Google Study Finds Gemini AI Tutoring Can Match Human Tutors in Some Maths Exercises

Can artificial intelligence help students learn mathematics as effectively as a human tutor?

A new study published by Google on October 29, 2026, suggests that AI tutoring may produce similar results to human tutoring in certain classroom activities.

The research involved more than 1,100 students aged 12 to 15 in the United Kingdom.

Students received support through an educational platform that combined Google’s Gemini AI technology with teaching resources developed by Eedi.

According to Google, students receiving AI tutoring performed similarly to those working with human tutors on platform assignments and intermediate quizzes.

However, the researchers did not observe corresponding improvements in standardised test scores.

The findings suggest AI can support mathematics education, but they also highlight important limitations.

How Was the Study Conducted?

Google worked with educational researchers and teachers to examine how AI tutoring affects student learning.

The researchers conducted a 10-week randomised controlled trial.

A randomised controlled trial is a study in which participants are assigned to different groups to compare the effects of an intervention.

In this case, the researchers compared three types of assistance.

The first group received static hints.

The second group received help from human tutors.

The third group received interactive AI tutoring using Gemini and Eedi’s educational resources.

Whenever students answered a mathematics question incorrectly, they received support according to their assigned group.

The researchers then compared their progress.

What Is Eedi?

Eedi is an educational platform focused on mathematics learning and assessment.

It helps identify common mistakes students make when solving problems.

For example, a student might incorrectly add two fractions because they do not understand how denominators work.

Instead of simply providing the correct answer, Eedi aims to identify the misunderstanding and help the student correct it.

The Gemini-powered tutoring system used this educational approach.

This distinction matters because the study did not test an ordinary chatbot working without educational guidance.

It tested an AI tutoring system specifically designed for mathematics learning.

How Did AI Tutoring Compare With Human Tutors?

According to Google’s research report, students receiving AI tutoring performed similarly to those receiving human tutoring on regular platform activities and intermediate quizzes.

This suggests that a properly designed AI tutor can provide useful instructional support.

However, researchers also measured performance using independent STAR standardised assessments.

The same learning gains did not appear in those standardised test scores during the trial.

Therefore, it would be misleading to conclude that Gemini was proven to be as effective as human tutoring in every educational setting.

The findings are more limited.

Did the AI Give Students Answers?

The research placed considerable emphasis on helping students think rather than simply providing solutions.

Google says Gemini guided students through mathematical problems instead of immediately revealing the answers in nearly 99% of tutoring sessions.

Students worked through the mathematics themselves in more than 85% of their conversations.

For example, if a student struggled with an equation, the AI could ask a guiding question or explain a related concept.

This approach encourages active learning.

It is different from using a chatbot to copy an answer into homework.

How Accurate Was the AI Tutor?

The research also examined how often human reviewers approved the AI’s instructional responses.

Google reports that the outright approval rate increased from 76% in an earlier study to 95% in the latest research.

This represents a substantial improvement in the quality of the tutoring responses under the study’s evaluation process.

However, the figure does not mean Gemini is 95% accurate on every mathematical question.

The measurement concerns human expert approval of responses in a particular educational setting.

Another important detail is that human experts reviewed every AI response in real time before it reached students.

This means the study involved significant human oversight.

Why Didn’t Standardised Test Scores Improve?

Google suggests one possible explanation involves the amount of material students completed.

Students receiving AI tutoring answered approximately 14 questions per session on average.

Students receiving simple hints answered approximately 19.

AI tutoring often involved longer discussions and more detailed explanations.

This may have helped students understand individual problems while reducing the number of topics covered during the 10-week trial.

However, this explanation remains a hypothesis.

The study does not establish a single confirmed reason for the difference.

Google says it is exploring longer trials covering an entire academic year.

What Did Google’s Study in Italy Find?

Google also published findings from research involving educators in Italy.

The study involved 31 schools, around 1,600 teachers, and a student population of more than 24,500 learners aged 13 to 18.

The research examined how teachers used Gemini for Education to plan lessons and support classroom activities.

According to Google, 79% of teachers reported feeling more creative when planning with Gemini, compared with 30% without it.

The study also reported a shift towards more collaborative classroom activities.

Student-reported participation increased from 77% to 84% in the reported comparison.

However, these findings involve classroom observations and self-reported experiences.

They should not be interpreted as proof that AI automatically improves academic achievement.

Can AI Replace Human Teachers?

The research does not support replacing teachers with AI systems.

In the UK study, human experts reviewed AI responses before students received them.

Teachers were also involved in monitoring student progress.

In Italy, teachers used Gemini to help design lessons and classroom activities.

This suggests AI may be most effective when it supports professional educators.

Teachers provide encouragement, classroom management, emotional understanding, and judgment that cannot be reduced to mathematical explanations.

AI can provide additional assistance, but it should not be treated as a complete substitute for human teaching.

What Could This Mean for African Students?

Many students across Africa face challenges accessing additional tutoring.

Private tuition can be expensive, and students in remote areas may have fewer learning resources.

AI tutoring could potentially provide additional explanations and practice exercises.

For example, a student in Kenya could ask an AI assistant to explain how to solve simultaneous equations.

The assistant might guide the learner through each step.

However, the results of the UK study cannot automatically be applied to every African classroom.

Curriculum differences, language, internet connectivity, access to devices, and teacher involvement all matter.

More local research would be needed to establish effectiveness.

How Should Students Use AI for Mathematics?

Students can use AI tools to support learning rather than replace independent thinking.

A useful approach is to attempt a problem before asking for help.

If the answer is incorrect, the student can ask the AI to explain the mistake without immediately revealing the solution.

The student should then try again.

This can help encourage understanding instead of memorisation.

However, AI-generated mathematical steps may contain errors.

Important answers should be checked against trusted learning materials.

QueryFun Take: AI Tutoring Works Best When It Encourages Thinking

Google’s research provides encouraging evidence that AI can support mathematics learning.

However, the findings are more complicated than saying AI is now better than teachers.

Students receiving AI tutoring performed similarly to students receiving human tutoring on certain activities.

They did not show the same improvement on standardised tests during the trial.

The most important lesson is that the quality of AI tutoring matters.

A useful learning assistant should ask questions, explain mistakes, and encourage students to solve problems independently.

For schools and learners, AI should be treated as an additional educational tool rather than a replacement for effective teaching.

Frequently Asked Questions

Did Google prove Gemini is better than human maths tutors?

No. The study found similar results on certain classroom activities and quizzes, but not on standardised test scores.

How many students participated?

More than 1,100 students aged 12 to 15 participated in the UK trial.

How long did the research take?

The UK randomised trial lasted 10 weeks.

Did students use Gemini without supervision?

No. Human experts reviewed every AI tutoring response before it reached students.

Why didn’t standardised test scores improve?

One possible explanation is that AI tutoring took longer per question, reducing curriculum coverage. Further research is needed.

Can ChatGPT or Gemini help students learn mathematics?

Yes. AI can explain concepts and provide guided practice, but answers should be verified.

Should schools replace teachers with AI?

The study does not support replacing teachers. Human oversight was an important part of the research.

Can this research apply to schools in Kenya?

It provides useful evidence, but local testing would be needed to determine how well similar tools work in Kenyan classrooms.