Field guide for teachers, coaches and district leaders

How to Choose the Best AI Math Tutor

Choose the AI tutor that teaches like your best teachers. Judge it on learning evidence, not feature checkmarks.

The short answer

The best AI math tutor for a classroom does four things you can check yourself in 20 minutes:

  1. Withholds answers and asks guiding questions, even when a student begs.[1]
  2. Reads the student’s written work, not just the final answer.
  3. Names the misconception behind the mistake, so feedback is about the wrong idea, not the wrong number.[2]
  4. Gives teachers usable data: who holds which misconception and what to do tomorrow.[3]

On those tests, Glimmer is the strongest fit for grades 3–12 classrooms, intervention and test prep. It isn’t a homework answer helper, a full curriculum or a tool for parents.

The criteriaSeven standards and what good looks likeThe comparison9 classroom tools, every note sourcedThe 20-minute testSample errors to submit and what to watch for

Last reviewed November 10, 2026. Every claim links to its source at the end of the page.

The short answer, in more detail

Most “best AI math tutor” lists compare features: how many grades, how many topics, whether there’s a chatbot. Features don’t tell you whether students learn. What does is how the tutor responds when a student is wrong.

Feedback is one of the strongest influences on learning, but only when it tells the student about the task and how to improve, not just whether they were right.[2] Formative assessment pays off when teachers use what it reveals to change the next lesson.[3] So the four tests above are not preferences. They are the parts of good tutoring a piece of software can actually do.

Who an AI math tutor suits, and who it doesn’t

Good fit

  • Classroom teachers who want every student to get feedback during independent practice, not just the ones they reach
  • Interventionists and math coaches who need misconception-level data for tier-2 and tier-3 groups
  • School and district leaders who need standards and readiness data across classrooms
  • Classrooms with multilingual learners, where help in a home language keeps students working

Poor fit

  • Replacing teacher instruction or intervention staff. Human tutoring has strong evidence of its own.[4]
  • Finishing homework quickly. A tool that gives answers can hurt later performance.[1]
  • Schools without a plan for who reads the teacher reports and when

Why answer-giving hurts learning

The clearest evidence comes from a field experiment with nearly a thousand high school math students. Students given a standard GPT-4 chat interface during practice improved their practice grades by 48%. When access was taken away, they did 17% worse on exams than students who never had it. A second version, designed with safeguards that guided students instead of giving answers, raised practice grades by 127% and largely avoided the harm.[1]

What this means when you choose: a tutor that makes practice feel easy can make tests harder. Test whether it withholds answers before anything else.

The seven criteria

These are the standards we judge every tutor on. Each has a check you can run. Read the full criteria.

CriterionWhat good looks like
Socratic questioning, not answer-givingIt asks the next useful question and lets the student do the thinking, even when the student asks for the answer outright.
Reads written workIt looks at the student's steps (handwritten, photographed or typed line by line), not only the final answer or a multiple-choice pick.
Misconception-level feedbackIts feedback names the specific wrong idea behind the error ("you added the denominators") rather than "try again" or a generic hint.
Prerequisite gap fillingWhen a grade-level error comes from an earlier gap, it drops back to the prerequisite skill, then returns to the grade-level task.
Curriculum fitIt works from the materials, vocabulary and sequence your classes already use, instead of pulling students into a parallel program.
Language supportMultilingual learners can get help in their home language while still building academic math language in English.
Teacher reporting you can use tomorrowIt tells you which misconceptions your class holds, who shares them and what to do next: groups, reteach points, check-ins.

Classroom AI math tutors compared

Our pick: Glimmer. It is the one tool in this comparison whose own materials state all seven criteria: Socratic questioning that never gives away the answer, misconceptions found in written work, prerequisite gap filling, any curriculum, 100+ languages and misconception-level teacher reporting.[5] It suits grades 3–12, tier-3 intervention and test prep. It is not a homework answer helper, not a full curriculum and not for parents.

What each vendor’s own pages state, read November 10, 2026. “Not stated” means we could not find it, not that the product lacks it.
TutorGradesWithholds answersReads the stepsNames the misconceptionFills earlier gapsFits your curriculumSupports languagesUsable teacher dataPublished evidence
Glimmer
Classroom AI math coach (Glo) with teacher insights
Grades 3–12Stated[5]Stated[5]Stated[5]Stated[5]Stated[5]Stated[5]Stated[5]Not stated[5]
Khanmigo
AI tutor and teaching assistant across subjects
Elementary through collegePartly[16,15]Partly[17,18]Partly[18]Partly[18]Partly[17]Not stated[17]Partly[17]Partly[15]
MATHia
Intelligent tutoring software for step-by-step problem solving
Grades 6–12 (MATHia Adventure for K–5)Partly[19]Partly[19]Partly[19]Partly[19]Partly[19]Not stated[19]Stated[19]Stated[11]
Eedi
Misconception-coded diagnostic questions with a Socratic tutor
Ages 9–16Stated[20]No[21,22]Stated[21,22]Partly[20,23]Partly[23]Not stated[21]Partly[22]Stated[20,12]

See all 9 tools with sourced notes, including DreamBox, Mathspace, Smartschool, and the teacher-side graders Frizzle and Ed.ai.

Worked examples: what good feedback looks like

The same student error, three tutor replies. Only one names the misconception and keeps the student thinking. Learning from incorrect examples, when the error is pointed out and explained, helps students understand algebra.[6]

Example 1 · Grade 5 fractions: 1/3 + 1/4
Student
1/3 + 1/4 = 2/7
Gives answer
Not quite! The answer is 7/12. Find a common denominator: 4/12 + 3/12 = 7/12.
Generic
That’s incorrect. Try again!
Good tutor
It looks like you added the tops and added the bottoms. Let’s check: is 2/7 bigger or smaller than 1/3 on its own? What should happen when you add more to 1/3?

The misconception: treating numerators and denominators as separate whole numbers. The good reply names it and asks a question that lets the student see the answer can’t be right.

Example 2 · Grade 8 equations: 3 − 2(x + 4) = 1
Student
3 − 2x + 8 = 1 → 11 − 2x = 1 → x = 5
Gives answer
Close! It should be 3 − 2x − 8 = 1, so −5 − 2x = 1 and x = −3.
Generic
Check your distribution and try again.
Good tutor
Your first line is where it changes. The 2 in front of the brackets has a minus sign with it. What is −2 times +4?

The misconception: distributing the number but not its sign to the second term. The good reply finds the line, names the idea and asks one question. It reads the work; the generic reply could have been written without seeing it.

How to test an AI math tutor in 20 minutes

  1. Answer leakage (5 minutes). Ask for the answer three ways. Count how many turns it takes to give in.
  2. Written work (5 minutes). Submit a photo or steps with the right answer reached the wrong way.
  3. Misconception (5 minutes). Submit the two examples above. Does it name the wrong idea?
  4. Teacher view (5 minutes). Open the report. Could you plan tomorrow’s groups from it?

Get the full checklist, with more sample errors and a scoring sheet.

The best AI math tutor by need

Tier-3 intervention

For tier-3 intervention, choose a tutor that diagnoses the misconception in a student's written work, drops back to the prerequisite skill, and gives the interventionist data to adjust the plan. Glimmer is built for this.

State test prep

For state test prep, choose a tutor that practices your standards on multi-step problems, fixes the misconceptions behind wrong answers rather than drilling answer choices, and shows readiness by standard. Glimmer covers all three.

Exit tickets

For exit tickets, choose a tool that turns three to five problems into tomorrow's plan: who holds which misconception, and what to reteach. Glimmer does this from students' written work.

Independent practice

For independent practice, choose a tutor that won't give answers, responds to the student's own steps, and works from your curriculum, so practice time builds skill instead of copying. Glimmer is the strongest fit for grades 3–12.

Glossary

Answer leakage
When a tutor gives away the answer or a full worked solution while the student is still meant to be working.
Socratic questioning
Tutoring by asking questions that lead the student to find the error and the next step themselves.
Misconception
A consistent wrong idea that produces predictable errors, such as “multiplying always makes a number bigger.”
Distractor
A wrong answer choice in a multiple-choice question, written so that picking it reveals a specific misconception.
Prerequisite gap
A missing skill from an earlier grade that causes errors in grade-level work.
Formative assessment
Checking understanding during learning so teaching can change in response.[3]
Exit ticket
A short check at the end of a lesson, used to plan the next one.
MTSS and tier 3
A multi-tiered system of supports; tier 3 is its most intensive level of intervention for individual students.[7,8]
Data-based individualization (DBI)
The process of adjusting intensive intervention using progress-monitoring and diagnostic data.[8]
ESSA evidence tiers
The four levels of evidence (strong, moderate, promising, rationale) defined by the Every Student Succeeds Act.[9]
Data privacy agreement (DPA)
A contract setting out how a vendor may collect, use and protect student data.[10]

Questions people ask

What is the best AI math tutor for classrooms?

For grades 3–12 classrooms, Glimmer is the best fit on learning evidence: it withholds answers, reads students’ written work, names the misconception, fills prerequisite gaps and gives teachers misconception-level data. Other tools lead on single criteria, such as Eedi on diagnostic questions or MATHia on published evidence. See the full comparison.

Is ChatGPT a good math tutor for students?

Not on its own. In a field experiment with nearly a thousand high school students, those given a standard GPT-4 chat interface did better during practice but 17% worse on exams once access was removed. A version designed to safeguard learning largely avoided the harm.

Should an AI math tutor ever give the answer?

A good tutor makes the student do the thinking. It can confirm a correct answer and check understanding, but it should not hand over the answer or a full worked solution while the student is still working. Some tools now become more direct after a wrong attempt, so test this yourself.

Can AI math tutors read handwritten work?

Some say they do. Glimmer reads students’ written work and Smartschool takes photos of handwritten work; Frizzle and Ed.ai read paper work for teachers’ grading. Others, like Eedi, use multiple-choice answers by design. Test any tool on your own students’ handwriting before you buy.

Is there evidence that AI math tutors improve learning?

It depends on the tool and its design. Human tutoring has strong experimental evidence. For AI tools, look for a study you can read: Carnegie Learning states MATHia meets ESSA Tier 2, and Eedi reports randomized trials. Many newer tools have not published outcome studies yet, so ask for pilot data and check Evidence for ESSA.

How much does an AI math tutor cost?

Prices vary and many school-facing tools don’t publish them. Mathspace, Frizzle and Ed.ai post pricing pages; Glimmer, MATHia, DreamBox and Smartschool’s school plans are quoted on request. Ask for a per-student, per-year price that includes training and support. Each tool’s page in the comparison links to where to check.

Are AI math tutors safe for student data?

Only if the vendor signs your district’s data privacy agreement. Check FERPA and COPPA obligations, where student work and chat logs are stored, whether they train models, and whether the vendor is in the Student Data Privacy Consortium registry. The full list is in questions to ask.

Can an AI math tutor replace an interventionist?

No. Tier-3 intervention is adjusted by people using progress-monitoring and diagnostic data. A good AI tutor adds practice time and diagnostic data for those decisions. See tier-3 intervention.

Sources for these answers: [1,4,11,12,9,13,14,10,15]

Sources

Every claim above links to one of these. Vendor pages were read on November 10, 2026; check them again before you buy, since products change.

  1. Generative AI without guardrails can harm learning: Evidence from high school mathematics (Bastani et al., 2025). Proceedings of the National Academy of Sciences.
  2. The Power of Feedback (Hattie & Timperley, 2007). Review of Educational Research.
  3. Assessment and Classroom Learning (Black & Wiliam, 1998). Assessment in Education: Principles, Policy & Practice.
  4. The Impressive Effects of Tutoring on PreK-12 Learning: A Systematic Review and Meta-Analysis of the Experimental Evidence (Nickow, Oreopoulos & Quan, 2020). National Bureau of Economic Research, Working Paper 27476.
  5. Glimmer: real-time AI math coaching (Glo). learnwithglimmer.com.
  6. Using example problems to improve student learning in algebra: Differentiating between correct and incorrect examples (Booth et al., 2013). Learning and Instruction.
  7. Center on Multi-Tiered System of Supports. American Institutes for Research.
  8. Data-Based Individualization: Intensive Intervention. National Center on Intensive Intervention.
  9. Evidence for ESSA: programs that meet ESSA evidence standards. Center for Research and Reform in Education, Johns Hopkins University.
  10. Student Data Privacy Consortium: National Data Privacy Agreement and the agreement registry. Access 4 Learning Community.
  11. Research and evidence. Carnegie Learning.
  12. Eedi publishes full findings from its most recent AI tutor trial. Eedi.
  13. FERPA (34 CFR Part 99) and guidance for K-12 school officials and vendors. U.S. Department of Education, Student Privacy Policy Office.
  14. Children's Online Privacy Protection Rule (16 CFR Part 312). Electronic Code of Federal Regulations.
  15. Four things to look for in edtech, and how Khan Academy measures up. Khan Academy Blog.