What changed this week
for AI at home and at school
Hey Big Thinker. University of Notre Dame researchers tested two AI detectors on published research abstracts, and light AI edits got flagged far more often than disguised AI. Almost all the disguised rewrites slipped through. So what should happen after a flag, at home and at school?
The paper didn't test student essays, and the authors say detectors may behave differently on classroom writing. Below: what the paper found, three moves for parents, teachers and school leaders, and a 20-minute activity where kids catch an AI claim that can't hold up.
IN THIS ISSUE
01 News: what happens after an AI flag
02 Practical tips: three ways to see the thinking
03 Activity: become an Internet fact-checker
NEWS
Your school needs a plan for what happens after an AI flag
WHAT HAPPENED
University of Notre Dame researchers posted a preprint that tested 642 published research abstracts, each with AI rewrites, on Pangram 3.2 and GPTZero, then disguised those with a humanizer tool. Fewer than 4% of those rewrites stayed flagged, while real abstracts with light AI edits (a stand-in for AI help that guidelines allow) got flagged 38% to 80% of the time, depending on the detector and how old the abstracts were.
WHY IT MATTERS
|
|
Honest AI editing, the authors conclude, carries a higher risk of sanction than disguising AI, and a detector score shouldn't stand alone as misconduct evidence. More than 40% of surveyed 6th- to 12th-grade teachers used these tools in the 2024-25 school year, according to a Center for Democracy and Technology poll reported by NPR. When Tanner Jones, English department chair at Basalt High School in Colorado, suspects AI use, he starts with questions like "Take me through your process here..." A kid who can answer has proof that holds up. The paper tested no student essays, and the authors say classroom writing may differ. It used one humanizer, and Pangram 4 (released in July) wasn't tested, so these results may not apply to it.
Sources: Notre Dame paper · Edutopia
THOUGHTS FROM THE EXPERTS
We think detectors were never the answer, and this paper gives us one more reason. The best proof of learning we know is a kid who can explain their own work to a parent or a teacher. We'd like every school leader to build that conversation into what happens after a flag.
— Justin & Will
PRACTICAL TIPS
Three ways to see the thinking
For parents, teachers and school leaders, with a note on who does what.
01
Have kids walk you through one piece of work
Parents can pick something from this week's school work, and teachers can do the same in a quick chat at a desk. Ask for the whole story, start to finish, from the first idea to the last edit. Try: "Walk me through how you got this idea."
02
Ask how your school checks for AI use
Parents can ask a teacher how they check this year, and what happens after a flag. School leaders can ask three teachers and compare, then write down the next step after a flag so every teacher handles it the same way. Try: "Do you grade drafts or just the final piece?"
03
Agree on a "show your thinking" habit
Parents and teachers can set this up with kids this week, at home or in class. Kids keep their messy first draft. If a question ever comes up, they've got something to point to. Try: "Before you ask AI, write one sentence in your own words."
ACTIVITY
Become an Internet Fact-Checker
AGES 11–14 20 MIN YOU NEED: AN AI CHAT TOOL
Kids test an AI's evidence against the source and catch a claim it can't back up. The skill is checking AI's answers outside the chat, then explaining how you know.
HOW TO DO IT
1
Think like a checker (5 min). Pick a claim an app ad might make, like "learn twice as fast." Write your own list of the proof you'd want, or type: "An app says kids [learn twice as fast]. What would that have to mean, and what proof would it need?"
2
Ask for the proof (10 min). Type: "Give me your best evidence that kids [learn twice as fast]. For each piece, tell me who was tested and give me a link." Now step outside the chat and open one link yourself, because you can't catch this from inside it. A link that won't open counts as a catch. So does a source that says something smaller than the AI did (one college class, say). If the source checks out, open another. Then go back and type: "I opened your source and found [what you found]. Which claim can't you back up?"
3
Explain your catch (5 min). Type: "Ask me two questions, one at a time, about what I found in [the source I opened]." Answer in your own words and save the chat, because it's your proof of process. Then close the screen and take 60 seconds to tell a grown-up or partner how you caught it. In the classroom: partners swap claims and do the 60-second explain-back.
TALK ABOUT IT
Which AI claim sounded surest, and what did the source say? If someone asked how you made your last piece of school work, what could you show them?
POLL
Does your school check for AI use?
Tap an answer to vote. We'll share the results next week.
READ MORE
Go deeper
Big Thinkers · What one district's draft AI policy says, and the questions kids should be able to answer.
Big Thinkers · One private school's two-hour day and its self-reported results, the kind of promise this week's activity puts to the test.
Big Thinkers · What students say about AI at school, before your school writes its next rule.
SHARE BIG THINKERS
Know a parent, teacher or school leader who'd find this useful?
Big Thinkers is free. Every week: one story, practical tips, and an activity to do with kids. Pass it on.
Or just forward this email.
