Walk into many classrooms today and you'll find a new presence: artificial intelligence. It tutors students one-on-one, drafts lesson plans for teachers, and marks practice quizzes in seconds. It also writes essays for students who would rather not — and nobody can fully tell when.
AI in education is neither the miracle its boosters claim nor the catastrophe its critics fear. Here is an honest look at what is actually happening in classrooms, what the evidence says, and what good use looks like.
What AI actually does in classrooms today
The most visible change is the AI tutor. Khan Academy's Khanmigo is the best-known example: an AI tutor and teaching assistant built on GPT-4 that guides students through problems Socratically — asking questions rather than handing over answers — and helps teachers with lesson planning and feedback. It is free for teachers, and it has been deployed in schools since the autumn of 2025.
Language learning got there even earlier. Duolingo Max, launched in March 2023, added GPT-4-powered features to the language app: "Explain My Answer" unpacks why an answer was wrong, "Roleplay" lets learners practise real conversations with an AI character, and video-call practice simulates live speaking.
Behind the scenes, teachers use AI for the unglamorous work: drafting quizzes, summarising reading material, generating worksheets for different ability levels, and giving faster feedback on drafts. Accessibility tools — real-time translation, reading aloud, speech-to-text — help students who learn differently or study in a second language.
The pattern: AI handles scale — one-on-one attention for every student, instant feedback, endless patience — while teachers handle judgment.
The personalization promise
The big idea is adaptive learning: software that adjusts to each student. Get questions right and it moves faster; struggle and it slows down, revisits the basics, and tries a different explanation. It remembers what you forgot and resurfaces it just before it fades — the digital version of a tutor who never forgets.
In plain English, it works like this: the system tracks every answer, builds a model of what the student knows, and picks the next question to be just hard enough — challenging but achievable.
Does it work? The honest answer: promising but uneven. Well-designed adaptive systems can help, especially for practising core skills like maths. But personalization is not magic — a confused student can get a personalised path to the wrong understanding, and no algorithm replaces a teacher noticing a child is having a bad day. The technology is a multiplier on good teaching, not a substitute for it.
The cheating debate
Then there is the elephant in the classroom. When ChatGPT arrived in late 2022, it could write a passable essay in seconds — and assessment as schools knew it wobbled. AI use among students is now widespread, and the line between cheating and legitimate help is genuinely blurry.
Schools first reached for detection tools. The best known is Turnitin's AI-writing detector, which claims 98% accuracy with under 1% false positives — but those figures come from controlled tests on pre-2022 human writing. Independent testing tells a more cautious story: false positives of 3–15% in real-world conditions (about 8% on average in one 200-essay test), rising to around 12% for non-native English writers — and detection rates falling to 48–72% once AI text is edited or paraphrased. Some universities are moving away from detectors entirely.
The deeper response is redesigning assessment itself: more in-class work, oral exams, drafts and process marks, and assignments asking for personal reflection that AI cannot fake. The emerging consensus: you cannot detect your way out of this — you have to teach through it.
What teachers and schools report
Teachers are adopting AI fast. A Gallup survey of 2,232 US teachers found 60% had used AI tools in the 2024–25 school year; a survey by the teaching marketplace TPT put the figure at 80% of educators using generative AI. Adoption roughly doubled in a year — from 32% in 2024 to 61% in 2025.
What do they report? Time saved, mostly: regular AI users say the tools cut hours of admin — lesson planning, worksheet generation, feedback drafts. But there is a governance gap: only about 18% of teachers say their school has clear written guidance on AI use. The technology arrived in classrooms years before the rules did.
Risks: bias, over-reliance, privacy
- Bias. AI systems learn from data reflecting human biases, and they can reproduce them — marking some dialects as "wrong," or generating stereotyped examples. In a classroom, that can quietly disadvantage the students most in need of support.
- Over-reliance. The core worry: students outsourcing thinking itself. If AI always supplies the first draft and the answer, students may never build the underlying skills. Calculators didn't end arithmetic — but writing is thinking in a way arithmetic isn't.
- Privacy. Student data is sensitive. When schools feed pupil work into commercial AI systems, questions arise: where does the data go, who can see it, how long is it kept? Districts need clear data agreements — and many are still catching up.
- Equity. AI tutors help most those who already have devices, connectivity, and support at home. Without deliberate effort, AI could widen the gaps it promises to close.
These are not reasons to ban AI — they are reasons to govern it, which is exactly what the AI ethics conversation is about.
What good use looks like
For students: use AI as a tutor, not an answer machine. Ask it to explain, quiz you, and challenge your reasoning — then verify anything important. A good rule: if you couldn't explain the answer yourself afterwards, you didn't learn it. And know your school's policy before using AI on assessed work.
For teachers: set a clear classroom policy and teach AI literacy alongside your subject — how these tools work (see how LLMs are trained), where they fail, and how to check their output. Design assessments AI can't do for students: in-class writing, presentations, process portfolios. Use AI for prep and feedback yourself, and be transparent about it.
For parents: talk with your children about honest use rather than just policing it. Ask what the school's rules are. And keep the goal in sight: AI should build your child's capabilities, not replace them.
Related reading
- How AI Agents Are Changing Everyday Work — the same assistant technology, reshaping offices instead of classrooms.
- The Ethics of Artificial Intelligence — the bigger questions of fairness and accountability in AI systems.
- Large Language Models Explained in Plain English — what's actually inside the tools teachers and students are using.
The takeaway checklist
- AI tutors like Khanmigo (free for teachers) and Duolingo Max now give every student one-on-one practice — the technology's clearest win.
- Adaptive learning personalises pace and difficulty, but it multiplies good teaching rather than replacing it.
- Detection tools like Turnitin's are far less reliable than claimed (false positives up to ~12% for some groups); redesigning assessment beats policing it.
- Teacher adoption is soaring (60–80% in recent surveys) while school policies lag — only ~18% report clear written guidance.
- Watch the risks: bias in AI outputs, students outsourcing thinking, student-data privacy, and equity gaps.
- Good use means: students use AI to learn, not to avoid learning; teachers set clear policies and AI-proof assessments; parents talk about honest use.
