The Promptathon Playbook: How to Run a One-Day AI Event That Actually Sticks

A promptathon is a one-day, on-site team event where employees learn to use generative AI on real work problems. Done well, it is one of the most effective enablement formats we run — far more effective than a recorded course, because the practice happens with their actual work, in front of their actual colleagues, with help available in the room.

Done poorly, it is a fun afternoon that produces nothing the next week.

This playbook covers the difference. It is the scope we run as an AdoptionLab.AI Promptathon, and the format we recommend to teams running it themselves.

What a promptathon is — and is not

A promptathon is: a structured event where small teams pick a real, current work problem and solve it (or partially solve it) using generative AI tools, with facilitation and expert help available throughout the day. Outputs are real artifacts — a draft policy, a prototype email sequence, a redesigned workflow, a working prompt template — that the team plans to use after the event.

A promptathon is not: a hackathon to invent a new AI product. It is not a recorded training session. It is not a competition with prizes (you can have prizes, but if the prize is the point, the day will tilt toward novelty over usefulness).

The most common mistake we see is treating the promptathon as a top-of-funnel awareness event. It can do that, but it does it badly. What it does well is move people from “I have used ChatGPT once” to “I know how to apply AI to the specific work I do” in a single day.

Pre-work (two to three weeks out)

Promptathons live or die in the prep. The day itself is structured improvisation — people learn fastest when they are working on something that already matters to them — but the structure has to be set up in advance.

Pick the right problems. Each team needs a single, real work problem that is small enough to make progress on in 4–6 hours but large enough that progress matters. Bad: “use AI to be more productive.” Good: “draft the first version of our 2026 sales playbook for the SMB segment.” Better: “redesign our customer onboarding email sequence — currently 8 emails over 30 days, mostly written in 2022 — using AI to draft and our team to revise.”

We typically work with sponsors three weeks ahead to surface 8 to 12 candidate problems and then pick 4 to 6. The number depends on team size — figure roughly one problem per team of 4 to 6 people.

Pick the right tools. A promptathon should use one or two AI tools, not five. The point is depth, not breadth. We pick whichever AI assistant the company has already deployed (Microsoft Copilot, ChatGPT Enterprise, Claude through your business agreement, or your CRM-native AI), plus one second tool only if there’s a genuine reason — usually because the work problem benefits from an image or document tool.

Set the data rules in advance. What data is allowed in the prompts? Anonymized customer data? Internal financials? Real names? This needs to be answered before the day, in writing, and reviewed by your governance owner. If your policy isn’t clear, see our AI Policy Template for Mid-Market Companies for a starting structure. Walking into the event without clear data rules turns half the day into a debate.

Brief the participants. A 30-minute pre-call one week out covers what to bring, what tool we will use, what data is in or out, and what the team’s specific problem is. People show up ready to work, not ready to figure out what they’re doing.

The day itself

A typical schedule for a 4 to 6 hour promptathon:

Hour 0:00 to 0:30 — Welcome and frame. The sponsor opens. The facilitator sets the rules. Each team introduces their problem in 60 seconds. Critical: nobody pitches a “vision,” they describe a real, current pain. The framing here matters because the rest of the day is built on it.

Hour 0:30 to 1:00 — Tool fluency. Quick, structured walkthrough of the AI tool. Not a demo of features — a working demonstration of three patterns: drafting, transformation (rewrite this email in this tone), and analysis (find the patterns in this document). Twenty minutes of demonstration, ten minutes of supervised practice on a low-stakes example. The point is to make sure everyone in the room can actually open the tool and use it before they are asked to solve their team’s problem with it.

Hour 1:00 to 3:30 — Working session. Teams work on their problems with their assigned AI tool. Facilitators rotate. The energy in the room is the strongest signal — if it drops at hour two, that’s usually a sign that one or two teams have hit a wall and need a re-scope. Have the facilitators check in with each team at the 30, 90, and 150 minute marks. Snacks help.

Hour 3:30 to 4:30 — Showcase. Each team presents their work in 5 to 7 minutes. The format we use: what was the problem, what did we try, what worked, what didn’t, what’s next. Not a polished pitch. The audience asks one question per team. The whole showcase runs about an hour for 5 to 6 teams.

Hour 4:30 to 5:00 — Debrief and commitments. Facilitator pulls the threads together: what patterns showed up across teams, what tools or prompts everyone wants to keep, what governance or policy questions came up. Each team commits to one specific next step they will take in the next two weeks, written on a card, photographed, sent to the sponsor.

The whole day runs five hours of clock time. Lunch and breaks bring it to six. We do not recommend stretching past that — energy drops sharply, and a tighter day is more memorable.

The debrief and follow-through (this is where most promptathons fail)

The day ends with everyone energized. Two weeks later, three of the six teams have done nothing with what they built. This is the gap between a fun afternoon and an actual capability shift, and closing it is the single highest-leverage thing you can do.

We schedule two follow-on touchpoints into every promptathon engagement:

Two-hour debrief, one week after. Facilitator and sponsor meet with each team lead individually for 15 minutes. What progress, what stalled, what’s blocking. This is the conversation that catches stalled work before momentum dies. The aggregate is then summarized in a one-page note to the broader sponsor group.

Two-hour planning session, three weeks after. Sponsors plus team leads. Convert the day’s outputs into next steps that have an owner, a deadline, and a metric. Some of these become small pilots. Some become process changes. Some become prompt templates that get added to the team’s shared playbook. This is also where you decide which problems graduated to needing structured pilots — those are candidates for AI Opportunity Sprint or Pilot Sprint work.

Without these touchpoints, the energy from the day evaporates. With them, the same energy compounds.

Why this format works

A promptathon works because it inverts the usual training pattern. Instead of teaching the abstract first and asking people to apply it later, it puts people in front of their real work and helps them figure out where AI fits, in real time. The learning is sticky because it is grounded in the actual problems they came in with.

It also works because of social proof. When a colleague walks across the room and shows you what their team got working in 90 minutes, that lands differently than a recorded video. People leave the day having seen four or five concrete examples of “people like me, doing work like mine, using AI like this.” That is a more durable mental model than any presentation.

Finally, it works because the room is mixed. Promptathons we run intentionally include skeptics. The skeptics are not converted by being told AI is great. They are sometimes converted by sitting next to a peer they respect, watching the peer do something useful in real time, and saying “huh — show me how you did that.”

When to run one

Promptathons are best as a kickoff to a larger engagement. The day produces concrete artifacts and a spike of organizational energy; the surrounding engagement converts both into durable change. Standalone promptathons work, but the half-life is shorter than people expect.

If you are starting an AI Pilot Sprint or scaling work through an AI Change Management Sprint, running a promptathon at week one usually shortens the rest of the engagement by two to three weeks because the team is already fluent enough to participate in the build.

If you want help facilitating one — including the pre-work, the day itself, the debrief, and the planning session — that scope is exactly the Promptathon engagement. Four to six weeks elapsed time, 18 to 28 total hours, $12,000 to $25,000.

Or take this playbook and run one yourself. Most of the value is in the pre-work and the follow-through, and both are within reach for any team willing to take them seriously.