Issue 02 · Full Free PreviewAfter SchoolWho Guides AI Use? Students use AI. How does learning support catch up?
2026 Issue 02No. 002
Issue 02 · Full Free Preview
Who Teaches Students to Use AI After School?
Students use AI. How does learning support catch up?
A US survey shows broad academic AI use while most teens still turn first to themselves or other people when stuck. Teen assistants, public curricula, and home-learning devices support different steps. We examine additional product value, institutional guidance, and the evidence needed for lasting trust and payment.
Observation period
to
Judgment
Start with the task where students need help
Fact check
Event dates and source dates were checked separately
ISSUE SUMMARY
Students use AI for assignments while retaining several ways to seek help. Around August 18, teen assistants, public curricula, and institutional partnerships advanced different parts of the process: immediate responses, understanding, and learning support. The industry question is where students need help and how products, schools, and families can connect it.
COVER STORY
Who teaches students to use AI after school?
When a student gets stuck, whom do they ask next? That question shapes how an AI education product enters learning: occasional explanations or sustained tutoring, help preparing a question for a teacher or help completing an entire assignment inside an app. This week’s survey, product launches, and curriculum developments provide a clearer basis for choosing.
A US survey released by Common Sense Media on August 18 found that 70% of respondents use AI for schoolwork. Yet only 10% turn first to an AI app or tool when stuck, while 65% first try themselves or ask a teacher, parent, or friend.[6] Another 12% first search the internet with an AI-generated summary available; the 10% figure refers specifically to AI apps or tools. Contact with AI is broad, and existing help-seeking relationships remain important.
NORC surveyed 1,017 Americans aged 13–17 about their own behavior. The bounded industry implication is that AI participates in learning in this sample but is far from most students’ first source of help. China and other markets need their own evidence; these proportions cannot be carried across borders.
Use also changes with the task. Among schoolwork AI users, 63% obtain answers in some form, while 77% also use it to brainstorm, check work, or receive feedback. These multiple-choice groups overlap.[6] A student may brainstorm thoughtfully and later request an answer under deadline pressure. Segmenting students simply by AI proficiency can miss changes caused by task and time pressure.
OpenAI launched ChatGPT for Teens the same day, explicitly targeting moments outside class when help is unavailable and combining learning guidance with teen protections in personal accounts.[4] The industry significance is that educational features are appearing inside general tools students may already use. Specialist products need to show where they provide additional help.
If a student can already obtain an explanation from a general assistant, a specialist service might identify a specific misconception, follow the school’s curriculum, or help a teacher see recurring difficulties. These are proposed differentiators, not established winners. Actual use and purchasing must establish willingness to pay; a feature count cannot do so.
Different objectives create tension. A student may want to finish quickly, a parent may want less homework supervision, and a teacher needs to know what the child understands independently. Faster completion may not meet all these needs. A supplier must decide which steps remain the learner’s responsibility and how unresolved difficulties return to a trusted person.
Public curricula play another role. Jiangsu launched its coverage project this week, and Vietnam issued a national framework.[1][2] These could help students verify, explain, and take responsibility even when they change products. Better judgment might also raise what schools and families expect from tools, depending on implementation.
Our central judgment is that opportunities for AI learning services should be sought along specific help-seeking processes. Platforms can respond promptly, schools teach evaluation and transfer, and families help recognize difficulties. Services that connect these roles, reduce real burdens, and demonstrate independent learning have a stronger case for lasting trust.
Using AI, turning to it first, and paying for AI tutoring are three behaviors that require separate explanations.
WEEK IN REVIEW
This week: products, curricula, and school partnerships
On August 17, Zuoyebang demonstrated multimodal upgrades to its AI Super Teacher and a product range including Z100 and X70 Ultra.[11] This was an iteration of home-learning services; Z100 had already been publicly launched in July. Independent results showing how explanation, diagnosis, and planning improve learning were unavailable for this issue.
On August 18, Jiangsu launched its province-wide AI literacy project with a requirement to start classes in autumn. Vietnam’s Decision 2422 took effect, covering grades 1–12. OpenAI launched its teen experience and CodeAI partnership, and Common Sense Media released its schoolwork survey.[1][2][3][4][5][6]
That day also saw GSENet’s Global Smart School Initiative launched at a conference co-hosted in Beijing by Beijing Normal University and UNESCO IITE. Roanoke College confirmed free access to Google AI training and career certificates for students, faculty, and staff.[7][8][9] One is a cooperation initiative; the other has reached access provision. Neither supplies comparable student outcomes.
On August 20, the US Department of Education issued classroom technology guidance emphasizing instructional value, independent evidence, educator judgment, and transparency for parents.[13] It is federal guidance, with state and local decision-making retained. The evidence obtained for this issue cannot establish comparable effectiveness across these products and curricula.
POLICY
Public curricula: building judgment across tools
The US survey identifies a specific gap in explanation. Of all respondents, 44% recalled teachers discussing when AI use is allowed, 30% safe use, 27% how it works, and 26% whether its information is trustworthy.[6] This describes reported classroom conversations, not a common international timetable. It suggests a need to teach both rules and understanding.
Jiangsu first expanded teaching resources. Its announcement records 94,000 participations in the launch and 9.03 million viewing instances across summer training, plus free curriculum resources and rural support through visits, cloud classrooms, and technical specialists.[1] These are instances, not unique trained teachers or evidence of teaching proficiency.
Where free resources are available, curriculum services need a clearer paid proposition. Our inference is that selecting age-appropriate activities, helping teachers prepare, and using student work to improve the next lesson may address implementation needs better than another generic video collection. A business depends on schools experiencing those problems, the service saving work, and continued adoption.
Vietnam sets a more detailed common foundation: 12 core periods per class each academic year across grades 1–12, with optional extensions. Four strands cover human agency, ethics, technology and applications, and system design. The framework is platform-independent and uses processes and multiple evidence sources without separate AI exams or scores.[2][3] It gives competence across tools a place in the curriculum.
For content and platform teams, fluency in one product is therefore only part of the value. Detecting errors and explaining evidence after switching models fits this goal more closely. Process assessment can itself become a burden when teachers lack time. Services should help obtain a small amount of interpretable learning evidence; more forms do not necessarily help.
US guidance asks which learning problem a product solves, for whom, when, for how long, and with what evidence, alongside ongoing review and removal of persistently ineffective tools.[13] Consumer products seeking school adoption need implementation and outcome evidence between user approval and institutional responsibility for instruction.
PEERS
The different roles of assistants, learning devices, and schools
ChatGPT for Teens offers Study Mode and reminders that detect apparent homework shortcuts and redirect students to Study Mode. Study Hours can make Study Mode the default at chosen times. Age prediction indicating under 18, or self-reporting age 13–17, triggers the teen experience. Age protections are built in; study schedules and some controls after parent linking require configuration.[4] The announcement establishes a design, while age accuracy, global coverage, and protection outcomes still need independent examination.
Specialist tutors should also reassess their comparison point. Among US schoolwork AI users surveyed, 68% used general-purpose tools and 14% used AI tutors.[6] Categories overlap and cannot establish revenue shares. Another question-answering interface may offer insufficient reason to switch. Curriculum fit, following up misconceptions, and teacher communication need to become tangible differences.
The CodeAI partnership adds educational support through an advisory council, Hour of AI, a student Builders Challenge, and support for the free year-long high school AI Foundations course and career content.[5] Our interpretation is that a curriculum partner provides access to educator experience and may build educational trust. Classroom adoption and user growth remain unreported.
Zuoyebang combines diagnosis, explanation, planning, and dedicated devices into a home-learning service.[11] Compared with a personal-account assistant, its proposition is closer to organizing sustained subject learning. Willingness to pay for curriculum fit, hardware, and the support experience needs separate testing. Teacher-like interaction does not establish that human tutoring has been replaced.
Roanoke’s partnership makes resources free, leaving value dependent on who helps learners select, complete, and apply them.[9] The Global Smart School Initiative includes professional development and family-school-community cooperation, with plans for evaluation, consulting, and certification.[8] To help institutional choices, such services need published criteria, interests, and results. A certification label alone cannot demonstrate effectiveness.
OPPORTUNITY
Specialist services need to demonstrate additional value
Begin with the help-seeking task. A student who has drafted an answer wants their reasoning checked; one unable to start needs a first step; a parent unfamiliar with the curriculum needs useful questions to ask. These can require different products. Complete one form of support and test whether learners return before relying on a broad promise of learning companionship.
Learning claims also need to confront mixed experiences. In the survey, 66% of schoolwork AI users felt it helped understanding, 39% felt they missed learning, and 38% reported fewer original ideas.[6] These perceptions can coexist and establish no causal effect. Satisfaction alone can miss the case in which work is finished but the learner still cannot do it.
Institutions can place a short independent task after support and ask learners which suggestions they adopted and why. This can test understanding and show teachers where to resume instruction. Reducing repeated diagnosis gives a service a more concrete adoption case. Records should fit the task and avoid unnecessary monitoring work.
Families may build trust by understanding when to intervene. A service can show where a child needs support and which tasks still require independent effort, alongside completion counts. The trade-off is that visible difficulty may reduce short-term feelings of success while making longer-term value easier to test. Payment still needs comparison with free alternatives and family time costs.
Schools also need to consider what happens beyond their devices. Among schoolwork AI users, 44% encountered school network or device blocking; of that group, 59% switched to personal devices.[6] This US sample suggests pairing access management with assignment design and family communication. It does not establish that every restriction fails or justify ignoring school rules.
This issue’s judgment
Earn lasting trust by helping at the step where a learner gets stuck.A prompt answer is one part. Additional curriculum fit, follow-up, and human support need to be tested through workload and independent learning.
NEXT WEEK
Next: independent learning and sustained use
If general platforms reliably improve independent performance and transfer while reducing human support costs, specialist tutors need sharper evidence of additional value. If learners bypass guidance, take answers, or fail to detect mistakes, curriculum fit and human support may matter more. Clicks, conversation time, and self-reported understanding cannot independently settle the comparison.
For Jiangsu and Vietnam, watch classroom activities, teacher preparation time, and how students evaluate outputs. If free resources sustain instruction at low cost, additional curriculum services may have limited room. Persistent gaps in local activities, feedback, and professional support would establish more specific needs.
For platforms, watch age-detection errors, default and optional settings, actual support through linked parents, and external tests. Roanoke needs evidence beyond access, while the smart-school initiative needs inspectable evaluation rules. These are questions posed as of August 23; later results must not be written back into the week.
SOURCES & LIMITS
Sources, evidence limits, and revision history
This issue covers August 17–23 and was rewritten on September 16, 2026, from the original week’s vantage point. It retains the major curriculum, product, survey, school partnership, and federal guidance events. The original pool had 81 records, 32 concerning China and 49 overseas, including duplicates. Detailed selection and exclusion records remain in the evidence document.
The US survey ran April 30–May 14, combined weighted probability and nonprobability samples, and reports an overall margin of error of about ±4.3 percentage points. It describes American teens’ reported behavior and perceptions, not learning causality or students in China or Vietnam. Product positioning, trust, and service opportunities are our interpretations. Comparable revenue, retention, and long-term learning results were unavailable.
Editorial checks
Implementation: Jiangsu launched a project and required autumn classes; Vietnam issued a formal grades 1–12 framework. Actual coverage and learning still require verification.
Historical Indian developments: Bharat Bodhan and Bharat EduAI Stack belong to India and date to February 2026. Karnataka’s education plans were disclosed August 15–16. Neither belongs in this week’s events.[12]
US Department of Education: The named August 20 remarks are from Assistant Secretary Kirsten Baesler. Linda McMahon’s five AI principles are background from 2025.
Sources and outcomes: The OpenAI launch URL has been corrected. The AP cross-check remains listed but was unavailable in this recheck; official originals support the product facts. Zuoyebang coverage describes a launch and does not independently substantiate teacher equivalence or a learning-gain percentage.
: The second issue was first published. Core facts were checked against original sources, together with geographic scope, actors, event types, dates, and titles.
: Added the US Department of Education's classroom-technology guidance, clarified the named official and the 2025 background, and excluded the Indian AI tutor commentary proposal. : The title and prose were rewritten for clarity while preserving the original length, structure, facts, and sources.
: Rewritten from the original week’s vantage point, adding counterevidence on help-seeking, roles for products and public curricula, distinct decisions, and future tests. Corrected the OpenAI source URL; original publication date retained. Contents and corresponding section headings were also edited to reduce consecutive questions.
