OpenAI says 79% of students use AI for homework. Only 47% say anyone showed them how.
A new ThinkYoung survey in OpenAI's EMEA education report finds a 32-point gap between students who use AI for schoolwork and students who were taught to use it well.
OpenAI published a report on November 10 with a number that’s hard to argue with either way. ThinkYoung, a youth research group, surveyed students and teachers for OpenAI and found that 79% of students said they’d used AI for homework or assignments. Only 47% said anyone had shown them how.
That’s a 32-point gap between using a tool and being taught to use it. This report is written by Tom Duff Gordon, OpenAI’s VP of global affairs for EMEA, and Leah Belsky, its VP of education. They call this the “AI guidance gap.” It’s the clearest finding in an otherwise soft piece of corporate research.
The gap, in numbers
| What ThinkYoung asked students | Share who said yes |
|---|---|
| Used AI for homework or assignments | 79% |
| Were shown how to use it | 47% |
| Gap between the two | 32 points |
OpenAI doesn’t say how many students ThinkYoung surveyed, which countries they came from, or whether the sample is representative of anything beyond itself.
What OpenAI is actually doing about it
The report isn’t just a survey writeup. It’s a case for OpenAI’s own EMEA education partnerships, named as Estonia, Slovakia, Greece and Italy. The partnerships share one goal: get frontier AI tools into schools, build literacy programs with governments, then actually measure what happens. Five themes organize the lessons: clear goals, teacher prep, local partnerships, critical literacy. The fifth is credible evidence.
On the evidence point, the report is honest about its own limits. It says “teacher adoption and reported time savings offer encouraging early signals,” not proof of anything, and that understanding where AI actually helps learning “requires sustained evaluation.” It also cites OECD education director Andreas Schleicher’s comments on PISA data, framing the core idea as AI making teachers more important, not less, because judgment and guidance are what AI alone doesn’t supply.
What’s missing from the report
No independent outlet has reported on this yet. It published only hours ago as of this writing, so there’s no outside check on the ThinkYoung numbers or on how the four country partnerships are actually going on the ground.
Those same four countries get named again, but with no hard numbers for any of them. No student counts. No before-and-after scores. Just “early experiences.” A full report, titled “Responsible AI in Education,” is linked as a PDF from OpenAI’s own site, and that’s where any real detail on the partnerships would have to live.
My read
I think the 79/47 split is the one number here worth remembering, and the rest is OpenAI positioning itself as the company that noticed the problem it also profits from. That’s not necessarily wrong. Someone has to supply AI literacy programs if schools are going to have them, and OpenAI naming specific countries it’s working with is more concrete than most AI policy posts.
My read is that the gap itself is the real story, not OpenAI’s response to it: a lot of students are already using these tools for graded work with nobody checking whether they understand what they’re doing.
This lands the same week Google is tightening the opposite lever, turning Search and Gemini off by default for K-12 accounts in its own classroom AI controls unless a school opts in. OpenAI wants schools to teach AI use properly; Google’s default is to switch it off until someone decides otherwise. Both responses assume the same thing: that schools currently aren’t managing this well, which is probably true.
Related: for another look at how OpenAI talks about who actually uses its products, see our post on OpenAI’s free-ChatGPT-Plus offer for veterans.
Note on sources
OpenAI’s own report page is the primary source for the 79%/47% ThinkYoung figures, the EMEA partnership countries, the five themes and the Schleicher/PISA reference; I read that page in full. I did not open the full “Responsible AI in Education” PDF or ThinkYoung’s underlying survey data, so sample size and methodology are unverified and I’ve said so above. No independent outlet has covered this report as of publication, so there’s nothing to cross-check it against yet.