← Back to Blog
🎓 AI in Education

AI in Education Digest — August 21, 2026

Two of the biggest companies on earth spent this week trying to own the moment a student sits down to revise — Google with a free year of its top AI plan, OpenAI with a teen-safe ChatGPT built around a Study Mode. And in the same 72 hours, two pieces of independent research landed that complicate the pitch: one found AI models hand over the answer almost every time unless you explicitly stop them, the other found most US teens have never been taught what AI actually is. Seven stories that matter.

1. Google gives college students a free year of AI as Gemini and Search fill up with study tools

Source: EdTech Innovation Hub · 21 August 2026

Google is handing eligible US college students twelve months of Google AI Pro at no cost — a plan it prices at $19.99 a month — while students in more than 140 other markets can claim a free year of the cheaper Google AI Plus tier. Both offers have to be redeemed by 31 December 2026, both require a valid payment method at sign-up, and both roll into a paid subscription automatically unless cancelled. Google is separately offering some students a bundle of AI Pro and YouTube Premium for $8.99 in the US, against $35.98 for the two bought separately.

The giveaway sits on top of a much larger product push. Gemini now has a dedicated student hub with study notebooks that read a student's own uploaded lecture notes, diagnostic quizzes designed to surface gaps, flashcards and a progress dashboard; Deep Research is moving into Gemini Live so a multi-step research task can run in the background. Search is gaining free practice quizzes globally in English — including for the SAT, ACT, AP, GRE, LSAT, MCAT, NEET and JEE, drawing on material from The Princeton Review, Careers360 and PhysicsWallah — plus AI-generated interactive simulations, file creation, and a Lens flow where a student photographs their working and gets coached through the mistake. Nick Fox, Google's SVP of Knowledge & Information, said the priority was to build products that steer students "toward true subject matter understanding and skill development rather than cheating or shortcuts."

Why it matters: A free year removes the last real barrier between a teenager and a frontier model — which makes how students use these tools, rather than whether they can afford them, the only question left.

Video angle: "Google Just Gave Students a FREE Year of AI — Here's What's Actually In It"

2. 70% of US teens now use AI for schoolwork — but only 27% have been taught what it is

Source: EdTech Innovation Hub, reporting Common Sense Media / NORC · 21 August 2026

Common Sense Media's Teens in the AI Era report, based on an NORC survey of 1,017 US teenagers aged 13–17 conducted between 30 April and 14 May 2026 across all 50 states, puts schoolwork AI use at 70%. Among those users, 77% say they use it for help that stops short of answers — brainstorming, checking work, getting feedback — while 63% say they get an answer in some form. Crucially, those are often the same students: the honest picture is not two camps but one behaviour that slides between the two depending on the deadline.

The classroom side has not kept up. Only 27% of teens say a teacher has ever explained what AI is or how it works, 44% have discussed when it should and should not be used, and just 26% have been taught how to judge whether what it tells them is accurate. Blocking is not working either: 44% of AI users had a tool blocked on a school network or device, and of those, 59% simply switched to a personal device.

Why it matters: Filters move the behaviour off the school network rather than out of the student's life — the only durable lever left is teaching, and most teens report they haven't had it.

Video angle: "70% of Teens Use AI for Homework. Only 27% Have Been Taught What It Is."

3. AI models handed over the answer in 97% of unprompted tutoring tests

Source: EdTech Innovation Hub, reporting Comprendo · 18 August 2026

Comprendo, an applied research firm that evaluates AI in education, ran 779 simulated eighth-grade algebra tutoring conversations across six frontier models — GPT-5.6 Terra, Claude Sonnet 5, Grok 4.5, DeepSeek V4 Pro, Kimi K3 and Qwen3.8 Max — with simulated students carrying misconceptions drawn from documented mathematics education research. In the 298 conversations run without any tutoring instruction, the models simply gave the student the answer 97% of the time, and not one conversation met Comprendo's full teaching criteria. Founder Trevor Selby summarised it bluntly: "Zero successful tutoring outcomes in 298 tries."

Adding an explicit tutoring prompt changed things, but exposed a trade-off rather than a fix. Prompted models diagnosed the planted misconception in 88% to 100% of relevant conversations, yet strict "clean win" rates — diagnosing the error, making no false claims, and leaving the student to finish the reasoning — ranged only from 28% to 54%, with DeepSeek V4 Pro highest and Claude Sonnet 5 lowest. Models that diagnosed best tended to do more of the student's thinking for them. Accuracy was not the problem: just three false statements across all 779 conversations. Cost predicted nothing — the cheapest model, at around $0.002 per conversation, scored highest. Comprendo is clear on the limits: these were general-purpose models, not commercial tutoring products, and no real students or classrooms were involved.

Why it matters: "Guide, don't give" is a prompt, not a property — and if the default behaviour is to hand over the answer, every study tool shipped this week is one careless prompt away from being a homework machine.

Video angle: "AI Tutors Gave the Answer 97% of the Time — Unless You Say This"

4. OpenAI launches ChatGPT for Teens with Study Mode, Study Hours and parental controls

Source: AP News, TechCrunch and Engadget, via EdTech Innovation Hub · 18 August 2026

OpenAI began rolling out ChatGPT for Teens on 18 August — not a separate app, but a default experience applied automatically when the company's age declaration and usage-pattern prediction place an account under 18. Lauren Jonas, OpenAI's head of youth and families, told Engadget there is "no need for a teen to create a new account or change anything." The education centrepiece is Study Mode: guiding questions that ask what the student already knows, step-by-step working instead of finished answers, built-in quizzes and visualisations, and homework reminders that nudge a teen toward it when the prompt looks like a request for answers rather than understanding.

Alongside it sit default blocks on self-harm, violence, eating disorders, explicit content, romantic roleplay and claims of sentience, plus opt-in family tools requiring both parent and teen to agree — Quiet Hours, Study Hours where Study Mode becomes the default, and safety notifications reviewed by OpenAI staff before a parent is alerted. Parents do not read chat logs by default. Global rollout runs over roughly two weeks. TechCrunch's caveat is the honest one: teens are practised at working around parental controls, enforcement here is behavioural rather than architectural, and none of it has been stress-tested at scale yet.

Why it matters: This is the first time a frontier lab has made "learn it" rather than "get it done" the default setting for under-18s — if it holds, it reshapes what a lot of teenagers do at 10pm the night before a deadline.

Video angle: "ChatGPT Just Changed for Every Teenager — What Study Mode Actually Does"

5. A four-week AI revision trial produced significant GCSE science gains in all three subjects

Source: EdTech Innovation Hub, reporting WhatWorked Education / Medly · 20 August 2026

An independent randomised controlled trial evaluated by WhatWorked Education put Year 9 and Year 10 students in mainstream English secondary schools through four weeks of AI-supported science revision. Both groups got 30 minutes a week; the control used their usual self-directed resources. Of 929 students who sat the pre-test, 644 completed the post-test — 332 using Medly, 312 in the control. The overall effect size was 0.33 and statistically significant, breaking down as 0.52 for Biology, 0.32 for Chemistry and 0.31 for Physics, with an adjusted mean post-test score of 13.95 against 11.39.

The dose-response pattern is the interesting part: each extra question answered was associated with roughly 0.18 additional marks, and exploratory effect sizes climbed to 0.46 for students doing 10 or more questions, 0.57 at 20 or more and 0.95 at 30 or more. There was no significant difference in effect between Pupil Premium and non-Pupil Premium pupils. Co-founder Dr Kavi Samra was careful to note this "wasn't a comparison with pupils doing nothing" — the control group was also revising. Roughly 31% attrition and a process evaluation drawing on only 6 of 39 teacher trials are real limitations, and a larger replication study is planned for next academic year.

Why it matters: This is one of the first properly controlled UK trials of AI revision to report subject-level gains — and the dose-response curve suggests the active ingredient is retrieval practice volume, not the AI itself.

Video angle: "An AI Revision App Just Beat Normal Revision in a Real GCSE Trial — By How Much?"

6. William & Mary launches an Applied AI major that requires no coding at all

Source: EdTech Innovation Hub · 20 August 2026

William & Mary has introduced a 30-credit Applied Artificial Intelligence major, leading to a BA in Interdisciplinary Studies and designed to be completed across four to five semesters through the School of Computing, Data Sciences & Physics. No previous coding, statistics or advanced mathematics experience is required, and the university describes it as the first Applied AI major at a public university in Virginia. Core requirements include Introduction to Data Science, Ethics in Data Science and Designing AI Agents, plus a capstone.

The framing is deliberate: the degree is about directing and evaluating AI systems rather than building them. "Students will learn AI by putting it to work," said Dan Runfola, Chair of the Data Science Department, while Dean Doug Schmidt said the major should prepare students "not only to use AI, but to help shape how it is adopted." It joins an existing AI minor and an AI & Ethics minor announced earlier this month. One footnote the university itself volunteered: its communications team used ChatGPT to help produce the announcement article, then reviewed and edited it before publishing.

Why it matters: If the valuable skill is judgement about AI rather than the ability to build it, the degree that teaches it stops belonging to computer science departments — and the applicant pool widens enormously.

Video angle: "This University Just Launched an AI Degree With No Coding Required"

7. Arizona Western College becomes the first in its region with a full AI integration plan

Source: KYMA · 19 August 2026

Arizona Western College has announced what it calls an "intentional, coordinated approach to integrate AI across all learning and operational models" — the first college in south-west Arizona to publish one rather than experiment piecemeal. The plan covers professional development for faculty and staff, certificate and degree programmes in AI and machine learning, community workshops and non-credit courses, advanced tooling and training through partnerships including Google, and a physical AI Learning & Innovation Lab as a hub for workforce development.

The rationale is regional rather than academic. AWC points to AI's impact across agriculture, healthcare, military and defence, government and manufacturing in the area, and frames the new Regional Center for Emerging Technology as an attempt to "get ahead of the curve." President Dr Reetika Dhawan said the aim is for students, staff, businesses and community members to "feel confident and supported as technology continues to evolve," while Dr Shadi Kilani argued the region's competitiveness depends on turning emerging technologies into "real opportunities for people, businesses, and communities."

Why it matters: Community colleges reach the adults whose jobs are changing right now — a coordinated plan at that level moves more people, faster, than another elite university AI institute.

Video angle: "A Community College Just Out-Planned the Ivy League on AI"

🎬 Best story for today's video

Google Just Gave Students a FREE Year of AI (What's Actually In It)

It's the highest-CTR story of the seven for one obvious reason — it's free, it's concrete, and any student watching can go and claim it today. But it doesn't collapse into a product demo, because the Comprendo research from three days earlier gives the video a genuine second act: free access to a tool that defaults to handing over the answer is not the same thing as free learning. That tension carries retention well past the first thirty seconds.

Google just handed college students a free year of its top AI plan and quietly rebuilt Search and Gemini around studying. Here's exactly what's in the box, the one catch nobody's reading — and why a study published three days ago says free access might be the least important part of this.

AI in education Google Gemini free AI for students Google AI Pro ChatGPT for Teens AI study tools AI tutoring GCSE revision A-Level revision edtech news 2026

Get this digest in your inbox every morning — the AI-in-education stories that actually matter, in five minutes.

Subscribe to the newsletter

📚 Turn today's AI news into an actual study system

Revision Lab has the study systems, timetables and past-paper strategies I use with my own students — built for the way you actually revise, not the way apps assume you do.

Visit Revision Lab →

Building your AI study stack? Browse top-rated revision and study-skills guides on Amazon. As an Amazon Associate I earn from qualifying purchases.