Most AI literacy sessions spend their first fifteen minutes on how the technology works and lose the room permanently. The people in front of you do not need to know how it works. They need to know what to do with it and what never to do with it.
The four things worth an hour
1. An accurate mental model, in ninety seconds
The version that holds up: these systems predict likely continuations of text based on patterns learned from enormous amounts of material. They are extraordinarily good at producing plausible language and have no independent way of knowing whether what they produced is true.
That single idea explains almost everything staff need to anticipate — why output sounds confident when it is wrong, why it invents citations, why it is excellent at drafting and unreliable at facts. It does more work than an hour of architecture.
2. What may never go in
The highest-value five minutes in the entire session. Specific to your organization, read aloud, and given to everyone in writing:
- Customer, patient, student, or employee personal information
- Credentials and access keys
- Anything under a nondisclosure agreement or a contract confidentiality clause
- Unreleased financials, legal matters, personnel matters
3. Verification, as a habit rather than a warning
"Always check the output" is advice everyone agrees with and nobody operationalizes. Give them a rule they can actually follow: anything factual, numeric, legal, or attributable gets verified against a source before it leaves your hands. Drafting, summarizing, and reformatting your own material do not need the same treatment. Staff follow rules with edges.
4. Where it genuinely helps, using their own work
Do not demonstrate with a generic example. Ask the room for a task they all do and dread, and work it live. Adoption happens in the moment someone sees their own Tuesday afternoon get shorter, and not before.
What to leave out of the first session
| Leave out | Why |
|---|---|
| Model architecture and training methodology | Interesting to a minority, and it costs you the attention of everyone else |
| The vendor landscape | Changes constantly, and staff should be using approved tools, not shopping |
| Advanced prompting frameworks | Belongs in role-specific training, not literacy |
| Speculation about the future of work | Raises anxiety and teaches no behavior |
| A tour of every feature | Feature tours are forgotten by Friday |
The second hour, later
Once the baseline exists, split by function. Finance, HR, communications, and operations have genuinely different risks and different high-value uses, and a session for each one is far more effective than a longer general session for everyone. Do this a few weeks after the first, so people arrive with real questions from their own attempts.
How to make it stick
- Give them a one-page job aid with the prohibited-data list, the verification rule, the approved tools, and who to ask. Most of the retained value of the session lives on that page.
- Have supervisors ask about it within two weeks. Training that is never mentioned again is filed as an event, not a change.
- Publish the approved-tool list where people work, not in a policy binder.
- Repeat annually. Both the tools and the staff turn over.
Questions people ask
How technical does AI literacy training need to be?
Barely at all. Staff need an accurate mental model of what the tools do, not an explanation of model architecture. Technical depth in a general session reliably loses the room in the first ten minutes.
Should we teach specific tools or general principles?
Both, in that order of durability. Principles survive tool changes; a session built entirely around one product's interface is obsolete when the interface changes, which it will.
How do we know the training worked?
Pick a behavior to observe: fewer confidential pastes reported, more staff asking before using a new tool, output being verified before it goes out. Satisfaction scores measure whether people enjoyed the hour.
What about staff who refuse to use AI at all?
They still need the awareness content, because they will receive AI-generated material from colleagues and vendors and need to evaluate it. Framing it as literacy rather than adoption lowers the resistance considerably.