🛝 The Safest Bets in Education Just Got Tested

Welcome to Playground Post, a bi-weekly newsletter that keeps education innovators ahead of what's next.

This week's reality check: The first randomized trial of AI in real classrooms found that giving teachers an AI assistant made classes less interesting to students. A Stanford analysis of 796 districts found closing schools didn't improve a single budget outcome. And in Texas, students who barely made the transfer cutoff went on to earn $7,000 less per year than the applicants who were rejected.

Data Gem

Even after historic post-pandemic learning declines, just 2% of Americans name education as the country's most pressing problem, according to Gallup polling covered by Chalkbeat.

Teachers Got an AI Assistant. Their Students Noticed.

The pitch for AI in teaching is time: lesson plans in seconds, feedback in minutes, hours back every week.

One of the first randomized controlled trials of AI in real classrooms just tested what that time costs.

Wharton's Alp Sungu, with University of Pennsylvania researchers including psychologist Angela Duckworth, followed 193 teachers and more than 2,800 middle and high school students at a private school chain in Turkey. 

Half the teachers were randomly given a ChatGPT-based teaching assistant customized to the national curriculum. 

Half taught as usual.

Over 10 weeks, teachers used the tool mostly for lecture notes, assignments, and exams.

Students whose teachers had the assistant rated their classes as less enjoyable, and less important than the control group did. 

The decline was modest, but largest for teachers who were already heavy AI users.

Overall achievement didn't change.

Then the researchers looked at the classrooms that were already struggling.

Among teachers whose students had lower marks before the experiment, achievement and confidence both fell, measured on externally administered exams that ruled out grading differences.

The teachers AI was supposed to help most were the ones it hurt.

"Teachers, just like students or coders, might be using AI as a crutch," Sungu said. "Instead of doing the actual work, they're using AI to delegate the task, and that lowers the quality of their teaching."

His theory: strong teachers treat AI output as a first draft and rework it. Weaker teachers use it as is. 

"When you start using AI-generated material, you're losing your personal voice," he said. "If everything is very uniform, it just becomes a bit more boring."

Two caveats. 

The study, released in June, is not yet peer-reviewed. And control-group teachers were free to use other AI tools, which Sungu says means the findings may understate the risk.

His conclusion isn't that AI ruins education. It's that access alone doesn't improve teaching, and right now access is what's being sold.

For education innovators, the study lands squarely on the industry's core sales pitch, because time saved turned out not to mean teaching improved. Efficacy-measurement tools that track learning outcomes rather than usage hours. And professional development aimed at the teachers who need guardrails most, since the trial suggests they're the least likely to add them on their own.

The Schools Closed. The Payroll Didn't.

More students graduated high school in 2025 than in any year in American history.

On the other side of that peak, enrollment is falling, and the default district response is to close underenrolled schools.

A Stanford researcher checked whether the default works.

Francis Pearman analyzed 796 California districts between 2011 and 2019, comparing districts that closed schools with statistically similar districts that didn't.

Closures cut enrollment by an average of 287 students per district.

They left funding deficits and the odds of balancing the budget unchanged.

The result held for districts in financial distress and for stable ones.

Spending fell by about $447 per student, but revenue fell at effectively the same rate, because California's funding formula pulls state money out as student counts drop, with no cushion.

And one number didn't move at all.

Closing schools had no effect on the total number of teachers and staff.

The buildings emptied. The payroll didn't. 

With more than three-fourths of school spending going to personnel, a closure that doesn't change staffing barely changes the budget.

There is a California asterisk. Strong union protections and state rules restricting what districts can do with surplus buildings boxed in local leaders, and the Education Next analysis warns that states with more staffing flexibility could see different results.

Which is the real lesson: a closure only saves money if the cost structure closes with it.

For education innovators, that lesson defines the closure-planning market a decade of projected enrollment decline is about to create. Scenario tools that model revenue loss next to savings, not just square footage. And facility-reuse platforms that convert empty buildings into teacher housing or community services instead of line-item liabilities. Districts will keep closing schools. The product is making the closure count.

Barely Making the Transfer Cutoff Cost Students $7,000 a Year

Transferring up is the mobility script of American higher education: start at a community college, move to a four-year school, earn more. 

More than a third of all college students transfer.

A new working paper tested the script on the students at its margins.

Researchers publishing through Brown University's Annenberg Institute tracked Texas students from 2000 to 2024: more than 50,000 who applied to transfer from two-year to four-year colleges, and about 22,000 who applied from four-year colleges to the state's flagships.

Students who barely cleared the GPA cutoff for admission, typically around 2.0, against students just below it who were rejected.

The admitted students earned about $7,000 less per year than the rejected ones.

The gap was still statistically significant six to 10 years later.

For the students on the margin, getting in cost them money.

"It is very surprising. It's not what I was expecting to find," said Lois Miller, an economics professor at the University of South Carolina.

Marginal transfers arrive at junior standing, expected to keep pace in coursework their first college may never have aligned. 

At flagships, many couldn't keep their majors: the business school was more selective than the university itself, and transferred credits often counted as general credit rather than toward a degree.

"They're basically not able to major in business anymore," Miller said. "If they had stayed at their less selective four-year college and majored in business, they would have eventually gotten a higher-paying job."

One caveat: the study compared only students at the cutoff. It says nothing about transfers with stronger records.

But the infrastructure failure underneath it is universal. Transfer students lose an average of 43% of their earned credits, according to the Government Accountability Office.

The systemic response is starting to assemble. More than 160 institutions joined a new consortium led by accreditor SACSCOC to build common transfer pathways for general education courses, the first effort of its kind led by an accreditor. 

The framework gets built this year, used by fall 2027, and measured by summer 2028.

For education innovators, the space between the paper and the consortium is the market. Transfer-fit and ROI advising tools that tell a 2.1-GPA student whether the move pays before they make it. Credit-equivalency engines that map courses to major requirements, not just general credit. And transfer-onboarding platforms, since colleges build first-year support for freshmen and hand arriving juniors almost nothing. The consortium will take two years to measure itself. Students are transferring now.

⚡️More Quick Hits

This week in education:

• Google's AI search tools rated "Unacceptable" for kids after a 2,600-search audit — Common Sense Media found 75% of children use AI for search, about a third of cited sources lacked meaningful editorial accountability, and repeated history questions produced different answers 43% of the time

• 19 states now admit students to college before they apply — Georgia sends automatic-admission letters to more than 130,000 seniors covering up to 45 institutions, and freshman enrollment rose 7% at technical colleges and 4% at four-year schools after the program launched

• 5.7 million students with disabilities now spend at least 40% of their day in general education — a GAO analysis found mainstream placement up roughly 25% since 2012-13, while districts say limited funding for paraeducators and special educators keeps inclusion understaffed

• Indiana set the nation's first standards for 4-day school week waivers — districts need an A accountability grade, $45,000 minimum teacher pay, free Friday programming, and transportation for students who opt out, while national data puts typical four-day savings at just 0.4% to 2.5%

To stay up-to-date on all things education innovation, visit us at playgroundpost.com.