Student AI Usage Statistics 2026
By Axis Intelligence Research
Co-author: Sarah Mitchell | Last updated: September 11, 2026 | License: CC BY 4.0
Seventy percent of U.S. teenagers use AI for schoolwork, and 95% of UK undergraduates use it in at least one way. Only 27% of those teens say a teacher has ever shown them how to use it safely. Axis Intelligence Research tracks that distance as the Student AI Guidance Index — currently 43 points in U.S. high schools and 57 points in UK higher education.
Quick Answer
How many students use AI in 2026? In the United States, 70% of teens use AI for schoolwork (Common Sense Media, August 2026). In the United Kingdom, 95% of full-time undergraduates use generative AI in at least one way and 94% use it for assessed work (HEPI/Kortext, March 2026). Axis Intelligence Research finds the gap between student adoption and institutional guidance — the Student AI Guidance Index (SAGI) — stands at 43 points in U.S. high schools and 57 points in UK universities as of September 11, 2026.
Key Findings
- Axis Intelligence Research finds the Student AI Guidance Index (SAGI) reads 43 points for U.S. high school students as of September 2026 — 70% use AI for schoolwork while 27% have been taught by a teacher how to use it safely.
- Axis Intelligence Research finds the UK undergraduate SAGI reading is 57 points in 2026, identical to its 2024 level, because institutional AI provision and student adoption each rose by exactly 29 percentage points over two years.
- Axis Intelligence Research estimates that roughly 1.84 billion tutoring-or-teaching requests reached ChatGPT every week as of July 2025, derived from OpenAI’s reported 18 billion weekly messages and the 10.2% tutoring share measured in NBER Working Paper 34255.
- Axis Intelligence Research calculates that 44% of all U.S. teens now obtain a schoolwork answer from an AI tool, combining Common Sense Media’s 70% adoption rate with the 63% of teen AI users who report taking an answer from the tool.
- In Anthropic’s analysis of 574,740 student conversations, Claude performed higher-order cognitive work — Creating (39.8%) and Analyzing (30.2%) — at roughly three times the rate of lower-order work (Applying, Understanding and Remembering combined: 22.7%).
Source : axis-intelligence.com/student-ai-usage-statistics/ — CC BY 4.0
How Many Students Use AI for Schoolwork in 2026?
Adoption is no longer the interesting number. It is close to saturation in every population that has been measured with a probability sample, and the remaining variance is mostly a question of how the survey worded “use.”
U.S. high school students
Common Sense Media surveyed more than 1,000 U.S. teens for Teens in the AI Era: Schoolwork and Skills That Matter, published August 18, 2026. Seventy percent report using AI for schoolwork. Among those who use it, 63% say they get an answer to their schoolwork from the tool: 25% use it as-is, 31% rewrite it to sound like themselves, and 35% improve it with what they already know.
Pew Research Center’s How Teens Use and View AI, published February 24, 2026 and fielded September 25–October 9, 2025, puts the schoolwork figure at 54% of all teens ages 13 to 17. The two numbers are not in conflict — Pew’s fieldwork predates Common Sense Media’s by roughly ten months, and Pew asked about chatbots specifically while Common Sense asked about AI generally.
Pew’s intensity breakdown is the one worth quoting. Ten percent of teens do all or most of their schoolwork with a chatbot’s help. Twenty-one percent do some, 23% do a little, and 45% have not used chatbots this way at all.
UK undergraduates
The HEPI/Kortext Student Generative AI Survey 2026, based on 1,054 full-time UK undergraduates surveyed by Savanta in December 2025, is the cleanest trend series in this field because it has now run three consecutive years with comparable wording. Use of AI in at least one way went from 66% in 2024 to 92% in 2025 to 95% in 2026. Ninety-four percent use generative AI to help with assessed work.
Latin America
The Digital Education Council’s AI in Higher Education Latin America Survey 2026 reports 92% student adoption across the region, against 86% in the Council’s 2024 global survey. Sixty-seven percent of LATAM students use AI daily or weekly, and 88% of frequent users name ChatGPT as their tool.
| Population | AI use for study | Sample | Fieldwork | Source |
|---|---|---|---|---|
| U.S. teens (13–17) | 70% | 1,000+ | 2026 | Common Sense Media, Teens in the AI Era, Aug 2026 |
| U.S. teens (13–17) | 54% | 1,458 | Sep–Oct 2025 | Pew Research Center, Feb 2026 |
| UK undergraduates | 95% | 1,054 | Dec 2025 | HEPI/Kortext Report 199, Mar 2026 |
| UK undergraduates (assessed work) | 94% | 1,054 | Dec 2025 | HEPI/Kortext Report 199, Mar 2026 |
| LATAM higher education | 92% | Not disclosed | 2025–26 | Digital Education Council, Mar 2026 |
Sarah Mitchell, AI/ML: Every one of these surveys is now measuring the same thing from a different angle and landing between 70 and 95. That convergence means the adoption question is answered and the interesting measurement has moved downstream — to intensity, to what task is being offloaded, and to whether anyone taught the student how to do it. A headline that says “X% of students use AI” in 2027 will be reporting the weather.
What Is the Student AI Guidance Index (SAGI)?
SAGI — the Student AI Guidance Index — is an Axis Intelligence Research metric that measures the distance, in percentage points, between how many students use AI for academic work and how many receive structured institutional support for doing so.
How SAGI is calculated
SAGI = (adoption rate) − (institutional guidance rate)
Two components, each weighted 1.0, expressed on a 0–100 point scale. A reading of 0 would mean every student who uses AI academically receives institutional guidance. A reading of 60 means six students in ten are operating without it. The adoption component is the share of students in a population who report using AI for academic work. The guidance component is the share reporting the most concrete institutional support item available in the same survey instrument — a teacher-led safety conversation for U.S. high schools, institutional tool provision for UK universities.
Because the guidance question differs between the two populations, the readings are reported side by side and are not averaged into a single global number. Merging a “did a teacher talk to you” item with an “does your university provide tools” item would produce a figure that means nothing in either country.
Current readings
| Population | Adoption | Guidance component | SAGI | As of |
|---|---|---|---|---|
| U.S. high school | 70% (uses AI for schoolwork) | 27% (teacher taught safe use) | 43 | Aug 2026 |
| UK undergraduate | 95% (uses AI at all) | 38% (institution provides tools) | 57 | Mar 2026 |
| UK undergraduate | 92% | 23% | 57 → 69 | Mar 2025 |
| UK undergraduate | 66% | 9% | 57 | Mar 2024 |
Source for all inputs: Common Sense Media (2026); HEPI/Kortext Reports (2024, 2025, 2026). Full derivation in the accompanying CSV.
The finding institutions will not enjoy
The UK series produces the result that makes SAGI worth publishing. Between 2024 and 2026, undergraduate AI adoption rose 29 percentage points, from 66% to 95%. Over the same two years, institutional AI tool provision also rose 29 percentage points, from 9% to 38%.
Universities have been running hard. They have not gained a metre. The 2026 SAGI reading of 57 is the same reading recorded in 2024, having peaked at 69 in 2025 when adoption jumped and provision lagged a year behind it. Axis Intelligence Research reads the 2025 spike as the catch-up year and the 2026 return to 57 as the steady state: provision is now tracking adoption at parity rather than closing on it.
That distinction matters for anyone writing an AI strategy. A gap that is closing needs patience. A gap that is moving in parallel needs a different instrument.
What Do Students Actually Use AI For?
The task mix among teens
Pew’s data separates the tasks. About four in ten or more teens have used chatbots to research a topic or to solve math problems. Thirty-five percent have used one to edit something they wrote — the lowest of the three, which cuts against the assumption that essay generation dominates.
Outside schoolwork, 57% of teens use chatbots to search for information and 47% for fun or entertainment. Sixteen percent report casual conversation and 12% report seeking emotional support or advice.
What university platform data shows
Survey data reports what students say. Anthropic’s Education Report, published April 8, 2025, reports what they did — 574,740 academic conversations filtered from approximately one million on accounts tied to higher-education email domains.
| Student task category | Share of conversations |
|---|---|
| Create and improve educational content | 39.3% |
| Technical explanations or solutions for assignments | 33.5% |
| Analyze and visualize data | 11.0% |
| Support research design and tool development | 6.5% |
| Create technical diagrams | 3.2% |
| Translate or proofread content | 2.4% |
Source: Anthropic Education Report, April 2025.
Anthropic also classified interaction style. Nearly half of student conversations — around 47% — were Direct, meaning the student sought an answer or an output with minimal back-and-forth. The remainder were Collaborative.
The Bloom’s inversion
Anthropic mapped what the model was doing against Bloom’s revised taxonomy. Creating accounted for 39.8% of operations and Analyzing for 30.2%. Applying came to 10.9%, Understanding 10.0% and Remembering 1.8%.
Axis Intelligence Research calculates the ratio of higher-order to lower-order cognitive work in that distribution at 3.08 to 1 (70.0% versus 22.7%). The taxonomy is normally drawn as a pyramid with remembering at the base. In student AI use, it stands on its point.
Sarah Mitchell, AI/ML: The reflexive read is that students are outsourcing the hard thinking, and Anthropic says as much. The read worth having is narrower. Bloom’s was written to describe what a learner does, not what a tool does on a learner’s behalf, and a student who asks for an analysis and then argues with it is doing something a conversation log cannot distinguish from a student who pastes it. That is precisely why the 47% Direct share is the load-bearing number in that report and the Bloom’s chart is the one that travels.
How Many Students Submit AI-Generated Work?
This is the number every academic integrity office wants and almost nobody measures with a probability sample. HEPI does.
Twelve percent of UK undergraduates report directly including AI-generated text in assessed work in 2026. That figure was 8% in 2025 and 3% in 2024 — a fourfold rise in two years, against an adoption rate that only rose by 29 points over the same period.
Read the two series together and the story is not that more students use AI. It is that the share crossing from assistance into substitution is compounding from a small base while assistance itself has plateaued.
Teens see the same behaviour in their peers. Pew found that 59% of U.S. teens think students at their school use AI chatbots to cheat at least somewhat often, including about a third who say it happens extremely or very often. Among teens who have used chatbots for schoolwork themselves, that figure rises to 76%.
Are Students Being Taught How to Use AI?
Twenty-seven percent of U.S. teens say a teacher has ever talked with them about how to use AI safely. Seventy-three percent say no teacher has.
In UK higher education, 68% of undergraduates believe AI skills are essential to thrive, while 48% feel their teaching staff are helping them develop those skills. Thirty-eight percent say their institution provides AI tools, up from 23% and 9% in the two prior years. On encouragement, the sector is split down the middle: 37% of students agree their institution encourages AI use and 36% disagree.
Bans and what happens next
Common Sense Media measured the enforcement side, and the result is the single most quotable finding in the 2026 literature. Among teens who use AI for schoolwork, 44% have had an AI tool blocked on a school network or device. Fifty-nine percent of those students simply switched to a personal device to get around the block.
Axis Intelligence Research notes that this makes network-level blocking a measurement problem as well as a policy one: a district that blocks AI tools does not reduce student AI use so much as it moves that use off instrumented infrastructure, where nobody can see it.
How Large Is Student AI Use in Absolute Terms?
Percentages describe populations. Volume describes load, and load is what platform operators and ministries actually budget against.
Axis Intelligence Research estimates approximately 1.84 billion tutoring-or-teaching requests reached ChatGPT per week as of July 2025.
Formula and inputs, both from How People Use ChatGPT (Chatterji, Cunningham, Deming, Hitzig, Ong, Shan and Wadman, NBER Working Paper 34255, September 15, 2025):
18,000,000,000 weekly messages × 10.2% classified as Tutoring or Teaching = 1,836,000,000 weekly messages
This is an estimate, not an observation, and it carries two conditions worth stating plainly. The paper’s classified sample excludes users who self-report as under 18, so school-age use sits outside it. And “tutoring or teaching” includes adult self-directed learning, not only coursework.
Two adjacent figures from the same paper give it shape. Tutoring or teaching accounts for 36% of the entire Practical Guidance category, the largest topic in ChatGPT. And among users who self-report an age, roughly 46% of all messages come from people aged 18 to 25 — the university band.
For comparison, computer programming accounts for 4.2% of ChatGPT messages and mathematical calculation 3.0%. Education is, by message volume, more than twice the size of coding on the consumer product.
Which Subjects Use AI the Most?
Anthropic compared its conversation mix against U.S. bachelor’s degrees awarded, using NCES data as the denominator. The distortion is severe and consistent.
| Field | Share of student conversations | Share of U.S. bachelor’s degrees | Axis-calculated ratio |
|---|---|---|---|
| Computer Science | 38.6% | 5.4% | 7.15× |
| Natural Sciences & Mathematics | 15.2% | 9.2% | 1.65× |
| Business | 8.9% | 18.6% | 0.48× |
| Humanities | 6.4% | 12.5% | 0.51× |
| Health Professions | 5.5% | 13.1% | 0.42× |
Source: Anthropic Education Report, April 2025 (shares); ratio calculated by Axis Intelligence Research.
Computer science students appear in the data at just over seven times their share of degrees awarded. Business, health and humanities students appear at roughly half theirs. Anthropic attributes this partly to Claude’s coding strength and partly to awareness in CS communities, and both explanations are plausible — which is why the ratio is better read as a platform-specific adoption signal than as a general statement about who studies with AI.
What Do Students Think AI Is Doing to Their Learning?
Students are markedly less triumphant about AI than the adoption numbers suggest, and markedly more specific about what worries them.
Among U.S. teen AI users, 66% say AI helps them understand schoolwork rather than just finish it faster. In the same survey, 38% say having AI leads them to generate fewer of their own ideas and 39% say they feel they are missing out on learning when they use it to complete assignments.
Asked which skills will matter most, teens named working well with others (55%), fact-checking and critical thinking (50%), and reading and understanding complex information (48%).
UK undergraduates split similarly. Forty-nine percent say AI has improved their student experience, citing time saved and better understanding. Sixty-five percent say assessment has changed significantly in response to AI. On loneliness, the sector is at a dead heat: 20% say AI makes them feel lonelier and 21% say less lonely. Around 15% use AI for companionship, advice, or to address loneliness.
In LATAM, 65% of students are concerned that AI may make learning too shallow and discourage critical thinking, 56% are concerned about data privacy, and 56% doubt the fairness of AI-based assessment — while 73% expect to use AI in their future jobs.
On AI’s wider impact, Pew found 36% of U.S. teens expect a positive personal effect over the next 20 years against 15% expecting a negative one. Framed around society rather than themselves, positive falls to 31% and negative rises to 26%.
Methodology
Collection. Every figure in this page was retrieved from its primary publisher on September 11, 2026. No figure is reproduced from an aggregator, a statistics roundup, or a secondary news summary. Where a report exists as both a press summary and a full document, the figure was taken from the publisher’s own page or PDF.
Sources used. Common Sense Media, Teens in the AI Era: Schoolwork and Skills That Matter (August 18, 2026; nationally representative, 1,000+ U.S. teens). Pew Research Center, How Teens Use and View AI (February 24, 2026; 1,458 U.S. teens and their parents, fielded September 25–October 9, 2025, margin of error ±3.3 points, Ipsos KnowledgePanel). HEPI/Kortext, Student Generative AI Survey 2026, Report 199 (March 12, 2026; 1,054 full-time UK undergraduates, fielded by Savanta in December 2025). Anthropic, Education Report: How University Students Use Claude (April 8, 2025; 574,740 academic conversations). Chatterji et al., How People Use ChatGPT, NBER Working Paper 34255 (September 15, 2025). Digital Education Council, AI in Higher Education Latin America Survey 2026 (March 13, 2026).
Axis-calculated figures and their formulas.
- SAGI = adoption rate − institutional guidance rate, each component weighted 1.0, expressed in percentage points. U.S. high school: 70 − 27 = 43. UK undergraduate 2026: 95 − 38 = 57. UK 2025: 92 − 23 = 69. UK 2024: 66 − 9 = 57.
- Weekly tutoring message estimate = 18,000,000,000 × 0.102 = 1,836,000,000 (as of July 2025).
- Share of all U.S. teens obtaining an AI answer = 0.70 × 0.63 = 0.441, reported as 44%.
- Bloom’s inversion ratio = (39.8 + 30.2) ÷ (10.9 + 10.0 + 1.8) = 70.0 ÷ 22.7 = 3.08.
- Subject over/under-representation ratio = conversation share ÷ degree share.
Scope. SAGI readings are computed only where a single survey instrument supplies both an adoption item and a guidance item for the same population. Readings for different populations are reported separately and never averaged, because the guidance questions are not equivalent instruments. Where survey fieldwork dates differ materially from publication dates, the fieldwork date is the one recorded in the dataset.
About This Dataset
The full dataset behind this page ships as a CSV with one row per observation and complete provenance columns: source organisation, source document, source URL, retrieval date, primary-source flag, and an Axis-calculated flag with a method note for every derived figure. Every number in the prose and every chart datapoint exists as a row.
Licence: Creative Commons Attribution 4.0 International (CC BY 4.0). Free to reuse, including commercially, with attribution.
Attribution line: Axis Intelligence Research, Student AI Usage Statistics 2026, 2026.
Cite This Research
APA Axis Intelligence Research. (2026). Student AI usage statistics 2026: Adoption, homework, study habits and the guidance gap. Axis Intelligence. https://axis-intelligence.com/student-ai-usage-statistics/
MLA Axis Intelligence Research. “Student AI Usage Statistics 2026: Adoption, Homework, Study Habits and the Guidance Gap.” Axis Intelligence, 2026, axis-intelligence.com/student-ai-usage-statistics/.
Chicago Axis Intelligence Research. “Student AI Usage Statistics 2026: Adoption, Homework, Study Habits and the Guidance Gap.” Axis Intelligence, 2026. https://axis-intelligence.com/student-ai-usage-statistics/.
Frequently Asked Questions
Why do student AI usage surveys report such different numbers, and which one should I cite?
Because they measure different populations with different wording at different times. Pew’s 54% is U.S. teens asked about chatbots, fielded in late 2025. Common Sense Media’s 70% is U.S. teens asked about AI generally, fielded in 2026. HEPI’s 95% is UK full-time undergraduates asked whether they use AI in any way at all. For U.S. school-age figures, cite Common Sense Media for currency and Pew for intensity breakdowns and margin of error. For higher education trend lines, cite HEPI, because it is the only series with three comparable annual waves.
Does blocking AI tools on a school network reduce student AI use?
The available evidence says it relocates it. Among U.S. teens who use AI for schoolwork, 44% have had a tool blocked on a school network or device, and 59% of those students switched to a personal device. The practical consequence for administrators is that blocked use becomes unmeasurable use — it leaves school infrastructure and therefore leaves every log an institution can inspect.
What share of students actually submit AI-generated text as their own work?
Twelve percent of UK undergraduates reported directly including AI-generated text in assessed work in 2026, against 8% in 2025 and 3% in 2024. This is the most methodologically solid substitution figure currently available, because it comes from a repeated probability sample with consistent wording. Detection-tool statistics are not a substitute for it and measure something different.
Which academic subjects show the heaviest AI use?
On Anthropic’s platform data, computer science students accounted for 38.6% of student conversations while representing 5.4% of U.S. bachelor’s degrees — an over-representation of just over seven times. Business (0.48×), humanities (0.51×) and health professions (0.42×) all appeared at roughly half their enrolment share. Treat this as platform-specific: Claude skews technical, and a ChatGPT-derived equivalent would likely compress the spread.
Are students using AI to learn, or to avoid learning?
Both, in measurable proportions. Sixty-six percent of U.S. teen AI users say it helps them understand schoolwork, and 63% say they take an answer from it — a quarter of them unchanged. On Anthropic’s data, roughly 47% of student conversations were Direct, meaning minimal engagement with the output. The honest summary is that assistance and substitution coexist inside the same student, often inside the same assignment.
How many students are taught AI literacy at school or university?
Twenty-seven percent of U.S. teens say a teacher has ever talked with them about using AI safely. In the UK, 48% of undergraduates feel teaching staff are helping them develop AI skills, while 68% consider those skills essential. This gap between perceived necessity and delivered instruction is what the Student AI Guidance Index measures.
What is the Student AI Guidance Index (SAGI) and how do I use it?
SAGI is an Axis Intelligence Research metric equal to a population’s AI adoption rate minus its institutional guidance rate, in percentage points. It currently reads 43 for U.S. high schools and 57 for UK undergraduates. Institutions can compute their own reading from any two internal survey items measuring the same pair, and compare against the national benchmarks published here. The formula is deliberately simple so that a reading is reproducible without access to our data.
Is AI use higher among older or younger students?
Higher among older ones, on every dataset with an age dimension. Common Sense Media’s 2026 census found AI use rising from 81% among 9- to 12-year-olds to 89% among 13- to 15-year-olds and 92% among 16- to 17-year-olds. In the NBER analysis of ChatGPT, roughly 46% of all messages from users who self-report an age came from the 18-to-25 band.
How many AI tutoring requests are sent worldwide each week?
Axis Intelligence Research estimates approximately 1.84 billion per week on ChatGPT alone, as of July 2025, calculated from OpenAI’s reported 18 billion weekly messages and the 10.2% tutoring-or-teaching classification share in NBER Working Paper 34255. This excludes users self-reporting as under 18 and covers consumer plans only, so it understates total student volume.