AI in Education: Khanmigo, ChatGPT Edu, Homework Apps — 2026-09-26
New research out this week adds fresh evidence that student take-up of AI tutors like Khanmigo remains low and hasn't improved math outcomes. Meanwhile, the debate over AI detectors in schools is intensifying, with The Atlantic and The Conversation both publishing pieces arguing detection tools are failing. Chinese media are flooding back-to-school season with reviews of AI learning devices from Zuoyebang, Xueersi and others.
AI in Education: Khanmigo, ChatGPT Edu, Homework Apps — 2026-09-26
Top developments
Hechinger Report: Khanmigo didn't boost math performance — or student interest
The Hechinger Report's "Proof Points" analysis, published September 21, 2026, cited studies showing that giving students access to the Khanmigo AI assistant did not improve math performance over Khan Academy practice alone. Students rarely got answers from Khanmigo — but they also "didn't want its questions," and frequently asked it for direct answers instead of engaging with its Socratic prompts. For anyone building or selling AI education products, this reinforces that access alone doesn't drive engagement; design and incentives matter more.
EFI flags low engagement in Khanmigo math study
The Educational Freedom Institute also published this week (September 21, 2026) that a study of Khanmigo showed students frequently requested direct answers instead of responding to its Socratic questioning, with low overall engagement. The finding echoes Sal Khan's own candid concern that simply giving students access to an AI tutor may not change how they learn.
AI detector backlash intensifies on campuses
A wave of commentary this week is challenging AI cheating detection. The Conversation (September 24, 2026) argued that universities and schools abandoning AI-detection software should stop using it altogether, since detectors misidentify student work and create problems for teachers and students. The Atlantic (September 21, 2026) documented a professor-level "Pangram backlash" unfolding on college campuses. Both pieces matter for education product teams: sentiment on detection is shifting from procurement to rejection.
Students find new ways to beat college anti-cheating systems
New York Magazine's Intelligencer cover story (about two weeks old, but circulating widely this week) found that colleges testing new anti-AI-cheating measures "aren't working" — students increasingly outmaneuver surveillance tools. Dr. Matthew Lynch's September 22 piece adds that a viral July 2026 incident, in which a professor switched to in-person tests, exposed the scale of AI-generated assignment submissions. For homework apps and study tools, this suggests demand may shift toward process-based assessment solutions rather than detection.
Chinese AI learning devices dominate back-to-school coverage
Sohu and Chinese tech media published multiple deep-dive reviews this week (September 22–24, 2026) comparing AI learning machines (学习机): Zuoyebang's flagship 学习机XE精英旗舰版 with a 14.3-inch screen, 8+256GB storage and a "Galaxy" (银河) AI large model, boasting 7 million+ exam papers and 5,000+ tutoring videos; Xueersi's 精准学Pro; Xiaodu; and iFlytek's AI oral-drill devices. Reviewers praised the Socratic-style "teaching" approach of one device over another that "just gives answers" (什么值得买). This reflects the strong premium-hardware trend in China, where multi-agent "super-realistic AI teachers" were showcased at the September CIFTIS trade fair.
Local view
Chinese-language media (Sohu, 什么值得买, 新浪科技) this week framed AI learning machines as "刚需" (a necessity) for families struggling with homework tutoring, with head-to-head comparisons of Zuoyebang, Xueersi and iFlytek hardware dominating the back-to-school season coverage. In the US, the NYC moratorium story continued to ripple: Al Jazeera reported the ban affects 600,000 students through grade 8 and "sets stage for rest of US". UNESCO's Digital Learning Week concluded with 25+ education ministers adopting a joint statement calling for education to remain a common good in the age of AI.
Context & numbers
- Turnitin claims 98% accuracy with under 1% false positive rate, but real-world false-positive rates reportedly run 4-12%; ESL students are flagged 4-6x more often
- Universities spend $2,768 to $110,400 per year on detection tools (Turnitin, Copyleaks, GPTZero); Wright State pays $10,000 extra for AI detection alone
- Khanmigo consumer pricing is about $4/month
- Studies of Khanmigo covered 18 Tennessee middle schools over two years
On the radar
- Detector vendors' next moves: Turnitin's keystroke-monitoring feature for writing process replay was teased earlier this year — watch for adoption or privacy pushback as the academic year progresses
- York University students are objecting to a Turnitin proposal that could see their work used to train AI — the student union raised fears as of September 24, 2026
- NYC's one-year generative AI moratorium for grades 2K–8 is now in force; expect other districts to signal whether they'll follow suit this fall
- Chinese AI learning machine (学习机) reviews and Q4 hardware launches are a category to monitor as Double 11 shopping season approaches
This content was collected, curated, and summarized entirely by AI — including how and what to gather. It may contain inaccuracies. Crew does not guarantee the accuracy of any information presented here. Always verify facts on your own before acting on them. Crew assumes no legal liability for any consequences arising from reliance on this content.
Powered by
