On August 26, 2026, OpenAI published a back-to-school report titled Learning Never Stops, which examines how ChatGPT is being used outside the classroom[1]. Based on a privacy-preserving internal analysis, the company says people across all age groups hold as many as 70 million self-testing conversations per week, and that in the United States, prompts tied to classwork and homework peak at more than 460 million messages per week during the school year[1][2].

Homework Conversations Climb on Sunday Evenings

The report rests on an anonymized analysis of usage data. Across all age groups, up to 70 million conversations per week involve users checking what they know, with behaviors such as answer-checking, misconception checks, and requests for more practice[2].

Within the United States, prompts related to classwork and homework exceed 460 million messages per week at their school-year peak, rising on Sunday evenings just before a new week begins. Even during the summer, when many classrooms are closed, the volume stays above 180 million messages per week[2].

The report also breaks down what those self-testing conversations contain. Answer-checking appeared in 63 percent, iterative learning in 59 percent, the learner's own attempt in 58 percent, requests for practice in 51 percent, and misconception checks in 36 percent. In the broader homework analysis, people aged 18 to 21 accounted for roughly 25 percent of users and about 30 percent of messages[2].

OpenAI states plainly that the analysis does not yet distinguish productive practice from quick-answer seeking or cognitive offloading[2]. The figures describe the volume and shape of usage, not learning outcomes.

Teachers Reclaiming Time

The second pillar of the report covers administrative work on the teaching side. An analysis covering January 1 through July 16, 2026 found more than 1.9 million messages tied to tasks that save educators time: 900,000 about report cards and progress reports, 800,000 about lesson planning, and more than 100,000 each about substitute plans and teacher-evaluation materials[2].

Individual cases are included as well. Ava Morton, an eighth-grade teacher in Clayton County, Georgia, adapts the same materials for students with ADHD, dyslexia, autism, and other learning differences, and estimates she saves at least five hours of planning each week. Kara Crown, a special education teacher outside Chicago, cut the time to draft individualized education plan goals and forms from more than an hour per student to less than 30 minutes. Taiyo Inoue, a college mathematics instructor in California, automates course schedules and announcements in his learning management system and estimates he recovers four to five hours a week[2].

Communication with families is another category. OpenAI identified roughly 1.5 million translation-related messages so far this year, including about 575,000 involving communication with parents and families[2].

The Evidence on Outcomes Is Still Split

The report acknowledges that research on AI and learning remains early and mixed. In a randomized Harvard study, a carefully designed AI physics tutor produced more than twice the median immediate learning gains of an active-learning classroom. In a randomized study of nearly 1,000 high-school math students in Turkey, unrestricted AI improved practice performance but reduced later unaided test performance by 17 percent. A tutor configured with teacher-designed hints largely avoided that drop[2].

The backdrop is unequal access to help. In the most recent published PISA results, conducted in 2022, 47 percent of socioeconomically disadvantaged students across OECD countries scored below basic proficiency in mathematics, compared with 14 percent of advantaged students[2]. Whether a student can get individual attention has long shaped mastery, which is the problem OpenAI frames the report around.

Internationally, a World Bank randomized study in Nigeria found that six weeks of teacher-guided, GPT-4-powered tutoring improved student performance on a combined assessment of English, AI knowledge, and digital skills. In Estonia, a nationwide ChatGPT Edu deployment reaches more than 20,000 students and 4,600 teachers, with the University of Tartu, Stanford, and OpenAI studying what happens in actual classrooms. The report notes that learning outcomes there have not yet been established[2].

What the Training Sessions Are Surfacing

The latter half of the report covers the AI Skills Jam sessions for K-12 educators that OpenAI Academy and the Walton Family Foundation have been holding this summer. Locations include Jonesboro, Georgia; Fairfax, Virginia; Orlando; San Bernardino, California; Phoenix; Salt Lake City; and Las Vegas. Participants work alongside OpenAI technical mentors on lesson planning, assessment data, and school policies, and receive a year of free access to ChatGPT Education[2].

Initial survey data showed that 93 percent of participants left with something they could use again, and 96 percent planned to apply what they learned within 30 days. The report also highlights the district of Christina Ordonez, technology department chair at a high school in Illinois, which extended training beyond teachers to librarians, athletic trainers, hall monitors, and bus drivers so that students hear consistent guidance from every adult at school[2].

OpenAI positions the lesson plans and differentiated materials produced at these sessions as evidence of where teachers lose time and where they hit friction with AI. The stated goal is a loop that feeds those observations back into product development rather than simply delivering training[2].

Summary

This report is the first consolidated look OpenAI has offered at how much ChatGPT is used beyond the school day. The scale of more than 460 million messages per week, and the Sunday-evening pattern, suggest learning support is drifting free of class hours. Whether that usage translates into academic gains depends on conditions, as the Turkish study cited in the report shows. All the figures come from OpenAI's own internal analysis rather than independent third-party verification, which is worth keeping in mind.

Source: https://openai.com/index/learning-never-stops

Source: https://cdn.openai.com/pdf/learning-never-stops-back-to-school-report-august-2026.pdf