AI in Education: Uses, Benefits, Risks, and Real-World Examples

Quick Answer: What Is AI in Education?

AI in education means using artificial intelligence tools to support teaching, learning and school administration. The most common uses are tutoring chatbots that answer questions, tools that help teachers plan lessons and mark work, adaptive practice that adjusts to a student's level, and systems that handle admissions or flag suspected cheating.

The evidence so far is mixed, and the design of the tool seems to matter more than whether it uses AI:

• Benefits: well-designed AI tutors that guide students instead of giving answers have shown learning gains in published trials, and teacher-facing tools can cut preparation time.
• Risks: a tutor that simply hands over answers can leave students no better off, or worse off, once the AI is taken away. Other risks include privacy, bias, unreliable AI-detection tools, over-reliance and unequal access.
• Governance: UNESCO has urged governments to regulate generative AI in schools, and the EU AI Act classes several education uses as high risk.

This guide explains the main uses, the benefits and risks with the research behind them, real-world examples drawn from published studies and vendor pages, and a practical checklist for schools, teachers, parents and students. Figures from papers that could not be opened in full are labeled as reported.

Introduction

Generative AI reached classrooms quickly. A Pew Research Center survey of 1,391 US teens, conducted from September 18 to October 10, 2024, found that 26% had used ChatGPT for schoolwork, double the 13% recorded in 2023 (Pew Research Center). Students are not the only ones adopting it. Teachers use AI to draft lesson plans and quizzes, and schools and exam bodies are deciding where AI belongs and where it does not.

Policy has not kept pace. UNESCO reported, citing a global survey of more than 450 schools and universities, that fewer than 10% had institutional policies or formal guidance on generative AI (UNESCO). That gap is why a clear picture of what AI can and cannot do for learning is useful before deciding how to use it.

This article covers AI in education in practical terms: what it is, how it is used, what the research says about benefits and risks, examples of real deployments and studies, and how to use it responsibly.

What Does AI in Education Mean?

"AI in education" covers several different kinds of tools, and mixing them up causes much of the confusion in the debate.

• Generative AI chatbots such as ChatGPT can explain topics, answer questions, draft text and generate practice problems.
• AI tutors are built for teaching. They use a generative model but are configured to guide the student, for example by asking questions instead of giving the final answer.
• Adaptive learning systems adjust the difficulty or sequence of material based on a student's responses.
• Teacher assistants help with planning, rubrics, quizzes, feedback and summaries.
• Administrative and decision systems support admissions, grading and test monitoring.
• Detection tools try to identify AI-written text. Their reliability is limited, as discussed below.

The distinction matters because the benefits and risks differ by category. A tutor designed to guide reasoning is a different thing from an open chatbot a student can ask to write an essay.

How AI Is Used in Education

1. Personalized tutoring and practice

An AI tutor can respond to each student's questions at any time and adjust to their level. UNESCO's own page on AI in education describes opportunities such as "personalized tutoring and smarter lesson planning" while also naming challenges: "over-reliance, ethical concerns, and declining deep reading skills" (UNESCO).

2. Lesson planning and teacher workload

Khan Academy says its Khanmigo assistant can help teachers with lesson planning, rubric generation, quiz creation and student progress summaries, and quotes a teacher saying a task that normally took about an hour was completed in no more than 15 minutes using the rubric generator (Khanmigo). That is a vendor-reported testimonial, not an independent measurement.

3. Feedback on writing and practice

AI can give quick comments on drafts, explain errors in a worked solution, or generate extra practice questions. The risk is that feedback tools can also write the work, so the design and the classroom rules around them matter.

4. Language learning and accessibility

Chat-based practice can give learners a low-pressure place to practice a language, and AI can support accessibility through summarization, simplification and translation. Quality varies by language and tool, so teachers should check outputs for accuracy.

5. Administration, admissions and assessment

Schools and universities also use or consider AI to evaluate learning outcomes, assign students to institutions or monitor tests. These uses carry the most regulatory attention, because errors can affect a person's opportunities. The EU AI Act's Annex III lists these as high-risk uses: systems used to determine access or admission, to evaluate learning outcomes, to assess the appropriate level of education a person will access, and to monitor and detect prohibited behavior during tests (EU AI Act, Annex III).

6. Detecting AI-written work

Many schools have turned to AI detectors. The research on their accuracy is cautionary, and we cover how they work and where they fail in how AI detectors work and which tools perform best in best AI detectors.

The Benefits of AI in Education

Learning gains from well-designed tutors

The strongest case for AI in education comes from studies of tutors designed around how people learn, not from chatbots in general.

Nigeria (World Bank). A randomized controlled trial in nine public secondary schools in Edo State tested a six-week after-school program in which students worked in pairs with GPT-4 under teacher supervision, using prompts designed to promote reasoning instead of shortcuts. As reported for the World Bank paper, the effect on English, the main outcome, was 0.23 standard deviations, with a larger 0.31 standard deviation effect on a broader assessment, and the authors estimated gains equivalent to 1.5 to 2 years of business-as-usual schooling (World Bank). The World Bank page was not directly readable when this article was checked, so confirm the numbers at the source before quoting them.

Harvard physics course. A randomized crossover trial published in Scientific Reports in June 2025 compared a custom AI tutor with in-class active learning for 194 undergraduate physics students, with each student serving as their own control across two weeks. As reported in summaries of the paper, students using the AI tutor achieved larger learning gains in less time, with median gains more than doubling, in about 49 minutes instead of 60 (Scientific Reports). The journal page was not directly readable, so treat these as reported figures. Both results come from specific settings and designs, so they do not show that any AI tool will produce similar gains.

Time savings for teachers

Teacher-facing tools aim to reduce paperwork. Vendors report time savings, as in the Khanmigo example above, but this article cites no independent measurement of workload effects, so treat such claims as promising and unproven at scale.

Availability and patience

An AI tutor is available outside school hours and does not tire of repeated questions. Khan Academy describes Khanmigo as "limitless patience" and says it is designed to guide learners to find the answer themselves (Khanmigo). Whether this helps depends on whether the tool really withholds answers and whether students use it that way.

Wider access, in principle

AI tutoring could extend help to learners who cannot afford private tutors. That promise depends on devices, connectivity, language support and cost. UNESCO emphasizes a human-centered approach to ensure AI "does not widen the technological divides within and between countries" (UNESCO).

The Risks of AI in Education

1. Unguided AI can harm learning

The most important cautionary study is a randomized controlled trial at a large high school in Turkey, published in PNAS in 2025 under the title "Generative AI without guardrails can harm learning." As reported in summaries of the paper, students given a standard GPT interface performed about 48% better on assisted practice sessions than the control group, and students given a version designed to promote reasoning performed about 127% better. But when students' unassisted exam results were compared, the unguarded version harmed learning, while the guardrailed tutor protected performance (PNAS). The exact exam figures could not be verified from the paper, so this article does not quote them.

The lesson is that doing well with AI help is not the same as learning. A tool that makes homework easier may make exams harder.

2. Over-reliance and skill erosion

UNESCO names over-reliance and declining deep reading skills among the challenges (UNESCO). Pew's survey shows teens themselves draw lines: 54% said using ChatGPT for researching new topics was acceptable, while only 18% said using it to write essays was (Pew Research Center).

3. Accuracy

Generative models can state wrong information confidently. A student without the background to spot an error may learn it. Teachers and students should check important claims against textbooks and primary sources.

4. Unreliable AI detection

Treating a detector score as proof of cheating can harm students. The research on detectors, including false positives, is covered in our articles on how detectors work and the best AI detectors. A detector should at most start a conversation, and never be the sole evidence for a penalty.

5. Privacy and data protection

Student data, including writing, questions and sometimes voice or images, may be processed by third parties. Schools should check what data a tool collects, where it is stored, whether it is used to train models and how it can be deleted. UNESCO's guidance calls on governments to protect data privacy (UNESCO).

6. Bias and fairness in high-stakes decisions

AI systems trained on historical data can reproduce its biases. That is part of why the EU AI Act treats admissions, learning evaluation, level assessment and test monitoring as high-risk uses (EU AI Act, Annex III).

7. Unequal access

Students without reliable devices, internet or paid AI subscriptions may fall behind those who have them. Free tiers and school-provided access can reduce but not remove the gap.

8. Environmental cost

AI uses energy and water. The footprint of a single question is small, but the total grows with scale. See our explainer on how much water AI uses for the numbers and why they vary.

Real-World Examples

These examples come from published studies and vendor pages, not from any first-hand testing.

What the examples have in common

The positive results involve tools designed to promote reasoning, with teacher supervision or structured sessions. The cautionary result involves open access without guardrails. A reasonable reading of the studies above is that this pattern matters more than any single number.

Policy and Regulation

UNESCO

In 2023 UNESCO published guidance on generative AI in education and research. It called on governments to regulate generative AI in schools, recommended an age limit of 13 for use of AI tools in the classroom, called for teacher training and set out seven key steps for governments (UNESCO). It also developed a separate policy-maker guide and competency frameworks for students and teachers.

United States

The US Department of Education's Office of Educational Technology published a report in May 2023 on AI and the future of teaching and learning. Its first recommendation was to emphasize humans in the loop, alongside others such as designing AI using modern learning principles, strengthening trust and developing education-specific guidelines and guardrails (Inside Higher Ed). This summary comes from secondary coverage because the report itself could not be accessed.

European Union

The EU AI Act classes the education uses listed in Annex III as high risk (EU AI Act, Annex III). Compliance dates have been the subject of revisions, so check current guidance from official EU sources if your school or product operates in the EU.

How to Use AI in Education Responsibly

For schools and districts

1. Write a clear policy. Say what is allowed, in which tasks, and how students should disclose AI use.
2. Keep humans in the loop. Decisions about grades, discipline and admissions should not rest on an AI output alone.
3. Review privacy terms. Check data collection, retention, training use and deletion before adopting a tool.
4. Train teachers. UNESCO's guidance calls for teacher training (UNESCO).
5. Pilot and measure. Test a tool with a small group and look at unassisted outcomes, not just satisfaction.

For teachers

1. Prefer tools that guide reasoning. The Turkey and Nigeria examples suggest design matters.
2. Assess understanding without AI at least some of the time. That shows what students can do on their own.
3. Teach students to verify. Ask them to check AI claims against sources.
4. Do not rely on detector scores. Use conversations, drafts and process evidence.
5. Check outputs before sharing. AI-generated materials can contain errors.

For parents

1. Ask what the school allows. Policies vary.
2. Prefer tools that explain, not just answer. Ask your child to explain the solution in their own words.
3. Check age limits and privacy terms. Khanmigo's page, for instance, says parent subscribers must be 18 or older with a US billing address (Khanmigo).

For students

1. Use AI to understand, not to avoid thinking. Try the problem first, then ask for a hint.
2. Follow your school's rules and disclose use when required.
3. Verify facts. Do not copy an AI explanation you cannot check.
4. Protect your data. Avoid sharing personal information with tools.

If you are looking for apps that balance learning with healthy screen habits, see our list of learning and gaming apps for good screen time.

Questions to Ask Before Adopting an AI Tool

A short vendor-evaluation list helps schools avoid choosing on marketing alone.

1. What is the tool designed to do? Does it guide students toward answers or simply provide them? The studies above suggest this design choice matters.
2. What evidence exists? Ask for independent studies, the age group and subject they covered, and whether results were measured with and without AI help.
3. What data does it collect? Find out what is stored, for how long, who can access it and whether it is used to train models.
4. Who is responsible for errors? Know how mistakes are reported and corrected, and what human review exists for any decision that affects a student.
5. What does it cost at scale? Free tiers and trials may end, and per-user pricing grows with enrollment.
6. Does it work for all your learners? Check language support, accessibility features and device requirements.
7. What happens if you stop using it? Confirm that you can export or delete student data.

How to Read Claims About AI and Learning

Headlines about AI in education often compress complicated studies. A few habits make claims easier to judge:

• Check what was measured. Better scores on AI-assisted practice are different from better scores on a later test taken without AI.
• Check the setting. A result from one school, one course or one country may not carry over to others.
• Check who is reporting. Vendor statements and testimonials are not independent evaluations.
• Check the comparison. A tutor compared with no help is a different test from a tutor compared with a good teacher or active-learning class.
• Look for the design details. In the studies discussed above, the guidance built into the tool seems central to the outcome.

Common Mistakes Schools Make with AI

1. Banning or embracing everything at once. Different uses carry different risks.
2. Judging a tool by student satisfaction. Practice scores with AI help can rise while unassisted results fall, as the Turkey study shows.
3. Using detector scores as proof. The tools have error rates, and the consequences for a wrongly accused student are serious.
4. Skipping the privacy review.
5. Ignoring equity. Provide access so benefits do not go only to students who can pay.
6. Treating vendor claims as evidence. Look for independent studies, and note that the positive results cited here come from specific settings.

Frequently Asked Questions

What is AI in education? It is the use of artificial intelligence to support teaching, learning and administration, including AI tutors, lesson-planning tools, adaptive practice, feedback tools and systems used in admissions or test monitoring.

What are the main benefits of AI in education? Published trials of well-designed AI tutors have reported learning gains, such as the Nigeria and Harvard studies described above, and teacher tools may reduce preparation time. These results depend on the design of the tool and the setting.

What are the biggest risks of AI in education? Unguided AI can harm learning, as a 2025 PNAS study of high school math found. Other risks include inaccurate answers, privacy concerns, bias, unreliable AI detectors, over-reliance and unequal access.

Does AI help students learn or just do their homework? Both are possible. The PNAS study found that unguarded access boosted practice performance but harmed unassisted exam results, while a tutor designed to promote reasoning protected learning (PNAS).

Will AI replace teachers? Nothing in the sources cited here argues for replacing teachers. The US Department of Education's report put "humans in the loop" first among its recommendations, and the positive studies involved teacher supervision or structured sessions.

Is it safe for kids to use AI? It depends on the tool, the age and the safeguards. UNESCO recommended an age limit of 13 for classroom use of AI tools (UNESCO). Check each tool's terms, data practices and age requirements.

Can teachers tell if students used AI? Not reliably with software alone. Detectors have error rates, and researchers caution against using them as proof. See our guide on how AI detectors work.

How many students use AI for schoolwork? In a Pew survey of 1,391 US teens, 26% said they had used ChatGPT for schoolwork in 2024, up from 13% in 2023 (Pew Research Center).

What does the law say about AI in schools? It varies by country. The EU AI Act lists admissions, learning evaluation, level assessment and test monitoring as high-risk education uses (EU AI Act, Annex III), and UNESCO has urged governments to regulate generative AI in schools. Check the rules where you live.

How should schools start with AI? Write a policy, review privacy terms, train teachers, pilot a tool with a small group and measure unassisted outcomes before expanding.

Conclusion

AI in education is neither a cure nor a threat in itself. The studies cited here suggest that results depend on design: tutors built to guide reasoning, used with structure and supervision, have shown gains, while open access to answers can improve practice scores without improving what students can do alone.

Schools, teachers and families can act on that. Choose tools that make students think, check unassisted learning, keep humans responsible for high-stakes decisions, protect student data, and treat detectors and vendor claims with caution. As policy catches up, UNESCO's call for regulation, teacher training and a human-centered approach is a reasonable baseline.

Sources

• UNESCO, Artificial intelligence in education and UNESCO: Governments must quickly regulate Generative AI in schools
• Pew Research Center, About a quarter of U.S. teens have used ChatGPT for schoolwork, double the share in 2023 (January 2025)
• Khan Academy, Khanmigo (vendor page; claims are vendor-reported)
• European Union, AI Act, Annex III
• Bastani et al., Generative AI without guardrails can harm learning: Evidence from high school mathematics (PNAS, 2025; figures as reported, paper not directly readable)
• Kestin et al., AI tutoring outperforms in-class active learning (Scientific Reports, June 2025; figures as reported)
• World Bank, From Chalkboards to Chatbots: Evaluating the Impact of Generative AI on Learning Outcomes in Nigeria (figures as reported; page not directly readable)
• Inside Higher Ed, Education Department issues recommendations on AI (May 2023, covering the US Department of Education report)