Table of Contents
- Key Pointers
- The Short Version
- Why the numbers vary so much
- The headline numbers
- How use breaks down by subject
- How use breaks down by age and school type
- What students actually use AI for
- What this means for teachers
- How student behavior is changing
- Where the numbers will likely go from here
- Wrap-up
- FAQs
- Sign Up for Quetext Today!
Key Pointers
- Survey data from 2023–2024 puts the share of college students who have used AI tools for coursework at roughly 50–66%, depending on the study and the school type.
- Among high schoolers, the share is lower but rising fast. Pew Research found about 19% of US teens aged 13–17 who had heard of ChatGPT had used it for schoolwork in late 2023.
- Use varies significantly by subject. Students reach for AI more often in writing-heavy classes (English, history, social sciences) than in math-heavy ones.
- The line between “research aid” and “essay writer” is where most of the educator concern lives.
- Detection technology has caught up partway, but the smart workflow combines AI scanning with human judgment.
The Short Version
Most independent surveys from the last 18 months put student AI use somewhere between 19% and 66%, with college students more likely to use it than high schoolers and writing-heavy assignments more likely to involve it than math-heavy ones. The data is messy because schools, ages, and survey methods vary. What’s consistent is the direction. Use is rising, awareness is rising, and the workflow most educators recommend now is AI detection plus human review rather than either one alone.
Why the numbers vary so much
Here’s the analytical reality: there is no single, universally accepted figure for “how many students use AI to write papers.” The published numbers range from below 20% to above 65%, and the spread isn’t sloppiness. It reflects four real variables.
Who’s being surveyed. A study of MIT undergrads will produce different numbers than a national sample of high school freshmen. Both are valid. They measure different populations.
What “use” means. “Used ChatGPT once for a quick definition” and “submitted a fully AI-generated essay” are both technically use. Many surveys don’t separate them. The ones that do show the gap.
When the survey ran. AI tool use has been climbing month over month since ChatGPT launched in late 2022. A survey from January 2023 captures a different reality than one from late 2024.
Whether students were anonymous. Self-reported use of any potentially-frowned-upon behavior shifts heavily based on whether the respondent thinks they’re identifiable.
The numbers below come from independent, methodologically transparent sources. Where the studies disagree, the disagreement is meaningful, not noise.
The headline numbers
A few survey results that have shaped the public conversation:
College students (US): A BestColleges 2024 survey on college students and AI found 56% of college students reported using AI for assignments or exams. About 22% said they had used it for an entire assignment, while the larger group used it for parts of one.
High school students (US): Pew Research Center’s data on US teens and ChatGPT for schoolwork reported that around 19% of US teens aged 13–17 who had heard of ChatGPT had used it for school assignments by late 2023. That figure has almost certainly risen since.
Higher education broadly: Tyton Partners’ Time for Class 2024 report on AI in higher ed found that 59% of students used generative AI at least monthly, with adoption climbing year over year.
The pattern is consistent: roughly half of college students use AI for coursework in some form, with a smaller but growing share of high schoolers doing the same.
How use breaks down by subject
The aggregate numbers hide an important pattern. Students don’t use AI evenly across their classes.
| Subject area | Reported AI use | Why |
|---|---|---|
| English / writing-intensive | High (~60-70% of users) | AI is strong at producing draft text |
| History / social sciences | High | Long-form essays, research summaries |
| Computer science | Moderate-to-high | Code generation use cases |
| Mathematics | Moderate | Less reliable on multi-step math |
| Physics / chemistry | Lower | Specialized notation, frequent errors |
| Studio arts / performance | Lowest | Form less compatible with AI output |
This pattern matters for teachers planning detection workflows. The classes where AI use is highest are also the classes where detection is the hardest, because human writing and AI writing are most stylistically similar in essay-form work.
The Quetext analysis on students using AI in schools breaks down the subject-by-subject pattern in more depth and is worth a read for educators trying to set realistic expectations by department.
How use breaks down by age and school type
A consistent pattern across surveys:
Younger high school students (13-15): Lower reported use, partly because of less independent assignment work and partly because of school AI access policies.
Older high school students (16-18): Use rises as college applications and AP courses introduce more independent essay assignments.
Undergraduate college: Highest reported use overall, particularly in first- and second-year general education courses.
Graduate students: Use is high but more often described as a “research aid” than an “essay writer.” The professional consequences of getting caught are heavier.
Online students: Higher reported use than in-person students, likely due to lower in-class accountability.
For broader context on AI use beyond education, the rundown on AI usage statistics for 2026 covers how the education numbers compare to enterprise and personal use.
What students actually use AI for
Survey data separates the casual use from the high-risk use:
Lower-risk use cases (most common):
- Generating ideas or outlines
- Defining terms or explaining concepts
- Summarizing long readings
- Improving grammar and clarity of writing they’ve already drafted
- Translating non-English sources
Higher-risk use cases (smaller share):
- Writing full essays
- Generating answers for online assessments
- Producing problem-set solutions to be submitted as original work
- Paraphrasing source material to hide direct citation
The lower-risk uses don’t necessarily violate any policy at most schools. The higher-risk ones do. The challenge for teachers isn’t to eliminate AI from the classroom. It’s to identify which kind of use is in front of them.
For the academic risk side specifically, the breakdown on risks of using AI for assignments covers the consequences students face when their use crosses into the higher-risk territory.
What this means for teachers
The data points to a workflow rather than a verdict. Three observations:
Banning AI doesn’t match the numbers. When the majority of students in a class are using AI in some form, a blanket ban produces a compliance problem teachers can’t actually enforce. Policies that distinguish between research-aid use and authorship use are more workable and more honest about the reality.
Detection alone isn’t enough. Even the strongest AI detectors flag confidently on raw AI output and weakly on paraphrased or hybrid drafts. The smart workflow is AI detection plus a quick read for the patterns AI generates (smooth transitions, generic structure, missing specifics).
The conversation matters more than the catch. Teachers who get the best results in the data report having explicit class-level conversations about acceptable use rather than treating AI as an automated cheating engine. Students respond to clear lines.
Try this: Run a submission through Quetext’s AI Detector and pair the signal with a quick read for the patterns this data describes. Where the detector and the read agree, you have an actionable signal. Where they disagree, you have an opportunity to ask the student about their process. If you’d rather start with a one-paragraph spot check before scanning a full stack, Quetext covers the first 1,000 words at no cost.
How student behavior is changing
Three trends visible in the most recent survey data:
The line between “research” and “writing” is blurring. Students who say they use AI “only for research” often paste AI-generated phrasing directly into their drafts. The intent stayed legitimate. The output drifted.
Free tools are dominant. ChatGPT free tier, Gemini, and Claude free tier account for the bulk of student use. Paid AI tool subscriptions are concentrated in graduate students and STEM majors.
Multi-tool workflows are growing. Students increasingly use one AI tool for ideas, another for paraphrasing, and a third to check whether the output reads as AI. The detection target is no longer a single model.
Awareness of detection has gone up. Most surveyed students know detectors exist and many actively try to circumvent them. The arms race is now part of the workflow on both sides of the assignment.
Where the numbers will likely go from here
Forward-looking analysis from the same survey programs points in one direction. Reported use has climbed every quarter since ChatGPT launched. Each new model release (GPT-4, GPT-4o, Claude 3, Gemini 2) raises the floor of what AI can produce. Each iteration also makes detection harder, which both increases student confidence in getting away with use and reduces the signal-to-noise ratio of single-tool verdicts.
The likely 2026 picture: most college students will use AI in some capacity for most writing-heavy assignments. The classroom question won’t be “is AI being used here.” It will be “what kind of use is acceptable in this class, and how do we structure assignments around that policy.”
Wrap-up
The honest answer to “how many students use AI to write papers” is: more than half of college students, around a fifth of high schoolers as of late 2023 (likely higher now), and use is climbing across every grade level and subject. The variance in the data is real but doesn’t change the direction. AI is part of student writing workflows, and the question for educators is no longer whether but how.
The workflow that works for most teachers: clear policy, AI detection on submissions, and human review on flagged work. The conversation about acceptable use matters more than the catch on any single paper.
Try try Quetext’s free AI scan on a paper before grading it. The first 1,000 words are free, which is enough to test the workflow before committing to a department-wide license.
FAQs
What percentage of students use AI to write papers?
Independent surveys from 2023–2024 put the share of college students using AI for assignments somewhere between 50% and 66%, depending on the study. About 19% of US high schoolers aged 13–17 who had heard of ChatGPT had used it for schoolwork in late 2023, according to Pew Research, and that figure has almost certainly risen since. The range reflects differences in how studies define “use,” which students they sample, and when the survey ran.
- College: roughly 50-66%
- High school: around 19% in 2023, rising
- Numbers depend heavily on survey methodology
Are students using AI mostly for cheating or for legitimate help?
A blend of results exists. The majority of students who say they use AI do so as a research tool such as generating outlines of content, finding explanations of concepts, or correcting grammar. A smaller, but significant portion of the users of AI use it to generate an entire essay; therefore, violating most academic integrity policies at schools. Teachers now face the new challenge of identifying where the research tool ends and the authorship has begun, as many students will not understand that they crossed over until the work does not pass the detector tests.
- Most of the users of AI have used it as a research tool frame.
- Minority of the users of AI have generated an entire essay.
- The location between the two is where most policies will have disputes.
Can teachers tell when a student used AI?
Yes. Quetext and other AI detectors work by categorizing the flagged content based on its statistical similarity to human LLM-created content. Quetext is the only service to do this effectively with ‘raw’ or unedited content created by LLMs. Using paraphrase/hybrid drafts makes detection much harder, and there continues to be an extensive amount of documented false positives for ‘real’ writing in all categories of this type of detection. Therefore, for optimum efficiency using these tools in educational settings, it is essential for educators to use AI detectors in combination with manual reading of LLM’s (LLM) signal patterns (standard forms, absent definitive specifics, exceptionally unordinary seamless transitions) for detection indication.
- LLM’s user unedited outputs for detection effectively.
- By paraphrasing an unedited original LLM-generated draft will not be detected very effectively.
- AI detectors should always be used in conjunction with manual review by a human for best effectiveness.
What subjects see the most AI use?
Classes with a heavy writing component (English, History, Social Science) have the highest amount of reported AI use primarily because writing is the typical format of work assigned to students, therefore lending itself to being able to be produced by AI tools. Classes that focus on Computer Science also use AI a lot because AI can produce and generate code. In comparison to writing-heavy classes, students taking Mathematics and Science classes report lower amounts of AI use because LLMs (Large Language Models) are not as reliable with multi-process methods for producing quantitative results. Although the gap between AI and Mathematics/Science classes is closing, the lowest amount of reported AI use is found in Studio Art and Performance Art classes.
- Classes that have a writing component have a high level of use,
- Classes that have a quantitative component have a lower level of use, however, the difference between the two is decreasing,
- Classes that require the student to perform have the least amount of reported AI use.
