Generative AI tool linked to lower student motivation | ETIH EdTech News - EdTech Innovation Hub

Generative AI tool for teachers linked to lower student motivation in trial

A randomized school trial has found that giving teachers access to a generative AI support tool reduced student motivation and did not improve average academic performance, raising questions about how schools deploy teacher-facing AI products.

The research, by Alp Süngü, Benjamin Lira Luttges and Angela L. Duckworth, was conducted across a network of middle and high schools in Turkey during the spring 2025 semester.

The study covered 193 teachers, 2,816 students and 14,198 student-course observations across 14 schools. Teachers were randomly assigned to continue business-as-usual teaching, receive access to a custom generative AI teaching support tool, or receive the same AI access with weekly reminders and usage feedback.

The tool was powered by GPT-4o and included curriculum materials from the Turkish Ministry of Education. Teachers could use it for lesson planning, assessment creation, student feedback, differentiated instruction and administrative communication.

Süngü, Assistant Professor at The Wharton School, framed the finding bluntly in a LinkedIn post: “Most teachers now use AI. But what does it do to students?”

Students found AI-assisted courses less engaging

Students taught by teachers with AI access rated their courses as less important, less enjoyable and less interesting.

The researchers found that teacher AI access reduced student intrinsic motivation by 0.11 standard deviations. The average effect on student confidence was also negative, although reported as marginal.

There was no statistically significant average effect on academic performance across the full sample. That overall result, however, hid a split between different teacher groups.

Students taught by lower-performing teachers scored 0.129 standard deviations worse when their teachers had AI access. They also reported lower confidence. For students of higher-performing teachers, the academic performance effect was directionally positive but not statistically significant.

Süngü wrote: “No average effect on academic performance. But the null hides two offsetting stories. Students of lower-performing teachers scored 0.13 SD worse — and lost confidence too! Effects on higher-performing teachers' students are directionally positive, but not significant.”

AI was used mainly to produce materials

The study’s usage data gives a clearer view of what teachers actually did with the AI tool.

The researchers found that most use was focused on creating teaching materials. Sixty-six percent of conversations involved tasks such as lecture materials, homework, exams, syllabus design and student reports.

Instructional support tasks, including differentiation, misconception correction, student feedback and stress management, accounted for 16 percent of conversations.

Most interactions were short. The median conversation was two user prompts, suggesting many teachers were using the tool to generate outputs rather than refine, challenge or adapt them through longer exchanges.

Süngü wrote: “What might explain these findings? We analyzed teachers' conversations with AI. The median conversation is just 2 prompts long, and 2/3 of use is generating course materials (lecture notes, homework, exams, etc.). Minimal iteration, minimal instructional support. This looks like a substitution of pedagogical effort, not augmentation.”

The motivation decline was larger among teachers who already reported heavier AI use before the experiment. Süngü said the effect was “~3x larger for teachers who were already heavy AI users.”

Researchers question simple productivity claims

The paper challenges one of the most common arguments for AI in schools: that tools which save teachers time will automatically improve learning.

The researchers suggest AI-generated materials may be technically competent but too generic, particularly when teachers accept outputs with limited adaptation. Süngü put the concern more directly: “AI-generated materials may be competent but generic. Students notice when their teacher's own voice disappears. And weaker teachers seem least equipped (or motivated) to critically adapt AI output.”

The study also found that teachers’ own views shifted differently depending on prior AI use. On average, access to the tool did not significantly change teacher beliefs about whether AI helps or harms student learning.

But teachers who had used AI more heavily before the experiment became more pessimistic after further exposure, while lighter users became more optimistic.

Süngü wrote: “Familiarity did not breed acceptance. Experience bred concern.”

The authors do not argue that AI has no role in schools. The paper points instead to the need for stronger training, more careful implementation and tool designs that push teachers toward pedagogical judgment rather than fast content production.

Süngü wrote: “This doesn't mean AI has no place in schools. It means productivity gains from other knowledge work don't automatically transfer to classrooms — where relationships, not just output, drive learning. Successful deployment might need guardrails and training, especially for the teachers most eager to adopt.”

The paper is dated June 25, 2026. The researchers note that the intervention lasted one semester and used one specific AI tool, meaning results could differ with longer exposure, different product designs, stronger guardrails or more discriminating assessments.