What the Latest Education Research Says About Learning, Teachers, and AI in Schools

Editorial Note

This article reviews several recent education studies and working papers released during 2026, including research published or made available in June and July.

The evidence is not equal in strength. Some findings come from peer-reviewed journal articles, while others come from EdWorkingPapers hosted by the Annenberg Institute at Brown University. Working papers allow researchers to share emerging evidence before formal journal publication, but their findings may change after additional review.

This article is intended for educational and informational purposes. It does not endorse a particular artificial-intelligence platform, educational technology company, tutoring product, school policy, or teacher-training program. Schools should independently review privacy, accessibility, academic-integrity, instructional, and legal considerations before adopting new technologies.

Artificial intelligence is becoming easier for schools to purchase, but the newest education research points to a more difficult question.

What actually helps students learn?

Recent studies suggest that simply placing an AI tool in front of students is not enough. Students may ignore it, use it for only a few minutes, misunderstand its feedback, or depend on it without knowing how to judge the accuracy of its responses.

The research also suggests that teachers remain central even when technology provides the initial explanation, feedback, or support. Human educators help students remain engaged, interpret information, decide what needs revision, and connect a digital activity to a larger learning goal.

Outside the AI debate, new findings also reinforce familiar educational principles. Principals can significantly influence how teachers experience leadership in their schools. Teacher candidates develop stronger instructional skills when they practice and receive coaching rather than only reading about teaching. Students respond differently to feedback depending on who provides it and how well they understand what to do next.

The larger message is not that technology has failed or that schools should reject AI.

It is that access, automation, and high-quality output should not be confused with learning.

AI Tutoring Did Not Improve Reading Achievement by Itself

A June 2026 working paper examined what happened when elementary students received access to an AI literacy platform.

Researchers conducted two randomized controlled trials. Students were assigned either to use the platform independently or to receive help from an in-person tutor whose role was to promote participation rather than teach the reading content directly.

The results revealed a major implementation problem.

Nearly half of the students expected to use the platform independently never used it. Students who did participate averaged only approximately two to five minutes of use each week.

Adding a human tutor increased engagement by between 71% and 80%, although the increase in actual weekly usage amounted to only about one to four additional minutes. Even with that human support, total use remained low, and the intervention did not produce measurable improvements in reading achievement.

The study does not prove that every AI tutoring system is ineffective.

It shows that giving students access to a platform does not guarantee meaningful use. A tool may be technically available while remaining educationally inactive.

Schools Cannot Assume Students Will Use Technology as Intended

Technology plans often begin with what a program can do.

An AI platform may be able to generate explanations, adjust reading materials, recommend lessons, provide practice questions, or respond to students immediately.

Those capabilities mean little when students do not use the platform long enough to benefit.

The AI tutoring study illustrates why implementation should be treated as part of the educational intervention rather than an administrative detail.

Schools need to consider when students will use the tool, how long sessions will last, what teachers will do during those sessions, how participation will be monitored, and what will happen when students disengage.

A school may purchase access for hundreds of students and still produce little learning if technology use is optional, confusing, poorly scheduled, or disconnected from classroom instruction.

The finding also challenges the idea that students naturally prefer digital learning.

Students may enjoy phones, games, videos, or social media without wanting to complete an academic activity simply because it appears on a screen.

Human Support Increased Participation but Did Not Guarantee Learning

The in-person tutors in the AI literacy study were not primarily responsible for teaching reading.

Their role was to keep students engaged with the digital platform.

That support increased participation, which demonstrates that human presence still mattered. A student may need encouragement, redirection, troubleshooting, accountability, or reassurance before using an automated system consistently.

However, the improvement in engagement did not produce a detectable improvement in reading achievement.

That distinction is important.

Engagement is necessary for many learning activities, but engagement alone does not establish that meaningful learning occurred.

Students can spend more time clicking through a program without thinking deeply, remembering information, or transferring new skills into classroom work.

Schools should therefore evaluate more than login rates, completed sessions, and total minutes.

They should also examine whether students can demonstrate the intended skill without the platform.

A Four-Year AI Chatbot Study Found Administrative Benefits, Not Academic Gains

Another July 2026 working paper followed an AI-enabled text-messaging chatbot at a large urban public university over four years.

Researchers examined the conditions that allowed the system to continue operating, how students responded to it, and whether it affected academic outcomes.

Students remained generally receptive to the chatbot. The strongest effects involved completing time-sensitive administrative tasks, such as responding to deadlines or navigating institutional requirements.

The researchers did not detect improvements in academic performance or student persistence. They also found that centralized institutional ownership and the ability to adjust communications were important for maintaining the system over time.

This finding provides a more realistic picture of what educational AI may do well.

An AI system does not need to transform instruction to be useful. It may help students remember a deadline, complete a form, find an office, understand a registration process, or take the next step in a complicated administrative system.

Those outcomes matter, particularly for students who do not have family members or advisers familiar with college procedures.

However, administrative convenience should not be presented as proof of better learning.

AI May Be Better at Removing Friction Than Replacing Instruction

Schools and universities frequently contain procedural obstacles that are difficult for students to navigate.

A student may need to complete financial-aid documents, register for classes, confirm enrollment, meet advising requirements, or submit information before a deadline.

Missing one small step can create consequences far larger than the mistake itself.

An AI chatbot may be well suited to addressing this type of problem because many administrative questions are repetitive, time-sensitive, and based on information the institution already possesses.

The technology can provide reminders and direct students toward resources without requiring a staff member to answer every basic question individually.

This does not mean the system can replace advisers.

Students facing academic difficulty, financial instability, disability-related needs, family emergencies, or uncertainty about their future may require judgment and personal support that an automated message cannot provide.

The research suggests a practical division of labor. AI may help manage routine communication, while trained professionals focus on situations requiring context, empathy, discretion, and complex decision-making.

Most Students Use AI, but Few Feel Prepared to Use It

A peer-reviewed 2026 study analyzed survey responses from 3,839 higher-education students across 16 countries.

Eighty-six percent reported using generative AI. Only 23% felt prepared to use it, and many reported inadequate guidance from their institutions.

The researchers concluded that student adoption had moved faster than institutional readiness. They called for clearer guidance, stronger AI literacy, and assessments better suited to an environment in which AI tools are widely available.

The authors also acknowledged important limitations. The findings relied on voluntary, self-reported, cross-sectional survey data. The study cannot determine whether AI use caused particular learning outcomes or whether the respondents represented all students equally.

The gap between use and preparation should concern schools.

Students do not necessarily wait for formal instruction before adopting a new technology. They experiment, learn from classmates, watch videos, and develop habits based on convenience.

By the time a school produces an official policy, students may already have used AI for months or years.

AI Literacy Requires More Than Prompt Writing

AI literacy is sometimes reduced to teaching students how to create a better prompt.

That is only one small part of responsible use.

Students need to know how to verify a factual claim, identify missing evidence, recognize fabricated citations, protect personal information, compare an AI response with reliable sources, and explain which parts of an assignment remain their own work.

They also need to understand when AI use interferes with the purpose of an assignment.

A student practicing multiplication, sentence construction, research evaluation, translation, or argumentative writing may be asked to complete work independently because the teacher needs evidence of the student’s own thinking.

Using AI to produce the answer may help the student finish faster while preventing the skill from developing.

A strong AI-literacy program should therefore begin with purpose.

Students should ask not only whether AI is permitted, but what they are supposed to learn and whether the tool supports or replaces that learning.

AI Produced Stronger Writing Feedback, but Students Did Not Revise More Effectively

A peer-reviewed study involving 70 graduate students compared teacher feedback with two forms of generative-AI feedback.

Students first wrote argumentative essays and then revised them after receiving feedback.

AI feedback generated with a step-by-step prompting method was rated as higher quality than both basic AI feedback and teacher feedback. It was more closely aligned with the criteria used to evaluate argumentative writing.

However, the higher-rated AI feedback did not produce significantly greater improvements in student revisions.

Teacher feedback, even though it received lower quality ratings under the study’s rubric, led to comparable gains in essay quality.

The researchers concluded that feedback quality alone was not enough. Students also needed to engage with the feedback, understand it, and apply it appropriately. They suggested that hybrid systems may be most useful when teachers help students interpret and use AI-generated feedback.

This result exposes a common misunderstanding about feedback.

The best feedback is not necessarily the response containing the greatest amount of detail.

It is the feedback that helps a learner take the next productive step.

More Feedback Can Create More Confusion

Generative AI can produce a long analysis within seconds.

It may identify weaknesses in structure, evidence, grammar, transitions, style, organization, and logic all at once.

That volume can appear impressive while overwhelming the student.

A teacher may provide fewer comments because the teacher knows which problem matters most at that stage of the learner’s development.

One student may need help developing a clearer claim. Another may need to connect evidence to the argument. A third may need to reorganize paragraphs before editing individual sentences.

The teacher can prioritize.

An AI system may identify several genuine problems without understanding which one the student is ready to address first.

This helps explain why a technically stronger response may not produce a stronger revision.

Schools should evaluate feedback by what students do with it, not by how sophisticated or comprehensive it appears.

Teachers Still Provide Meaning and Context

Teacher feedback carries information beyond the written comment.

Students may know what the teacher emphasized during class, how the current assignment connects to earlier lessons, and which skills will matter in the next unit.

The teacher can also ask a follow-up question, notice confusion, revise an explanation, or provide a model when a student does not understand.

AI feedback is often delivered as a finished product.

The student receives a response but may not know whether it is accurate, appropriately prioritized, or aligned with the teacher’s expectations.

A hybrid approach could use AI to identify possible areas of concern while preserving the teacher’s responsibility to determine what matters instructionally.

This does not require teachers to approve every AI-generated sentence individually. It requires schools to avoid presenting automated feedback as a complete substitute for professional judgment.

Teacher Preparation Improves When Candidates Practice

Recent education research also reinforces the importance of practice in teacher preparation.

An April 2026 working paper randomly assigned 149 preservice teachers to two forms of online professional learning.

One group completed a conventional module involving readings and reflection questions. The second group completed a practice-based module that included examples, simulated instruction, live coaching, and opportunities to perform the targeted teaching skill.

The practice-based module produced stronger skill development across several measures, including observations from student-teaching classrooms.

The target skill involved modeling mathematical thinking aloud for elementary students, including students with disabilities. The researchers found evidence that candidates transferred what they practiced into actual classroom settings.

The finding may sound obvious, but it has significant implications.

Teacher education frequently asks candidates to read about good instruction without giving them enough structured opportunities to perform it, receive feedback, and try again.

Knowing what an effective teacher should do is not the same as being able to do it during a live lesson.

Teacher Training Should Resemble the Work Teachers Perform

A teacher may understand a written explanation of classroom questioning and still struggle to ask productive questions while monitoring student behavior, checking understanding, managing time, and responding to unexpected answers.

Teaching is a performance profession.

It requires knowledge, but it also requires rapid judgment and the coordinated use of several skills at once.

Simulated classrooms can provide candidates with a lower-risk environment in which to practice. A candidate can attempt an explanation, receive coaching, identify a weakness, and repeat the activity before working with a full classroom of children.

Simulation should not eliminate supervised classroom experience.

Real students bring backgrounds, personalities, relationships, learning needs, and reactions that an artificial environment cannot reproduce completely.

However, simulation can help candidates prepare for those experiences instead of treating the first live lesson as the first opportunity to practice.

Online Teacher Education Does Not Need to Be Passive

The study also challenges the assumption that online preparation must consist primarily of videos, readings, quizzes, and discussion posts.

Digital teacher education can include practice, observation, feedback, and coaching.

The crucial question is not whether the training occurs online.

It is whether participants are asked to perform the skill they are expected to learn.

This principle extends beyond teacher preparation.

Students generally develop writing by writing, speaking by speaking, solving problems by solving them, and conducting experiments by making decisions within scientific investigations.

Instruction may begin with an explanation, but competence develops through guided performance.

Principals Significantly Influence Leadership-Related Working Conditions

A June 2026 working paper examined how principals affected teacher working conditions in Illinois.

The researchers used annual state data and patterns surrounding principal turnover to estimate differences attributable to school leadership.

They found substantial variation among principals, particularly in conditions directly connected to leadership.

Principals had much smaller effects on working conditions outside their immediate influence. The researchers did not find evidence that estimated improvements in working conditions translated into meaningful differences in teacher turnover.

The findings support two conclusions that may appear contradictory.

Principals matter, but principals do not control everything.

A school leader can affect communication, trust, expectations, professional support, decision-making, and the way teachers experience administration.

The principal may have less control over salaries, regional housing costs, state accountability systems, district staffing rules, student needs, building conditions, and other structural pressures.

Better Leadership May Not Immediately Solve Teacher Turnover

Schools often treat teacher retention as a direct test of principal quality.

Leadership certainly affects whether teachers feel respected and supported. However, a teacher may leave a well-led school because of salary, family relocation, retirement, certification requirements, commuting distance, workload, childcare, or a better opportunity elsewhere.

Similarly, a teacher may remain in a poorly led school because leaving would create financial or personal hardship.

Turnover is therefore an imperfect measure of leadership.

Districts should still monitor retention patterns, but they should also gather direct information about communication, trust, professional autonomy, workload, and access to instructional support.

A school can improve as a workplace even when those improvements do not immediately produce a visible change in turnover.

The Latest Research Does Not Support Replacing Teachers With AI

Across the new studies, AI performed useful functions.

It generated detailed writing feedback. It delivered administrative reminders. It offered personalized literacy activities. It created the possibility of support at a larger scale.

The research did not show that these systems could replace teachers.

Students needed human encouragement to use an AI tutoring platform. Better AI feedback did not automatically produce better essay revisions. An administrative chatbot helped with procedural tasks but did not improve academic performance or persistence.

The evidence points toward a supporting role.

AI may help teachers create materials, identify patterns, generate practice activities, provide initial feedback, translate information, or reduce repetitive administrative work.

Those uses become educationally valuable only when they are connected to a clear instructional system.

Technology Should Be Evaluated as Part of a Learning Design

Schools sometimes evaluate AI tools as isolated products.

They examine the available features, price, interface, and technical specifications.

A stronger evaluation begins with the learning environment.

Who will use the tool? What will students be doing immediately before and after using it? What knowledge should they gain? Who will review the output? How will teachers know when the AI is wrong? What information will be collected? Can students with disabilities access it? Will multilingual learners receive reliable support?

A platform that performs well in a controlled demonstration may produce little benefit when placed inside a crowded classroom with limited time and inconsistent internet access.

The quality of the product matters.

The quality of the implementation may matter just as much.

Schools Need Clearer Rules for Student AI Use

The finding that most surveyed students used AI while relatively few felt prepared suggests that institutional silence is not preventing adoption.

It is creating inconsistent adoption.

One teacher may permit AI brainstorming. Another may prohibit all use. A third may require disclosure, while another may have no written expectations.

Students moving among these classrooms may struggle to understand what counts as responsible assistance and what counts as academic misconduct.

Schools should clearly distinguish among uses such as brainstorming, tutoring, translation, editing, source discovery, data analysis, answer generation, and submitting AI-produced work.

A single rule stating that AI is either “allowed” or “not allowed” is unlikely to address the range of activities involved.

The policy should also explain when students must disclose their use and how they should describe what the tool contributed.

Assessment Must Change Without Abandoning Standards

AI has made some traditional assignments easier to complete without demonstrating genuine understanding.

That does not mean schools should abandon essays, homework, or independent research.

It means assessments may need additional evidence of student thinking.

A teacher can require students to submit outlines, source notes, drafts, revision explanations, oral defenses, classroom writing, or reflections on decisions made during the assignment.

These practices make the learning process more visible.

They also reduce the temptation to treat AI detection software as the primary solution.

AI detectors can produce uncertain or incorrect results, and a score alone does not establish what a student did.

Assessment design should focus on evidence the teacher can examine directly.

The Strongest Pattern Is the Continued Importance of Human Guidance

The latest studies examine different subjects and age groups, but they point toward a shared principle.

Learning depends on what students do with information.

An AI platform can provide an activity, but students must participate. A feedback system can identify weaknesses, but students must understand and apply the suggestions. A teacher-training module can explain instructional practice, but candidates must rehearse the practice and receive coaching.

Human guidance supports that transition from access to action.

This does not make technology unnecessary.

It makes educational design necessary.

How New To Education Covers Education Research

New To Education reports on education studies to help readers understand what the evidence shows, what it does not show, and how findings may affect students, teachers, families, and school leaders.

Research coverage should distinguish peer-reviewed studies from working papers, surveys from experiments, and measurable outcomes from assumptions about what a technology might accomplish.

This distinction is particularly important for artificial intelligence.

AI products are evolving faster than long-term education research can evaluate them. Schools may need to make decisions before researchers have definitive answers.

That makes careful implementation, local evaluation, transparency, and teacher involvement even more important.

Key Takeaways

Recent education research suggests that access to artificial intelligence does not automatically improve student learning. In two randomized trials, many elementary students barely used an AI literacy platform, and the intervention did not improve reading achievement. Human tutors increased engagement but did not produce measurable academic gains.

A four-year university study found that an AI-enabled chatbot helped students complete some time-sensitive administrative tasks. Researchers found no detectable improvements in academic performance or persistence.

A peer-reviewed international survey found that 86% of participating higher-education students used generative AI, but only 23% felt prepared to use it. The findings suggest that student adoption is moving faster than institutional guidance.

AI-generated writing feedback was rated more highly than teacher feedback in one experiment, but it did not lead to greater improvements in student essays. Students’ ability to interpret and apply feedback remained essential.

Teacher candidates developed stronger instructional skills when online modules included simulated teaching practice and coaching rather than readings and reflections alone.

Principals significantly affected leadership-related working conditions, although those effects did not clearly translate into lower teacher turnover.

The overall research supports using AI to complement educators rather than replace them.

Frequently Asked Questions

Does the latest research show that AI improves student learning?

Not consistently. Some systems can provide useful support or feedback, but recent studies show that access alone does not guarantee regular use or stronger academic outcomes.

Did AI tutoring improve elementary reading scores?

The June 2026 working paper discussed in this article found no measurable improvement in reading achievement. Human tutors increased students’ engagement with the platform, but overall use remained low.

Are students already using generative AI?

Yes. In one peer-reviewed survey involving 3,839 higher-education students across 16 countries, 86% reported using generative AI.

Do students feel prepared to use AI responsibly?

Most respondents in that study did not. Only 23% reported feeling prepared, and many believed their institutions had not provided sufficient guidance.

Is AI feedback better than teacher feedback?

AI feedback received higher quality ratings in one small experiment involving graduate-student writing. However, it did not produce greater improvement in essay revisions than teacher feedback.

Can AI replace teachers?

The studies reviewed here do not support replacing teachers. Human educators remained important for engagement, feedback interpretation, instructional judgment, practice, and student support.

What does the research say about teacher preparation?

Teacher candidates benefited more from online modules that included simulated practice and coaching than from modules relying mainly on reading and reflection.

Do principals affect teacher working conditions?

Yes. Recent working-paper evidence suggests that principals can significantly influence leadership-related working conditions. Their effects appear smaller in areas controlled by larger district, labor-market, or policy systems.

Are all of these studies peer-reviewed?

No. Some are peer-reviewed journal articles, while others are working papers. Working papers should be treated as emerging evidence rather than final scientific consensus.

Final Thoughts

The newest education research offers a warning against simple conclusions.

Artificial intelligence is neither a guaranteed solution nor an educational failure.

Its value depends on what students are asked to do, whether they use it, how teachers guide the activity, and whether the resulting work demonstrates actual learning.

The research also reinforces something schools have known for generations.

Teaching is not the delivery of information alone.

Students need encouragement, explanation, feedback, practice, accountability, and opportunities to apply knowledge. Teachers need preparation that allows them to rehearse difficult instructional skills. School leaders need to create working conditions in which educators can perform that work effectively.

AI may help schools complete some of these tasks faster.

It may expand access to feedback and reduce administrative friction. It may provide students with another place to ask questions.

However, the strongest educational system will not be the one with the most AI.

It will be the one that understands where technology adds value, where it creates risk, and where a knowledgeable human being remains essential.

Support New To Education

New To Education publishes independent reporting and analysis about education research, school policy, artificial intelligence, teachers, student learning, careers, and developments affecting communities.

Readers can support this work by sharing the article, subscribing to New To Education updates, joining the New To Education community, exploring our educational and professional services, or using the support options available through the website.

https://newtoeducation.com/

New Research Finds Students Rely on AI in Fundamentally Different Ways

https://www.newtoeducation.com/view-blog/new-research-finds-students-rely-on-ai-in-fundamentally-different-ways-6a5da7cf80bd5

New Education Research Shows Students Need More Than AI Access. They Need AI Readiness

https://www.newtoeducation.com/view-blog/new-education-research-shows-students-need-more-than-ai-access-they-need-ai-readiness-6a4f6fa07c40d