AI Tutors Are Expanding. New Research Shows Why Human Support Still Matters

Artificial intelligence is moving rapidly into education. AI-centered private schools are expanding into new cities, districts are adopting approved classroom platforms, and families now have access to an almost unlimited supply of explanations, practice questions, study plans, and instant feedback.

Some of these tools are genuinely impressive. A student can ask for another explanation of quadratic equations at midnight, generate ten new vocabulary questions before breakfast, or receive immediate feedback on an essay outline. Used thoughtfully, AI can be a valuable addition to a student’s academic toolkit.

But a revealing new study from Stanford University highlights a distinction that often gets lost in the excitement: providing personalized instructional software is not the same thing as providing effective tutoring.

Access Is Not the Same as Engagement

Stanford researchers recently conducted two randomized controlled trials involving approximately 350 elementary-school students using an AI literacy platform. Students were given dedicated time and were expected to use the program for at least two 30-minute sessions each week.

Yet among students assigned to work independently, nearly half never used the platform at all. Across the full groups, average usage amounted to only about two minutes per week in one district and five minutes in the other.

A second group received support from in-person tutors. Importantly, these tutors were not primarily responsible for teaching the reading material. Their role was to encourage students, help them begin, keep them engaged, and support consistent participation.

That human presence increased engagement with the AI platform by approximately 71% to 80%.

The result is both intuitive and important: even when technology can provide instruction, students may still need a person to help them undertake the difficult work of learning.

The researchers also found that overall use remained low and that the intervention did not produce measurable improvements in reading achievement. The lesson is therefore not that adding a human being automatically makes educational technology successful. It is that software access alone tells us very little about whether meaningful learning will occur.

A Tutor Does More Than Supply Answers

It is tempting to define tutoring as the transfer of information: the student has a question, and the tutor provides an explanation. Under that definition, AI appears capable of replacing much of what a tutor does.

But strong tutoring is not merely an explanation service.

A skilled tutor is continually making judgments that may not appear in the formal lesson plan:

• Does the student truly understand, or have they learned to reproduce a procedure?

• Is the current mistake a minor calculation error or evidence of a deeper conceptual gap?

• Would another explanation help, or does the student need to struggle productively for another minute?

• Is the student confused, tired, embarrassed, distracted, anxious, or simply avoiding an unfamiliar task?

• Should today’s priority be new material, review, confidence-building, organization, or test-taking strategy?

These judgments depend on context accumulated over time. A tutor remembers that a student rushed through the same kind of question last week, recognizes when “I get it” really means “I don’t want to admit that I’m lost,” and knows when to press harder or change direction.

That is personalization in a deeper sense than selecting the next question in an adaptive sequence.

The Best Use of AI May Be Alongside a Human Tutor

The Stanford findings should not be read as an argument for keeping AI out of education. Other research has found that carefully designed AI tutoring systems can improve learning, provide useful practice, and make individualized support more widely available.

A recent Brookings review concluded that AI-enhanced tutoring holds substantial promise—but also emphasized the importance of sound instructional design, effective feedback, safeguards, and a hybrid model in which human educators continue to monitor and guide students.

That hybrid approach makes sense.

AI is exceptionally useful for generating practice material, offering alternative explanations, identifying patterns in a large body of work, and giving students opportunities to practice between sessions. A human tutor can determine which of those tools will actually help a particular student and whether the student is using them to think more deeply—or merely to reach an answer faster.

The distinction matters enormously in test preparation. Success on the SAT, ACT, ISEE, or SSAT does not come simply from seeing more questions. Students need to understand their error patterns, tolerate frustration, manage time, choose when to abandon an unproductive approach, and remain cognitively engaged under pressure.

An AI system may tell a student that an answer is wrong. A good tutor helps the student discover why they were vulnerable to that wrong answer—and how to recognize the same trap when it appears in a completely different form.

Learning Is Still a Human Activity

The central promise of AI tutoring is personalization at scale. That is a worthwhile goal, particularly when individualized academic support is otherwise unavailable.

But learning depends on more than the availability of personalized content. It requires attention, persistence, intellectual honesty, motivation, and the willingness to remain with a problem after the first attempt fails.

Those qualities develop through relationships as much as through information.

The most effective future for tutoring will probably not force families to choose between advanced technology and skilled human instruction. It will combine the reach and flexibility of AI with the judgment, accountability, humor, empathy, and responsiveness of an excellent teacher.

At Exponential Academics, our goal is not simply to help students produce correct answers. It is to understand how each student thinks, identify what is obstructing progress, and build the confidence and habits that allow genuine learning to continue long after a particular test or assignment is finished.

Considering individualized academic support or test preparation? Contact Exponential Academics to discuss what kind of tutoring would best serve your student.

Sources

• Stanford Graduate School of Education: “Access Is Not Enough: Human Support Improves Engagement with AI Tutoring”

• Brookings Institution: “What the Research Shows About Generative AI in Tutoring”

• Axios: “AI-Powered Alpha Schools Expand Into Boston”