exam), a 28-point swing that says more about the exam than about what the model knows.
Paul et al., 2026
Single-sourceSeven popular detectors were tested on unedited AI-written text. Four did better than 50%; none reached 65%. The mean was 39.5%.
Perkins et al., 2024
Single-sourceIn a corpus of 91 university policies on generative AI, total prohibition appeared in 14, but accuracy checking and hallucinations in only 3%.
Díaz et al., 2026
Single-sourceAsked how a large language model works, only 33.33% of students identified next-token prediction; most chose “LLMs summarize the web”.
Johri et al., 2026
Single-sourceBenchmarks flatter AI too: on offensive cyber tasks, 21 of 22 models cheated, and an average 41.5% pass rate fell to 26.1% once passes showing any sign of cheating were removed.
Kouremetis et al., 2026
Single-sourceBoth sides of this page are sourced to the same standard: a figure, a unit, a citation and a confidence label. We have not weighted them for you, and where the research could not answer, the card says so rather than filling the gap.
Many figures here are cited second-hand, as their source states them: survey results from Vodafone Stiftung, the OECD, UNESCO, the WEF, McKinsey and the Digital Education Council, and studies by Liang, Weber-Wulff, Newton & Essex, Chirikov, Mogavi and Benton, each via the paper named on the card.
