AI is already in our students’ homework and our lesson plans. We will go through some studies conducted over the span of 3 years of AI age and pick the responsibilities that can be safely devolved by teachers in favor of AI.
The short version
- The same kind of AI tool can help students learn more or less. What decides it is how a teacher sets up its use.
- AI is best at saving teachers time, giving students patient practice and coaching new tutors.
- Teachers are still needed to check real learning, decide when a student should struggle and keep students going.
- Most schools still have no written rules for AI. Six simple steps help us fill that gap.
AI has taken education by storm. In December 2025, the research group RAND surveyed about 1,200 US students, from middle school through college (RAND).
- 62% of students said they used AI for homework help. In RAND’s survey seven months earlier, the figure was 48%.
Teachers are using AI too. In early 2026, Gallup surveyed 2,069 US public school teachers (Gallup).
- Only 18% of teachers said their school had given them formal guidance, such as a written policy, on how to use AI at work.
So AI is already changing how students learn and how teachers work. Students use it for homework, and teachers use it to plan lessons. But most of this is happening without any clear plan from schools.
Many of us still ask whether AI will replace teachers. The research points to a more useful question. Which parts of learning can we hand to AI, and which parts still need a teacher?
Here we will walk through studies from 2023 to 2026 to see where AI does a job better than any person can. We’ll also see where it quietly backfires.
Why the Future of Education Isn’t AI Versus Teachers
Two recent studies used the same kind of tool, a general AI chatbot, but their results presented a stark contradiction.
A chatbot on its own: students remembered less
In the first study, a researcher randomly split 120 university students into two groups (Barcaui, Social Sciences & Humanities Open). Both groups studied the same material.
One group studied with ChatGPT. The other group used traditional study methods, with no AI.
Forty-five days later, both groups took a surprise test on what they had studied.
- 57.5% vs 68.5% correct on the surprise test. Students who had studied with ChatGPT remembered less than students who studied without it.
The ChatGPT group had AI help the whole time they studied. Yet less of it stuck. The author calls this kind of unlimited ChatGPT use a “cognitive crutch”. Students lean on the tool instead of doing the thinking themselves.
A chatbot led by teachers: students learned more
The second study ran in Edo State, Nigeria (De Simone et al., World Bank). First-year senior secondary students joined a six-week after-school English program. Other students who weren’t selected for the program formed the comparison group.
Students in the program also used an AI chatbot. The difference was the setup. Teachers picked the topics, set the tasks and led every session.
At the end of the six weeks, both groups took the same test.
- Much higher test scores for students in the program than for students who weren’t selected. They learned in six weeks what normally takes about 1.5 to 2 years of school, by the World Bank’s estimate.
The researchers measured the gain as 0.31 standard deviations. A standard deviation measures how spread out students’ scores are. A rise of that size would move a typical student from the middle of the class to ahead of about 62 out of every 100 classmates.
What the two studies tell us
Same kind of technology. Opposite results. In the first study, students used AI alone. In the second, teachers decided how AI was used.
The tool didn’t decide the outcome. The teaching around it did.
Technology can carry a teacher’s know-how to more students. It can’t stand in for it.
What Technology Does Best in the Classroom
AI earns its place in four jobs. Each one either frees up a teacher’s time or helps a teacher reach more students.
1. It gives teachers hours back
In 2025, Gallup and the Walton Family Foundation surveyed 2,232 US public school teachers (Gallup and Walton Family Foundation). Six in 10 had used an AI tool for work in the 2024–25 school year. They used it to make worksheets, adapt materials for different students and build quizzes.
- 5.9 hours saved each week by teachers who use AI at least once a week. That adds up to about six weeks over a school year.
Teachers in the survey said they spend that time on better feedback, more tailored lessons and talking with parents. Some simply got their evenings back.
2. It offers patient practice
An AI tutor doesn’t get tired of the questions. In Ghana, about 500 students took part in a study of Rori, an AI math tutor that runs on WhatsApp (Henkel et al.).
Students in the Rori group got a phone for one hour a week during their study hall period. They used Rori to practice math on their own. The comparison group followed the same lessons and class hours, without Rori.
After the first year, the researchers compared how much each group’s math scores had grown.
- 0.36 standard deviations more math growth for students who used Rori. The researchers say that is roughly an extra year of learning.
Rori works on basic phones and slow data connections. The extra cost was about $5 per student.
These are first-year results, so we should treat them as promising, not final. Notice the setup, too. Rori ran inside the school day, in a set study period.
3. It coaches new tutors in real time
Stanford researchers built a tool called Tutor CoPilot to help human tutors during live math sessions (Wang et al., Stanford NSSA).
The trial ran in online tutoring, where tutor and student talk through a chat window. It included more than 700 tutors and about 1,000 elementary students from underserved communities. Each tutor was randomly assigned to work with the tool or without it.
During a session, Tutor CoPilot offered the tutor three possible replies. Each one modeled what an experienced teacher might say next, such as a guiding question. The tutor could pick one, edit it, ask for new ones or ignore them.
The researchers then checked how often students mastered the topic of each session.
- 4 percentage points more students mastered the topic when their tutor had the tool.
If 60 out of 100 students mastered a topic without the tool, about 64 out of 100 did with it.
- +9 percentage points for students whose tutors had lower ratings or less experience. The tool cost about $20 per tutor per year.
These tutors gained the most because the tool handed them the moves experienced teachers already make.
Tutors with the tool were also 10 percentage points more likely to ask students to explain their thinking. An analysis of over 350,000 chat messages showed the same shift. They relied less on general praise like “good job.”
The AI didn’t teach the child here. It helped a person teach better.
4. It opens doors for more learners
UNESCO’s 2023 report on technology in education found that digital tools have greatly widened access to teaching and learning materials (UNESCO GEM Report). Free videos, open courses and AI chatbots now reach learners who live far from a good school.
What Only Teacher Expertise Can Do
Every result above had a teacher’s judgment behind it. In Nigeria, teachers led the sessions. In Ghana, Rori ran in a set study period. At Stanford, tutors chose which suggestions to use. These four jobs are where that judgment can’t be handed to a machine.
1. Tell doing from learning
A finished assignment isn’t proof that a student learned something. The OECD makes this point directly in its 2026 Digital Education Outlook (OECD). A student can do a task well with AI’s help and still not learn from it.
The report warns that when students hand their thinking to general chatbots, they may stop checking their own understanding. Over time, that can stop them from building real skill.
Teachers can spot this gap in a way a chatbot can’t. For example, a student gives the right answer but can’t explain how they got it.
2. Protect useful struggle
Some struggle is the learning itself. Think of music scales or basic math facts. Nobody gets fluent by watching someone else practice.
One study shows what happens when AI does the hard part for too long (Darvishi et al., Computers & Education). About 1,600 university students in 10 courses used an online study platform. On it, they reviewed each other’s work and wrote feedback.
For the first four weeks, every student had a built-in AI helper. It checked each review as they wrote it and suggested how to make it better. For the next four weeks, the researchers randomly turned the helper off for some students.
- Weaker reviews without the AI helper. Those students wrote shorter feedback that was less relevant to the work. Their reviews were also more similar to what they had written before.
They had leaned on the helper’s tips instead of learning to write good feedback themselves.
So a teacher has to judge the instance. When should a student keep working on a hard problem alone? When are they truly stuck and need a hint? AI can’t make that call well, because it can’t see the student the way a teacher can.
RAND suggests a simple way to split the work, often called a flipped classroom (RAND). Students first learn the new topic at home, for example by reading or watching a short lesson. They may use AI for that. Then they do the practice problems in class, without AI, while the teacher watches and helps.
3. Keep students going
Students keep going on difficult work when a person they trust is there to help and cheer them on. A chatbot can answer questions, but it rarely gives a student a reason to keep trying.
The OECD’s PISA 2022 tests back this up (OECD). PISA tested nearly 700,000 15-year-olds in 81 countries and economies. It also asked them about school during the COVID-19 closures. Of all the closure-related factors it looked at, having teachers available to help had the strongest link to math scores.
- 15 points higher in math where students said they had good access to help from their teachers. For scale, the OECD average math score in 2022 was 472.
4. Judge which tools earn a place
UNESCO found little solid, independent evidence that digital technology improves education. Products change about every 36 months, faster than anyone can test them. Much of the evidence comes from the companies selling them.
Yet schools rarely ask for proof that a product works before they buy it. In one survey across 17 US states, only 11% of teachers and administrators asked for peer-reviewed evidence before adopting a product.
The OECD adds that most general AI tools weren’t designed for learning. It calls for tools built with teachers, not only for them.
AI works best as an amplifier. Teachers decide what it amplifies.
The Real Risk: AI Is Moving Faster Than Guidance
The biggest risk isn’t bad technology. It’s useful technology used without clear school rules. In most schools, nobody has written down when students may use AI, what they may use it for or how they must show their own work.
Think of the everyday questions. Can a student ask a chatbot to fix their essay? Can a teacher use AI to grade? Right now, each person ends up guessing.
Students are unsure
The RAND survey from the introduction shows how unsure students feel.
- 62% of students used AI for homework help by December 2025.
- 67% agreed that using more AI for schoolwork harms critical thinking. In February 2025, 54% agreed. Students keep using it anyway.
Only about a third of students said their school had one rule on AI for the whole school. For many, the rules change from one teacher to the next.
Teachers are unsure too
The Gallup survey of 2,069 teachers found the same gap.
- 69% of teachers have never been told by their school how to use AI when tutoring a student one-on-one.
- 58% of teachers have never been told how to use AI when grading work or giving feedback.
Tutoring and feedback are where a teacher works most closely with each student. Yet these are the tasks where teachers get the least advice on using AI.
The pattern holds outside the US. In the OECD’s 2024 TALIS survey of lower secondary teachers, reported in its 2026 Digital Education Outlook, 37% used AI for their job. And 72% worried AI makes cheating easier. A student can ask AI to write an essay or solve homework, then hand it in as their own work.
Too few teachers to go around
Then there’s the teacher shortage. UNESCO estimates the world needs 44 million more teachers by 2030 to give every child a primary and secondary education (UNESCO). One in three of them is needed in sub-Saharan Africa.
With teachers this scarce, it’s tempting to see AI as a replacement. But the studies in this guide don’t support that. In Nigeria, students learned more when teachers led the AI sessions. At Stanford, AI helped tutors teach better. When students used ChatGPT on their own, they remembered less.
Clear rules help too. Schools with an AI policy got more time back from AI than schools without one.
- 26% more time saved per teacher in schools with an AI policy (Gallup and Walton Family Foundation).
How to Balance AI and Teacher Expertise in the Classroom
We don’t need to wait for a school or district to hand us a plan. Here are some steps we can take until clear standard rules arrive.
1. Hand AI the admin first
Let’s start where the payoff is clear and the risk is low. We can use AI chatbots such as ChatGPT, Gemini, Claude or Microsoft Copilot to draft worksheets, exit tickets, grading rubrics and emails to parents.
We should treat every draft like work from a new teaching assistant. Before anything reaches a student, we check the facts, the level and the tone.
2. Set a “hints, not answers” rule for student AI
In Nigeria, teachers kept the AI on track by choosing the topics and leading each session. We can build some of that guidance into the tool itself in a few minutes.
Most chatbots let us save standing instructions or build a custom assistant. Examples include custom GPTs in ChatGPT, Gems in Gemini and Projects in Claude. Open the one your students use and paste in something like this:
“You are a tutor for Grade 8 math. Never give the final answer. Ask one question at a time. Ask the student to explain each step before you give the next hint.”
Before students use it, check the tool’s age rules and your school’s privacy policy.
3. Keep core practice AI-free
Some skills students must be able to do on their own, such as basic math, spelling or writing a clear paragraph. Let’s choose those skills first. Then we set aside class time where students practice them with AI switched off, while we watch and help.
This is the flipped classroom RAND recommends. Students can learn new content at home, with AI’s help if they like. The hard practice happens in class, with a teacher in the room.
4. Reinvest the saved time in students
Time saved on paperwork only matters if it reaches students. So we should decide in advance where it goes.
The best use is work only a teacher can do. Write comments on students’ work that tell them what to fix. Take five minutes to sit with one student and ask how they’re doing. Gather the three or four students who are stuck on the same topic and go over it together. This is where a teacher’s time makes the biggest difference.
5. Ask for evidence before we adopt a tool
Before we bring a new AI product into class, we can ask the seller four questions:
- Has someone outside the company tested it?
- Were teachers involved in designing it?
- Does it give students hints or finished answers?
- What happens to student data?
The last question matters more than it seems. UNESCO found that only 16% of countries clearly protect student data privacy in their education laws.
6. Write a one-page AI policy with our team
Let’s keep it short enough that students will actually read it. It should cover three things:
- When students may use AI
- What they may use it for
- How they show their own thinking
For training, UNESCO’s AI Competency Framework for Teachers can be a great starting point (UNESCO). It lists 15 skills teachers need, grouped into five areas. The first area is a human-centered mindset: people stay in charge of decisions and responsible for them, not the tool.
The Future of Education Belongs to Teachers Who Use AI Well
We don’t have to choose between AI and teachers. The research shows each one is good at different jobs. AI is strong at saving time, providing practice and reaching more students. Teachers decide what counts as learning, when struggle helps and why a student keeps trying.
When a teacher guides how AI is used, it helps more students learn more, as it did in Nigeria. When students use AI alone, they get help at every step but may remember less, as the ChatGPT trial showed.
That puts teachers in a stronger position than the headlines suggest. Teachers aren’t competing with the tool. They’re the ones needed to make it work.
Sources
- Schwartz, H. L., & Diliberti, M. K. (2026). More Students Use AI for Homework, and More Believe It Harms Critical Thinking: Selected Findings from the American Youth Panel. RAND Corporation. https://www.rand.org/pubs/research_reports/RRA4742-1.html
- Malek Ash, A., & Senseman, K. (2026). Most Teachers Receive No Formal Guidance on AI Use. Gallup. https://news.gallup.com/poll/710534/teachers-receive-no-formal-guidance.aspx (accessed October 2026)
- Barcaui, A. (2025). ChatGPT as a Cognitive Crutch: Evidence From a Randomized Controlled Trial on Knowledge Retention. Social Sciences & Humanities Open. Elsevier. https://www.sciencedirect.com/science/article/pii/S2590291125010186
- De Simone, M. E., Tiberti, F., Barron, M., Manolio, F., Mosuro, W., & Dikoru, E. (2025). From Chalkboards to Chatbots: Evaluating the Impact of Generative AI on Learning Outcomes in Nigeria. World Bank. https://openknowledge.worldbank.org/bitstreams/cd9cca5b-2ffd-4be1-8313-0f4d2ed6ef4f/download
- Gallup & Walton Family Foundation. (2025). Teaching for Tomorrow: Unlocking Six Weeks a Year With AI. Walton Family Foundation. https://static.waltonfamilyfoundation.org/df/fb/eba12807470a9402d7433cc47dba/teaching-for-tomorrow-unlocking-six-weeks-a-year-with-ai-report.pdf
- Henkel, O., Horne-Robinson, H., Kozhakhmetova, N., & Lee, A. (2024). Effective and Scalable Math Support: Experimental Evidence on the Impact of an AI-Math Tutor in Ghana. Communications in Computer and Information Science. Springer. https://doi.org/10.1007/978-3-031-64315-6_34
- Wang, R. E., Ribeiro, A. T., Robinson, C. D., Loeb, S., & Demszky, D. (2025). Tutor CoPilot: A Human-AI Approach for Scaling Real-Time Expertise. National Student Support Accelerator, Stanford University. https://nssa.stanford.edu/studies/tutor-copilot-human-ai-approach-scaling-real-time-expertise (accessed October 2026)
- UNESCO. (2023). Global Education Monitoring Report 2023: Technology in Education: A Tool on Whose Terms? UNESCO. https://gem-report-2023.unesco.org/ (accessed October 2026)
- OECD. (2026). OECD Digital Education Outlook 2026: Exploring Effective Uses of Generative AI in Education. OECD Publishing. https://doi.org/10.1787/062a7394-en
- Darvishi, A., Khosravi, H., Sadiq, S., Gašević, D., & Siemens, G. (2024). Impact of AI Assistance on Student Agency. Computers & Education. https://doi.org/10.1016/j.compedu.2023.104967
- OECD. (2023). Decline in Educational Performance Only Partly Attributable to the COVID-19 Pandemic (PISA 2022 results). OECD. https://www.oecd.org/en/about/news/press-releases/2023/12/decline-in-educational-performance-only-partly-attributable-to-the-covid-19-pandemic.html (accessed October 2026)
- UNESCO. (2023). The Teachers We Need for the Education We Want: The Global Imperative to Reverse the Teacher Shortage (fact sheet). UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000387001
- Miao, F., & Cukurova, M. (2024). AI Competency Framework for Teachers. UNESCO. https://doi.org/10.54675/ZJTE2084