What We Know
Artificial intelligence is entering classrooms faster than researchers can establish whether it consistently improves learning.
On November 10, 2026, Stanford University's SCALE Initiative published an updated review of research examining the educational effects of AI.
The November 2026 review analyzed 52 studies identified as providing strong causal evidence, an increase from 20 studies in Stanford's March review.
The researchers considered findings involving student-facing AI systems, educator-facing tools, learning outcomes, and implementation strategies.
The results challenge two common assumptions: that artificial intelligence automatically improves education and that its use necessarily damages learning.
Instead, the evidence suggests that the effects depend substantially on how students use AI and how the tools are designed.
AI appears more promising when students use it to understand concepts, practice skills, or receive feedback than when they use it to complete academic work on their behalf.
Even then, benefits are inconsistent.
Editorial Note
This article provides independent analysis of Stanford University's November 2026 evidence review.
New To Education is not affiliated with Stanford University or the researchers involved.
Many studies included in the review examined postsecondary students rather than American K–12 classrooms. The conclusions should therefore not be interpreted as proof that the same effects will occur across every school, grade level, or academic subject.
The Difference Between Completing Work and Learning
Consider two students assigned a mathematics problem.
The first asks an AI system to provide the answer and copies the solution.
The second attempts the problem independently, identifies where they became confused, and asks AI to explain the next step.
Both students used artificial intelligence.
However, the educational experiences are fundamentally different.
The first student may complete the assignment without developing the ability to reproduce the reasoning.
The second student may use AI feedback to address a misunderstanding.
Stanford's review indicates that AI is generally more promising when it supports the learning process rather than substitutes for it.
A completed assignment is not necessarily evidence of mastery.
Schools should distinguish between the quality of work produced with assistance and a student's ability to demonstrate understanding independently.
What Did the Research Find?
The Stanford review examined research involving general-purpose chatbots, specialized tutoring systems, feedback tools, and other educational AI applications.
Some interventions improved immediate academic performance.
Others produced no measurable learning advantage compared with conventional educational materials or existing platforms.
A particularly important finding involved students' perceptions.
Students sometimes reported that AI made learning easier, clearer, or more enjoyable, even when the research detected no corresponding improvement in achievement.
This distinction matters because satisfaction and learning are different outcomes.
An application can be engaging without producing durable academic progress.
The Risk of Becoming Dependent on AI
One study discussed in the review examined mathematics practice with unrestricted AI assistance.
Students using an unrestricted system performed better during supported practice but worse on later assessments completed without AI.
The researchers found that unrestricted assistance could create dependence rather than transferable understanding.
This result should not be generalized to every AI tool.
However, it demonstrates why educational systems should evaluate whether students retain knowledge when assistance is removed.
A useful test is straightforward: can the learner explain the concept or solve a comparable problem independently?
If not, the technology may be completing more of the intellectual work than intended.
Can AI Tutors Replace Human Educators?
The research does not support a simple conclusion that AI tutoring should replace human instruction.
Some AI systems delivered useful support. Others offered little improvement.
The review also identified differences in how AI affected educators, depending on their professional experience and the tasks involved.
Human educators perform responsibilities that extend beyond providing explanations.
They identify misunderstandings, adapt instruction, encourage participation, and evaluate whether students are developing confidence and independence.
AI may help deliver practice materials or immediate feedback, but teachers remain responsible for evaluating the educational value of those activities.
The most useful question may not be whether AI can replace a tutor.
It may be how technology can help educators provide more effective instruction.
What Schools Should Measure
Schools considering educational AI should look beyond adoption statistics.
The number of students using a platform does not establish whether academic performance improved.
Useful measures include independent assessments, retention of knowledge, students' ability to explain their reasoning, and changes in achievement over time.
Institutions should also consider whether AI benefits different student groups equally.
The Stanford review found that evidence concerning educational equity and long-term learning outcomes remains limited.
A district should be cautious about promising sustained academic benefits before those outcomes have been demonstrated.
Practical Recommendations for Educators and Families
Students can be encouraged to attempt difficult work before consulting AI.
When they use AI, they should ask for explanations, hints, and feedback rather than automatically requesting completed answers.
Teachers can design activities requiring students to explain their reasoning, revise their thinking, or demonstrate skills without assistance.
Families can ask children to explain what they learned after completing an AI-supported activity.
Schools should provide guidance about acceptable use, academic integrity, and the limitations of generated information.
AI may become a valuable educational tool, but effective use requires deliberate instructional design.
Key Takeaways
- Stanford's November 2026 review examined 52 studies providing strong causal evidence.
- AI is more promising when supporting learning than when completing work for students.
- Positive student experiences do not necessarily produce higher academic achievement.
- Benefits vary across tools, tasks, and educational settings.
- Long-term learning effects remain insufficiently understood.
- Human educators remain important in evaluating and guiding learning.
The Bottom Line
The evidence does not justify either unconditional enthusiasm for AI or blanket rejection.
Artificial intelligence can assist learning under appropriate conditions.
However, a student who produces better work with AI has not necessarily become a stronger independent learner.
The challenge for schools is to use technology in ways that strengthen understanding rather than replace intellectual effort.
Education should be measured by what students can ultimately do—not merely by what their technology can produce.
Support New To Education
For families considering additional academic support, human-led tutoring remains one option for identifying learning gaps, practicing skills, and evaluating understanding.
New To Education connects learners with independent educators offering online and, where available, face-to-face instruction.
Families can explore educators through NTE Nearby and evaluate whether a tutor's qualifications and teaching approach match their needs.
Related Articles
- What the Latest Education Research Says About Learning, Teachers, and AI in Schools
- How to Use AI Tutoring Without Replacing Real Learning in 2026
Sources
Comments (0)
No comments yet. Be the first to share a thought.