The one-day AI workshop has a shelf life of about three weeks. Here's what the research says actually changes teaching practice — and how to buy PD that does it.
Matthew Kelso, Ed.D · July 2026 · 5 min read
Picture the standard version. An August in-service day. A presenter demos a chatbot writing a lesson plan, the room is half amazed and half offended, everyone gets a handout of fifty prompts. By late September, four teachers are using any of it. Three of them were using it before the workshop.
This isn't an AI problem. The research on professional development has said the same thing for two decades: one-off, lecture-format training produces almost no durable change in classroom practice. What works is sustained, active, content-specific support with feedback loops — coaching, not assemblies. Recent studies on AI-focused PD land in exactly the same place. The format matters more than the topic.
A workshop on a new gradebook teaches a fixed tool. AI tools change monthly, and the skill being taught isn't a button sequence — it's judgment. When is the output good? What can't the tool be trusted with? What does a teacher do when a student's essay reads like a press release? Judgment doesn't transfer in a slideshow. It develops the way it always has: try it on your own class, get it wrong, talk to someone, adjust.
That loop needs time and it needs a person on the other end. The encouraging news from the last two years of research: coaching scales better than it used to, because some of the between-session support can now be carried by the tools themselves. A teacher stuck on Wednesday doesn't have to wait for the October follow-up.
Cohorts, not audiences. Eight to fifteen teachers who share a subject or grade band, meeting on a rhythm across a semester. Content-specific matters: what an English teacher needs from AI has almost nothing in common with what a chemistry teacher needs.
Classroom assignments between sessions. Every session ends with something to try before the next one. The session after opens with what happened. This single design choice separates PD that changes practice from PD that fills a requirement.
A named person for the stuck moments. Coach, lead teacher, instructional facilitator — the title doesn't matter. Somebody answers the Tuesday 10:15am question, because that's when the real questions happen.
Leaders in their own track. Principals who've never touched the tools can't observe AI-integrated lessons or coach the teachers using them. A building-leader track — how to evaluate, support, and scale what teachers are doing — is the difference between one enthusiastic classroom and a building-level shift.
Set the measures before the first session, and make them about practice, not satisfaction. Exit surveys measure catering. Better signals: the share of staff actively using AI in planning or feedback (not "aware of it" — using it), teacher hours returned per week, and pre/post confidence on the specific tasks the cohort trained on. Revisit quarterly and cut what isn't moving.
A fair benchmark from the field: a semester-long cohort should move active-use rates well past half of participants, where the standalone workshop typically strands you near a fifth. If your provider won't commit to measuring, that tells you what they expect the measurement to show.
This is the model behind our three K-12 tracks — teachers, building leaders, district operations — and it's why we don't sell single-day engagements without a follow-through plan attached. The day is the kickoff. The semester is the work.