Build a small, useful AI learning programme around work your team actually does. This editable plan combines short lessons, role-based practice and observed skills checks. It is for workplace learners, managers and facilitators; no coding is required.
Format: four weekly 60-minute sessions, plus a suggested 15 minutes of practice between sessions. Adapt the pace to your team. You will leave with: a shared safe-use checklist, a reusable prompt, a checked work sample and an individual next-step plan.
Before you start
- Choose an AI tool and account your organisation permits for this activity. Check its current data controls and your internal policy before using it.
- Agree which information can be used. Start with the fictional workshop notes in the companion exercise, not client records, employee details or confidential work.
- Choose a facilitator and a human reviewer. One person may do both in a small team.
- Check access and accessibility in advance. Provide a paired demonstration or paper-based review activity if someone cannot use the tool; do not confuse access barriers with a skills gap.
- Pick one low-risk task for the pilot: an internal summary, a draft email based on invented details, or a checklist made from a public document.
Define the outcome
Complete these fields before inviting the team:
| Planning field | Your decision |
|---|---|
| Team and facilitator | [Write here] |
| Dates, location and approved tool | [Write here] |
| One task we want to practise | [Write here] |
| Who will use the output | [Write here] |
| Information allowed / not allowed | [Write here] |
| Human reviewer and sign-off point | [Write here] |
| Accessibility or access support | [Write here] |
| Evidence we will collect | [Write here] |
Example learning objective: Given fictional meeting notes, a participant can produce a short brief that preserves confirmed facts, marks missing information as unknown, and identifies at least one claim that needs checking before sharing.
This is an example objective, not a promise of performance. Replace it with the observable task you need.
Week 1 — Set boundaries and write a clear request
Learning objective: Explain when this tool is appropriate and write a request with a task, context, constraints and output format.
60-minute agenda: 10 minutes on permitted use and limitations; 10 minutes of facilitator demonstration; 25 minutes of paired practice; 10 minutes reviewing outputs; 5 minutes recording a next step.
Activity: Use the companion meeting-notes exercise. First request a summary without much direction, then specify an audience, word limit, required headings and an instruction not to invent missing details. Compare the two outputs rather than assuming the longer prompt is better.
Evidence: Save the revised prompt and a note explaining one change. Check that each participant can identify a piece of information they must not paste into the tool.
Between sessions: Improve one prompt using fictional or approved public information. Record what changed and whether it helped.
Week 2 — Check the output against the source
Learning objective: Separate supported facts, assumptions and unknowns; correct an output using the original source.
60-minute agenda: 10 minutes on checking claims; 10 minutes on the worked example; 25 minutes of individual review; 10 minutes of peer comparison; 5 minutes recording a next step.
Activity: Run the companion 45-minute exercise inside the session. Use the remaining time for local policy and questions. If access is unavailable, review the deliberately flawed sample draft on paper.
Evidence: Keep a marked-up draft with unsupported claims identified, a corrected brief, and an explanation of what still needs human confirmation. A fluent answer alone is not sufficient evidence.
Between sessions: Use the same checking checklist on another fictional or public-source task. Do not send an AI-generated draft externally as part of this practice.
Week 3 — Practise a role-specific task
Learning objective: Apply the same request-and-review process to one relevant, low-risk task without changing its safety boundaries.
60-minute agenda: 10 minutes choosing tasks; 10 minutes setting acceptance criteria; 25 minutes drafting and revising; 10 minutes peer review; 5 minutes recording a next step.
Choose one track, then adapt it:
| Role or team | Practice task | What the reviewer checks |
|---|---|---|
| Administration | Turn fictional meeting notes into actions | Owners and dates match the notes; unknowns stay visible |
| Customer support | Draft a reply using an invented customer question and approved public policy | No invented refunds, guarantees or policy exceptions |
| Marketing | Turn a fictional product brief into two audience-specific drafts | Product claims match the brief; tone and audience fit |
| Operations | Create a checklist from a public process description | Steps preserve source order and name human approval points |
| People and learning | Draft an agenda for an imaginary internal learning session | Activities match objectives and offer accessible participation |
Evidence: Retain the source, prompt, draft, reviewer comments and revised output. Keep personal or confidential information out of the learning record. Do not use this track to evaluate real employees or automate employment decisions.
Between sessions: Write a reusable prompt and a separate review checklist for your task. Test it on a second permitted example.
Week 4 — Demonstrate, reflect and choose the next step
Learning objective: Complete a comparable task, explain the checks performed, and identify where further support is needed.
60-minute agenda: 10 minutes of recap; 25 minutes for a fresh fictional task; 15 minutes of evidence review; 10 minutes setting the next practice goal.
Activity: Use the human AI-skills rubric with a new example of similar difficulty. Ask each participant to explain their decisions. Compare evidence with the starting task, not with another person's confidence or typing speed.
Evidence: Record a level for each observed skill, one concrete example, support provided, and one practice action. Mark unobserved skills as “not observed”; do not treat them as zero.
Follow-up: Agree a review date and an owner. More independent use is a separate manager/policy decision, not an automatic consequence of completing training.
A reusable prompt structure
Copy and adapt the following. These are instructions to a tool, not a guarantee that it will follow them.
Task: [What you want done]. Audience: [Who will read it]. Source material: [Permitted information only]. Constraints: Use only the supplied facts; label unknowns; do not infer dates, owners, permissions or numerical claims. Output: [Headings, length and format]. After the draft, list claims the human reviewer should verify.
Review quality, not just completion
Use these checks at the start and end of the programme:
- Did the person choose permitted information and recognise when to stop or ask for help?
- Did the request describe the task and required output clearly?
- Did the person check important claims against the source rather than asking the model whether it was correct?
- Did they improve the draft using specific feedback?
- Is there a named human owner before anything is shared or used?
If you measure task time, define the task consistently, include checking and correction time, and pair any time comparison with a quality check. A faster unchecked draft is not a successful outcome.
Copyable follow-up worksheet
| Field | Notes |
|---|---|
| Participant or voluntary identifier | [Write here] |
| Task and source used | [Write here] |
| One skill demonstrated, with evidence | [Write here] |
| One skill needing practice | [Write here] |
| Access or support needed | [Write here] |
| Next practice action | [Write here] |
| Review date and support person | [Write here] |
Collect only the learning information you need, explain who can see it and agree when it will be deleted. Do not put sensitive learning records into an unapproved AI tool.
Continue with the companion resources
Use the 45-minute AI workshop exercise for a complete fictional practice task and the human AI-skills assessment rubric to record observable learning needs.
For facilitated delivery, explore Caelan Huntress's existing workshop curriculum. This template is a reusable learning resource, not a substitute for your organisation's policies or an accreditation programme.