This week's read
Real
The Education Department just told districts to stop counting screen time and start counting evidence.
The read
On August 20 the US Department of Education issued a Dear Colleague letter on education technology and screen time, signed by Kirsten Baesler, assistant secretary in the Office of Elementary and Secondary Education. It is guidance, not enforcement, and it says so. The core argument is that recreational technology and instructional technology are different things, and that "the public policy debate should not be framed as a choice between technology and no technology, nor should every screen be viewed as serving the same purpose." It asks states and districts to judge a tool by demonstrated learning outcomes rather than by minutes of exposure, to look for evidence of effectiveness at procurement and at renewal, to build a process for checking whether a tool met its goal, and to be willing to remove tools that repeatedly fail. It endorses outcomes-based contracting by name, and asks vendors to publish outside evaluations and disclose their products' limitations. Where it breaks: it carries no force of law, it names no consequence, and it does not reconcile itself with the administration's own health-side messaging. A May Surgeon General's advisory urged cutting youth screen use, and the HHS secretary has called digital learning tools addictive and urged scrapping one-to-one device programs. Two federal positions, pointing opposite directions, in the same year.
5
Questions the letter says every
ed tech product should answer
0
Force of law. It binds nobody
and names no penalty
2
Federal positions on screens this
year, pointing opposite ways
US Department of Education, Dear Colleague letter, 2026-08-20. The five questions: what learning problem, when, for whom, how long, and what evidence.
What it means for your schools
Nobody has to do anything because of this letter, and that is exactly why it is useful to you. If your board is heading toward a minutes-per-day cap, you now have a federal document arguing that minutes are the wrong measurement and that the right one is whether the tool did the thing it was bought to do. Let it cut both ways, because it does. The same letter that protects a good tool from a blunt cap also says you should be willing to pull a tool that keeps failing, and asks you to say in advance what evidence would count. Most districts cannot answer that for the software they already own. The five questions are free, they are short, and they now live in a federal document you can hand to a skeptical board member or a vendor without it reading as your personal opinion. Ask them about the next renewal on your desk rather than the next purchase, because the renewals are where the money actually is and where nobody is looking.
US Department of Education →
The rundown
Real
Student newsrooms are policing AI with typing-speed analysis, and the kids who are not cheating are writing worse on purpose
The 74 interviewed three high school journalists on August 20, and the practices they describe are the story. At the Berkeley High Jacket, editors keep strict control of documents, look through writers' search histories, and analyze typing speed and word choice against previous work. Staffers have been let go over suspected AI use. School-issued Chromebooks block ChatGPT and Claude outright, and any attempt to reach them is flagged to a teacher through GoGuardian. Cid Frydman, a rising senior who says she does not use AI to write, describes deliberately slimming down her writing and holding back detail so it will not read as too polished. Tova Weiss, an editor at the Youth Environmental Press Team, watches peers "adjust their writing voice" and add typos on purpose to prove authorship, and calls it "absolutely devastating to watch how many people are intentionally damaging or diminishing the quality and authenticity of their work." Frydman also describes the reputational mechanics: get caught once and you cannot get a recommendation letter from any teacher, and teachers tell colleagues to watch that student. This is three students in three self-selected newsrooms reporting their own impressions, not a survey, and one of them says on background that ordinary copying is still far more common than AI. If your building suspects AI use, write down what happens next: who decides, on what evidence, and what the student can do about it. The 74 →
Hype
"Learn, unlearn, relearn" is a 1970 futurist slogan, and the people quoting it have the author wrong
Audrey Watters published "A Technology of Unlearning" on August 22, tracing the slogan that anchors a great deal of AI-era professional development back to its source. The line usually rendered as "the illiterate of the 21st century will not be those who cannot read and write, but those who cannot learn, unlearn, and relearn" is routinely attributed to Alvin Toffler. The attribution is wrong twice over. The sentence originates with the psychologist Herbert Gerjuoy, whom Toffler was quoting, and Toffler's own line in Future Shock (1970) is the more modest "By instructing students how to learn, unlearn, and relearn, a powerful new dimension can be added to education." Her argument is that a fifty-six-year-old futurist trope is being recycled as though it were an insight about generative AI. Take the citation fix as fact and the broader thesis as argument, because Watters is an avowed critic with a long-held position. She also has no product and no AI consulting practice, which is the reason she is on this newsletter's source list at all. When the case for a large and expensive change rests on a quotation nobody in the room has checked, the case is thinner than it sounds. Second Breakfast →
A read on trending research
Watch
Half the kids handed an AI tutor, with class time set aside to use it, never opened it.
Two randomized controlled trials from the Stanford SCALE Initiative, published through the Annenberg Institute at Brown: 355 students in grades 1 to 5 across two districts, every one of them given an AI literacy platform and dedicated session time, twice a week for thirty minutes. The randomized contrast was whether a student used it alone or alongside an in-person tutor whose job was engagement and accountability rather than instruction. The two cards below are the whole finding.
What was scheduled
The dose tied to benefit
30 minutes per week
Twice weekly sessions
Class time blocked out
Platform provided to every student
What the district paid for
What actually happened
Same students, same weeks
Nearly half never opened it at all
2 to 5 minutes per week among those who did
Human support added 1 to 4 minutes
No improvement in reading achievement
Access was never the constraint
Robinson, Gormley, Trindade Ribeiro & Loeb, "Access is Not Enough: Human Support Improves Engagement with AI Tutoring," EdWorkingPaper 26-1451, June 2026. Research Integrity Gate 78/100, Watch, with a major flag on generalizability: one product, two districts, grades 1 to 5. The human-support lift is 71 to 80 percent in relative terms, which is arithmetically true and describes moving from almost nothing to slightly more than almost nothing. The achievement null is uninformative rather than damning, because the delivered dose never approached the threshold. Full scored research list on
the research page.
What it means for your schools
Every AI tutoring pitch you will hear this year is priced and argued on access. Seats, licenses, coverage, how many students can reach it. This is the study that went and measured the next thing along, which is whether they touch it, and the answer was two to five minutes a week from the half who opened it at all. These were not defiant teenagers finding a workaround. They were first through fifth graders with the time blocked out for them. Then the researchers tried the expensive fix, a human in the room whose entire job was to get them to engage, and bought between one and four more minutes. Sit with how much that costs per minute. The move here is not to avoid AI tutors. It is to stop accepting access as evidence of anything, from a vendor or from yourself. Before the next renewal, ask for minutes of actual student use per week, per student, not logins and not licenses, and ask what the vendor believes the dose needs to be for the outcome they are promising you. If they cannot tell you the dose, they have not run this study. If they can tell you the dose and not your usage, you are about to find out you are already below it.
Annenberg Institute →
Apply it
Every story in this short issue is about counting the wrong thing. The Education Department told districts that minutes of screen exposure is the wrong number and demonstrated learning is the right one. The student journalists showed what happens when a school counts typing speed and detector scores instead: honest kids writing worse on purpose to stay under suspicion, which shows up in your data as compliance. Watters found a slogan that has been driving professional development for years while nobody checked who said it or what he actually said. And the strongest study of the week counted the number almost nobody asks a vendor for, minutes of actual student use, and found two to five a week from the half of children who opened the thing at all. Four different rooms, four confident numbers, and in each one the number everybody was watching was not the number that mattered.
Your one move this weekPick one AI or ed tech tool your district already pays for and find out how many minutes students actually used it last month. Not seats, not logins, not licenses. Minutes. You will either get a number that surprises you or you will discover nobody can produce one, and both of those are the answer. It costs one email, and it is the exact question the federal letter now backs you up on asking.
That's it for this week. See you next Thursday.
Katie
Brought to you by Planning Partner
A studio for tomorrow's lesson.
Quality AI lesson planning that lifts instructional quality and cuts teacher workload at the same time. It closes the gap between your curriculum and student-ready instruction, and gives leaders visibility into how your instructional model is actually being implemented, without piling more onto teachers.
See Planning Partner →
The Margin · A weekly AI-in-education digest for education leaders
Compiled by AI against a bar a human set
My Planning Partner · @myplanningpartner