The best practice for live AI use is a solo mock session where you run your actual planned prompts, watch the outputs, and make notes on what to adjust — the same way you’d rehearse any part of a workshop.
Why Generic Practice Doesn’t Work
Many educators try to prepare for live AI use by taking a course, reading tutorials, or experimenting with random prompts. That builds general familiarity, but it doesn’t prepare you for the specific moment in your specific workshop when you ask AI to do something specific. The surprise that throws you off-screen is almost always a prompt that behaved differently than you expected — not a gap in your general AI knowledge.
The fix is specificity: practice with the exact prompts, in the exact context, that you plan to use live.
How to Run a Solo Mock Session
Set aside 30 minutes before your workshop. Open the same AI tool you’ll use on-screen — Claude, ChatGPT, or whatever you’ve chosen. Walk through your session agenda and identify every moment where you plan to use AI. For each one, type the prompt you intend to use live and read the output carefully.
Ask yourself three questions: Is this output accurate enough to share? Is it the right length and format? Does it match the level of my audience? If the answer to any of these is no, adjust the prompt and run it again. A few iterations in private is worth far more than a perfect-seeming plan you’ve never actually tested.
Also note where the AI surprised you in a good way — outputs that were better than expected. Those moments are often worth building into the session deliberately.
What This Means for Educators
Think of this like a tech check before a Zoom call. You wouldn’t run a client presentation without testing your screen share and audio. AI is the same category of tool — it needs a pre-flight check. The good news is that unlike Zoom audio issues, AI prompt surprises are highly correctable in advance. You can test and refine until the output is predictable.
The Simple Rule
Never show a prompt to your students that you haven’t already run at least once yourself. That single rule eliminates the vast majority of on-screen surprises.