Design a First AI Session People Can Finish

A beginner session works best when its goal fits into a sentence. “Turn these fictional project updates into a meeting agenda, then check the draft” is a goal people can attempt. “Learn AI” leaves the facilitator with too many choices and the participants with no clear way to know when they have finished.

The thirty-minute plan below is a suggested format for an adult workplace group. It uses invented administrative material and assumes the organization has approved the tool being demonstrated. Adjust the pace and access arrangements for the people attending. The aim is one completed exercise and an explanation of the final edits, not a claim that half an hour makes anyone proficient.

Choose an output people already understand

A meeting agenda is a useful practice object because participants can judge its basic structure without learning a new profession. Prepare a short set of fictional updates: the venue decision is still open, the draft invitation needs review and one team member has a question about the schedule. State that the meeting lasts thirty minutes and that no deadline has been agreed for sending invitations.

Keep the material short enough to reread during the exercise. Include one missing detail that the draft should leave unresolved. This gives the review a purpose beyond fixing punctuation. Participants can notice whether the assistant invents a deadline or treats a suggestion as a decision.

Write the completion criteria beside the input. The agenda should contain the three topics, allocate a total of thirty minutes and mark the unresolved decision clearly. Anyone should be able to compare the output with those criteria. Avoid a task whose quality can only be judged by the facilitator's personal taste.

Set up access before the session

Check the tool on the same kind of account participants will use. Confirm that the expected input method works and that any required feature is available. Ask the organizer to handle account permissions before the meeting. Spending the opening ten minutes on password resets changes the lesson, even if the rest of the material is excellent.

Prepare a fallback containing the same fictional input and a saved sample answer. A participant without access can still inspect the answer, identify a problem and propose a revision. Make it clear when the group is viewing a prepared example rather than fresh output. If the service is unavailable for everyone, use that fallback to teach the review step and schedule hands-on access separately.

Offer the instructions in a readable document and read the essential steps aloud. Let people choose to work in pairs where appropriate. Check whether anyone needs a different way to view or enter the material. These are preparation choices, not assumptions about who is capable of learning the tool.

Minutes zero to five: explain the boundary

Open with the task and show what finishing looks like. Say that all material is fictional and ask participants to stay with it during the exercise. Explain which tool is approved for this session and where questions about real workplace data should go. Avoid inviting people to paste their own customer records as an impromptu demonstration.

UNESCO's guidance on generative AI in education emphasizes human agency and privacy. Those considerations are relevant here even though this plan is for a workplace group. Participants should understand the exercise, have a way to ask questions and retain responsibility for reviewing the output.

Ask the group to name one sign of a usable agenda. Take two or three answers, then point to the written criteria. This brief exchange establishes the object of the lesson without becoming a general debate about every possible use of AI.

Minutes five to ten: model one attempt

Show the sample input and a plain instruction: “Create a thirty-minute agenda from these updates. Use only the supplied facts. Label open questions and do not invent deadlines.” Read the result slowly enough for people to compare it with the source. Narrate one check, such as adding the allocated minutes or finding the still-open venue decision.

If the output is flawless for the exercise, show the prepared imperfect example as well. Explain that it is a teaching sample. The point is to demonstrate a review move, not to produce an entertaining failure on demand. If the live output contains an unexpected problem, acknowledge it and decide whether it fits the lesson before following it into a long detour.

Keep the demonstration short. Participants need time to operate the tool or work through the fallback themselves. A session filled entirely with the facilitator's screen does not show whether the instructions can be followed independently.

Minutes ten to eighteen: let people try

Give participants the same input and ask them to make one attempt. People working in pairs can have one person enter the prompt while the other checks the result against the source. Switch roles for the revision. That gives both people a defined contribution instead of leaving one as a silent observer.

As you circulate, ask what the person is trying to change. Avoid taking over the keyboard at the first hesitation. If several people are stuck at the same point, pause the room and clarify the instruction for everyone. Keep a note of that obstacle so the next version of the session can address it earlier.

Allow one purposeful revision. A participant might ask the tool to shorten the agenda descriptions or label an unresolved item more clearly. Have them keep the original answer so they can compare the versions. Repeatedly generating new answers without a reason makes it harder to explain what improved.

Minutes eighteen to twenty-five: inspect and explain

Ask everyone to find one sentence or agenda item that deserves a check. They should point to the source fact that supports it, identify information that is missing or explain why the wording needs a human edit. The final artifact can include a brief note describing that decision.

NIST's Generative AI Profile identifies confidently incorrect content as a risk. The review step makes that concern concrete without telling beginners that every output is wrong. They are learning to examine particular claims and commitments in the context of a specific task.

Invite two participants to explain their edits. If someone says only that the second version “sounds better,” ask which requirement it now meets more clearly. If the group finds no errors, ask which checks support that conclusion. An acceptable result still benefits from an account of how it was assessed.

Minutes twenty-five to thirty: transfer the habit

Close with a short teach-back. Each person should be able to describe the input, the instruction, one check and the human decision that followed. Ask for a possible next practice task using approved material. Capture questions that need the organization's policy owner rather than trying to settle them casually in the room.

After the session, review whether participants completed the task and could explain their changes. Collect feedback about pace and clarity separately. Enjoyment, confidence and demonstrated task completion are different kinds of feedback; none should be silently substituted for the others.

For the next session, change one dimension. You might use longer updates, add an ambiguity or ask participants to write their own completion criteria. Keep the exercise small enough that review remains visible. A repeatable habit of setting a goal, inspecting an answer and explaining an edit is a practical place for beginners to start.