03 — Resources

Designing AI Training That Sticks

A working checklist for HR and L&D leaders building AI training inside their organizations. Most AI training is lecture-heavy, vocabulary-focused, and forgotten within a week. This is how to build the kind that changes what people actually do on Monday.

"If a participant can't tell you what they're going to do differently this week, your training was information transfer — not training."

A — Why most AI training fails

Three patterns that kill the work.

B — Six standards

The rules a good AI workshop has to meet.

These are the standards used to audit every workshop on this site. If you're designing your own AI training, hold it to the same bar.

1. At least 50% of time is active learning.

Pair & share, hands-on tool work, drafting activities, discussions. Lecture blocks longer than 20 minutes get flagged for revision.

2. Every concept gets a practical exercise.

If you teach prompt structure, attendees write a prompt. If you teach AI policy, attendees draft a policy section. No concept is lecture-only.

3. Learning objectives use measurable verbs.

"Understand," "know," "be aware of," and "appreciate" are not measurable. Use Bloom's taxonomy verbs — apply, analyze, evaluate, create — that a third party could watch and verify.

4. Discussion questions are open-ended only.

No yes/no questions. No rhetorical questions. No "is AI policy important?" Use "what's one scenario you could see happening at your company without an AI policy?"

5. Pre-work is realistic.

Under 30 minutes. Specific, useful, and connected to the session — not assigned reading nobody does. If participants don't do the pre-work, the session should still work.

6. Application planning is required, not optional.

End every session with 5–10 minutes for participants to write: one thing I'll start doing this week, one thing I'll stop, one question I still have. Without it, training is information transfer.

C — Writing AI learning objectives

From vocabulary to behavior.

The single most common failure in AI training is a learning objective written as "participants will understand generative AI." There's no way to assess that. Bloom's taxonomy gives you measurable verbs that change what the training has to do.

Remember
Verbs: define, list, identify, name, recall, recognize
Example: Participants will list the four most common AI security risks in workplace settings.
Understand
Verbs: explain, describe, summarize, compare, contrast, illustrate
Example: Participants will explain the difference between AI hallucination and AI bias, with examples from their own industry.
Apply
Verbs: apply, use, demonstrate, execute, implement, perform
Example: Participants will apply the nine-step prompt framework to draft a working prompt for a real task from their current job.
Analyze
Verbs: analyze, differentiate, examine, investigate, deconstruct
Example: Participants will analyze an AI-generated draft and identify three accuracy concerns before publication.
Evaluate
Verbs: evaluate, judge, critique, justify, assess, recommend, prioritize
Example: Participants will evaluate three AI tools against their company's data privacy requirements and recommend one for adoption.
Create
Verbs: design, construct, develop, formulate, produce, build
Example: Participants will develop a v1 AI policy section tailored to their organization's data sensitivity and tool approval needs.

Rewrites — before and after

Before

Participants will understand AI policy.

After

Participants will draft three sections of an AI policy (purpose & scope, approved tools, data privacy) using the worksheet template.

Before

Participants will know how to use ChatGPT.

After

Participants will apply the nine-step prompt framework to produce one workplace-ready output (draft email, meeting summary, or process document).

Before

Participants will be aware of AI risks.

After

Participants will identify three AI risks specific to their role and recommend one immediate safeguard for each.

D — Activity formats that work

Six formats that actually move people.

These are the formats used across the talks on this site — every one of them maps to a Bloom's level above Remember.

Pair & share (Understand)

Two-minute partner conversations on a specific question. Surfaces what's already happening in attendees' organizations. Good for opening a session.

Hands-on tool time (Apply)

Participants open ChatGPT, NotebookLM, or whatever tool the session covers, and produce something real in 10–15 minutes. Non-negotiable for any AI session over an hour.

Spot-the-flaw exercises (Analyze)

Show an AI-generated artifact with intentional errors — a brief with fabricated citations, a draft with hallucinated statistics. Participants find what's wrong. Builds the review reflex.

Hard-case discussion (Evaluate)

Small groups pick from 2–3 messy scenarios with no clean answer ("an employee included client financial data in a ChatGPT prompt — what do you do?"). Surfaces exactly the decisions a policy needs to make.

Pitch role-play (Apply / Evaluate)

Pairs use a fill-in-the-blank planner to practice pitching a new AI tool to a skeptical leader. One plays the proposer, one plays the boss, then switch. Builds confidence in advocacy.

Draft-a-policy workshop (Create)

Small groups use a worksheet to draft three policy sections in 15 minutes. Each group shares their strongest decision and toughest unresolved question. Closes a policy session with a working artifact.

E — Module structure

A 90-minute workshop in shape.

A working timing template. Adjust ratios for shorter or longer sessions, but never drop below 50% active.

Time Section Mode
0:00 – 0:08Opening + objective previewLecture
0:08 – 0:18Pair & share openerActive
0:18 – 0:35Core concept + cautionary talesLecture + discussion
0:35 – 0:60Hands-on exercise or drafting activityActive
0:60 – 0:78Share-back + hard-case discussionActive
0:78 – 0:90Application planning + wrap-upActive

Active time: 65 of 90 minutes — 72%.

F — Audit your own module

Ten questions to stress-test it.

  1. Does every learning objective use a measurable Bloom's verb?
  2. Is at least 50% of session time active?
  3. Does every concept have a practical exercise tied to it?
  4. Are all discussion questions open-ended?
  5. Is pre-work under 30 minutes and clearly useful?
  6. Does the session include hands-on tool time (for AI sessions specifically)?
  7. Is there a spot-the-flaw or quality-check exercise for AI output?
  8. Is there a hard-case discussion with no clean answer?
  9. Does the session end with written application planning?
  10. Could a participant explain what they'll do differently this week — using a sentence, not just a feeling?

If you can't answer yes to all ten, the module needs another pass.

G — Want help designing yours?

Bring this thinking to your team.

If your L&D team is building AI training and wants a working session to pressure-test the design, the half-day workshop format covers exactly this material — using your real modules as the working examples.

Book a working session →

← Back to Resources