Personalized learning means fitting instruction to each student: what a student is taught next, how fast, and with what kind of help. For decades it has been easier to describe than to deliver. Schools have tried differentiated instruction, leveled readers and self-paced modules, but in a class of 25 or more, adjusting work to every child’s pace and needs has been exhausting for teachers and often out of reach.
Artificial intelligence has changed what software can do in that role. Adaptive programs can adjust the difficulty of practice as a student works, and AI tutors built on large language models can explain an idea a second or third way, in conversation. The research on AI for personalized learning is now strong enough to show both what works and what can go wrong, and it points in a consistent direction: AI helps when it is built and used to make students think, and it can harm learning when it simply supplies answers.
For district and school leaders, the decisions that follow are practical ones: which tools to buy, what data they collect, who can override them, how to pay for them once the first grant ends, and how to make sure the students who need the most help are not the ones left out. This article covers each in turn. The wider picture of AI in schools is in edCircuit’s AI in Education: A Guide for Schools and Colleges.
What is personalized learning?
There is no single agreed definition. A 2017 RAND Corporation study describes it as instruction focused on meeting students’ individual learning needs while drawing on their interests and preferences, and notes that the basic idea has long been part of American K–12 teaching. Schoolwide models usually add several features: an up-to-date record of each student’s strengths and goals, learning paths that differ from student to student, progress based on demonstrated mastery rather than time in a seat, and flexible use of staff, space and time.
Personalized learning is related to, but broader than, differentiated instruction. Differentiation usually adjusts the same lesson for groups of students by readiness or interest. Personalization can change the content, the order, the pace and the support for each student, and at its fullest it asks students to help set their own goals. Technology is not required for any of it; a teacher conferring with one student about a piece of writing is personalizing. What software adds is the ability to do some of that adjustment for many students at once.
How does AI for personalized learning work?
Most AI-powered personalized learning tools share a common cycle:
- Data collection. The system records a student’s answers to assignments, quizzes and practice questions, and in conversational tools, the questions the student asks.
- Pattern recognition. Algorithms estimate what the student has mastered, where the gaps are, and which errors recur.
- Adaptive adjustment. The tool offers easier or harder material, a different explanation, or more practice on a weak skill.
- Feedback to teachers. Dashboards summarize each student’s progress so a teacher can decide where to step in.
Programs such as ALEKS in mathematics and Accelerated Reader in reading have offered forms of computer-based tailoring for many years. Generative AI adds something those systems lacked: a tutor that responds in ordinary language, can be asked “why,” and can be instructed to guide a student toward an answer rather than give it. That last capability matters more than any other, as the research below shows.
Does AI-personalized learning work?
The evidence is more encouraging than it was a few years ago, and it carries a clear warning.
Before generative AI, the largest American study of schoolwide personalized learning found modest effects. The RAND study, which followed schools in the Gates Foundation-backed Next Generation Learning Challenges program, estimated that students gained about 3 percentile points in mathematics compared with similar students elsewhere. Reading showed a similar trend that was not statistically significant. The researchers also found that none of the schools looked as radically different from traditional schools as the theory predicted, and that the harder practices, such as regular goal-setting conversations between students and teachers, were the least often in place.
Recent studies of AI tutors report larger gains when the tool is designed to teach:
- A World Bank working paper (De Simone and colleagues, 2025) studied a six-week after-school program in Nigeria in which secondary students used Microsoft Copilot. English scores rose by 0.23 standard deviations, which the researchers equated to 1.5 to 2 years of typical schooling.
- In a study in an introductory physics course at Harvard, published in Scientific Reports in June 2025 by Kestin and colleagues, students learned more from an AI tutor built on teaching principles than from an in-class active-learning session. The study involved one university course, so its results do not transfer automatically to younger students.
- The Tutor CoPilot study (Wang and colleagues, 2024) took a different approach: the AI advised human tutors during live sessions rather than teaching students directly. Students whose tutors had the tool were 4 percentage points more likely to master topics, and 9 points more likely when their tutor had been rated lower, at a cost of about $20 per tutor per year. The paper is a preprint.
The warning comes from a study published in PNAS in June 2025 by Bastani and colleagues. In a field experiment with about 1,000 high school students in Turkey, students who used GPT-4 freely during math practice did better on the practice problems but 17 percent worse on the later exam than students who had no access. A version set up as a tutor, with safeguards against handing out answers, largely prevented that harm.
Taken together, these studies suggest that the design of the tool and the way it is used matter more than the presence of AI. A tool that makes students attempt, explain and check their work can add real learning. A tool that lets students skip that work can make practice look better while making learning worse. For a district comparing products, that difference is the first thing to test.
What personalized learning in schools can offer
When the tools are well chosen and teachers stay in charge, AI-supported personalization can serve several groups of students that a single pace of instruction serves poorly:
- Students with gaps in fundamentals can get targeted practice and explanation until they reach mastery, without being held to the class’s pace.
- Students who learn quickly can move ahead instead of waiting for the rest of the class.
- Families can see progress between report cards, where a platform offers a parent view.
- Teachers can spend less time preparing several versions of the same material and more time working with small groups and individual students.
The RAND findings are a reminder that the software is only part of the model. The schools there found the technology easier to adopt than the human practices around it, and those practices, such as regular goal-setting conversations and up-to-date records of each student’s progress, are where much of the promise lies.
Risks school leaders should weigh
- Misplaced students. An adaptive system works from the answers it sees. A student who was tired, distracted or new to the format can be moved to easier material that he or she does not need, and the misplacement can persist if no one checks it. Teachers need the authority, and the time, to override a placement.
- Answers instead of learning. The PNAS study shows how a general-purpose chatbot can raise practice scores while lowering what students actually learn. Tools for students should be set up to guide rather than to answer.
- Unequal access. Per-student subscriptions, devices and home internet all cost money. Without deliberate planning, the districts and students with the fewest resources are the most likely to go without.
- Privacy. Personalization depends on detailed records of what each student does. Federal law, including FERPA and, for children under 13, COPPA, and many state student privacy laws govern how vendors may collect and use that data.
- Too much automation. If students spend long stretches following a program’s prompts, learning can narrow to a sequence of exercises. Discussion, collaboration and the relationship with a teacher remain part of what school is for.
Policy and funding
For many school boards, the conversation about AI begins in the budget office rather than the classroom.
Federal funds. Pandemic relief money from ESSER, which paid for many adaptive platforms, is gone, and districts now weigh software against staff, infrastructure and other interventions. In July 2025 the U.S. Department of Education told grantees that formula and discretionary grant funds may be used for AI-based high-quality instructional materials and AI-enhanced high-impact tutoring, provided the use meets each program’s requirements. The same guidance stresses privacy and involving parents in decisions about new technology.
Subscription costs. Many adaptive platforms charge per student, so costs grow with enrollment and renew every year. A district that starts a platform with one-time money needs a plan for paying for it in the second and third year, or for ending it without disrupting students mid-course.
Contracts. Data-use clauses, renewal terms and price escalators deserve a careful reading. Districts should know where student data is stored, who can see it, whether it is used to train the vendor’s AI, and how it is returned or deleted when the contract ends.
State guidance. States have moved quickly. As of August 2026, 35 states and Puerto Rico had issued official guidance on AI in K–12 schools, according to the tracker kept by the group AI for Education, up from 28 states and the District of Columbia a year earlier, as Stateline reported. The guidance varies widely in what it covers, so district leaders should read their own state’s document before setting local rules. edCircuit’s AI Rules for Schools: A Back-to-School Guide describes what a clear school policy contains.
What other countries have learned. National rollouts carry their own risks. South Korea introduced AI digital textbooks in English, math and computer science in 2025, but by August only about 30 percent of schools had adopted them, and on August 4, 2025, the National Assembly removed their status as official textbooks, as The Korea Herald reported, leaving their use to individual schools. Teachers and parents had objected that the program was rushed. A district-by-district approach, as in the United States, is slower and less even, but it allows a district to pilot, measure and adjust before committing every school.
Questions school leaders should ask before adopting AI personalization
- Can teachers see why the system placed a student where it did, and can they override it?
- Is the student-facing tool set up to guide students to answers, or to supply them?
- What evidence does the vendor have, from what kind of study, with which students?
- What data is collected, how long is it kept, and is it used to train the vendor’s AI?
- How will students and parents understand the role of AI in shaping learning?
- What will it cost in the third year, and which students might lose access if funding falls short?
What each group needs to do
Teachers keep the final say on instruction. AI placements and recommendations are a starting point to check against what a teacher knows about a student.
Curriculum directors need to make sure AI-driven pathways follow the district’s curriculum and standards rather than the vendor’s sequence.
Technology directors and administrators are responsible for privacy review, contracts, infrastructure and staff training, and for measuring whether a tool is working after it is adopted.
Parents should be able to see their child’s progress and understand how AI influences what their child is taught.
Students need to learn how to use these tools well, which is part of AI literacy. edCircuit’s AI Literacy for Students: A Grade-by-Grade Guide sets out what that looks like at each stage.
Parent checklist: questions to ask about AI-powered personalization
- How is AI being used? Is it setting the pace, giving feedback, tutoring, or all three?
- Who reviews the results? Does a teacher check the system’s decisions about my child?
- What happens to my child’s data? Where is it stored, for how long, and who can see it?
- Can I see my child’s progress? Does the platform offer a parent view or regular updates?
- What protects against errors and bias? How would a wrong placement be noticed and corrected?
- What if I opt out? Will my child still receive the same level of support?
Parents looking for guidance on AI at home, by age, will find it in AI for Students: A Parent’s Guide by Age.
From automation to student agency
The most ambitious versions of personalized learning have always aimed for more than adjusting difficulty. They ask students to set goals, track their own progress and reflect on how they learn, so that personalization is something students take part in rather than something done to them. AI tutors that respond in conversation could support that work, for example by helping a student plan practice or explain a mistake in his or her own words.
The same tools can also pull in the opposite direction, toward students following a program’s prompts with little understanding of why. The research so far suggests the outcome depends on choices schools make: which tools they buy, how those tools are configured, and whether teachers have the time and authority to stay involved. Districts that pilot carefully, measure results, protect student data and plan for equitable access are in the best position to gain from AI-supported personalized learning while keeping teachers at the center of it.
Related reading
Subscribe to edCircuit to stay up to date on all of our shows, podcasts, news, and thought leadership articles.
