What the Jams actually were
OpenAI Academy and the Walton Family Foundation ran free, day-long AI Skills Jams for K-12 educators across nine US communities over the summer and fall, drawing more than 1,600 teachers, administrators and district leaders. Sessions combined introductory and advanced ChatGPT training with guided activities, on-site mentors and dedicated building time, structured so attendees left with a working artifact rather than just a slide deck of tips they would forget by Monday.
The findings come from OpenAI's own publication on the program, titled From Curiosity to Classroom Outcomes, based on feedback from 802 of the attendees. That is roughly half the total participant base, a response rate solid enough to draw directional conclusions from, though it is still OpenAI measuring satisfaction with OpenAI's own event, a framing worth keeping firmly in mind when reading the headline numbers that follow and before anyone treats this as independent third-party research.
The numbers that matter
Ninety-two percent of respondents said they left with something reusable, 96 percent planned to use what they learned within 30 days, and 97 percent said they would recommend a Jam to a colleague. Of the 741 respondents who said they built something reusable, 29 percent produced lesson and classroom planning materials such as small-group schedules, assessments and pacing guides, 18 percent built skills, projects or apps, 16 percent created images or visuals, 13 percent developed templates or communications, and 10 percent built websites, with some projects spanning more than one category.
Those categories tell you where teachers see the fastest payoff: planning and prep, the unglamorous work that eats hours every week, beats flashier builds like apps and websites by a wide margin. That ordering should inform how any district designs its own internal AI training, since it confirms teachers gravitate toward tools that save time on tasks they already do, rather than tools that let them attempt something entirely new on top of an already full plate. A training program built around prep time first will likely see faster adoption than one that opens with ambitious app-building demos.
What teachers actually built
Francesca Summers, a fourth-grade teacher at Kemp Elementary School in Jonesboro, Georgia, used ChatGPT to adapt district materials into 30 to 40 minute lessons starting from learning objectives and student work, saying she starts with the standard rather than the AI tool itself. Karl Craddock, a chemistry and physics teacher at William Fremd High School in Palatine, Illinois, built simulated experiments and 3D models for concepts like magnetic fields, a use case that would have required specialized simulation software a few years ago.
Cesar Lopez, an administrator in San Bernardino City Unified School District, built an AI assistant that flags unanswered emails before the end of his day so he can review and respond himself, saying the goal is to offload tedious work and get more time back with colleagues and students. At Township High School District 211 in Illinois, which serves more than 12,000 students, an athletic trainer built a custom GPT used by staff across five schools to produce consistent session reports, a small but telling example of how far these tools now reach beyond the classroom proper.
Where the model breaks down
Nearly a third of participants wanted more examples and demonstrations, and 29 percent wanted more hands-on practice time, both pointing at the same underlying issue: a single day is not enough to move a novice from curiosity to confident independent use. OpenAI's own analysis recommends separate programming tracks for first-time and experienced users, with role- and subject-specific examples, plus more unstructured time to experiment and clearer starting guidance for people who arrive without a plan.
The published results are also explicit about their limits. They measure what participants intended to do and how they felt about the day, not whether the tools they built actually improved student outcomes or reduced workload over a semester. OpenAI excluded duplicate submissions from one city and responses from an unrelated event aimed at older adults, a reasonable cleaning step that still leaves the underlying measurement as self-reported intent rather than observed classroom behavior.
The case for running this in-house
The structure here, a single day combining direct instruction with protected building time and on-site mentors, is a template any district or corporate L&D team can replicate without OpenAI's name attached to it. The expensive part is not the content, which is now widely available, but the staffing: mentors circulating the room to unblock people who get stuck matter more to completion rates than the quality of the slides, based on what these numbers suggest about satisfaction and follow-through.
District and training leaders weighing whether to buy an off-the-shelf version of this program or build their own should price out the mentor-to-participant ratio first, since that is the variable OpenAI's design leaned on hardest and the one easiest to under-resource in a budget-constrained internal rollout. A session that skips the mentors to save money becomes a weaker program wearing the same slide deck, cheaper on paper and far less effective in the room.
What to measure next time
Any organization running a similar session should add a 60- and 90-day follow-up that asks whether the reusable artifact actually got used repeatedly, not just once, and whether it changed time spent on a specific task like lesson prep or grading. Intent-to-use numbers this high are encouraging, but they are also exactly the kind of metric that looks great in a launch announcement and fades fast without a second measurement point.
Pair that follow-up with a direct comparison against a control group that did not attend a Jam, something OpenAI's publication does not appear to include anywhere in its reporting. Without that comparison, a 92 percent satisfaction rate tells you the day was well run and well received by the people in the room. It does not yet tell you the day was worth the opportunity cost of pulling more than 1,600 educators out of their schools and classrooms to attend it, which is the number a superintendent actually has to defend to a budget committee.
