VetsLearn · 2-hour workshop guide · Facilitation guide v2, 2026-10-09
At a glance
Length: 120 minutes. Room: laptop, projector and internet for the tool demo. The first 70 minutes work offline. Group: sized for about 6 to 30. Prep: read the primer, run the Skill Translator and Benefits Finder demos that day, print the participant worksheet (last page) one per person.
Session objectives
The session is built so that, by the end, each participant has practiced how to:
- Tell a generated claim from the evidence that supports it.
- Judge each claim in a draft as supported, unsupported or contradicted by its source.
- Write an Operator prompt with four parts: Context, Task, Standards, Evidence.
- Run one tool against a task they own and list what they would check before using the result.
These are practice goals. The session does not test or certify anyone.
Source lessons
- Foundations 01, “What AI actually is”. Tutorial format: a chaptered video of about 11 minutes in 9 chapters, a “Try it first” step and a practice workspace on the page. Planned reading 10 minutes, practice 15 minutes.
- Foundations 03, “Prompt like an Operator”. Planned reading 6 minutes, practice 15 minutes.
- Foundations 04, “When AI is wrong”. Planned reading 3 minutes, practice 20 minutes.
- Follow-on for participants: Foundations 02, “Five useful prompts and how to check them”. It is not taught in this session.
Timed agenda
| Time | Block | What happens |
|---|---|---|
| 0:00–0:10 | Try it first | One line about yourself. Show the source facts and the flawed draft from Foundations 01. The room judges each claim before you explain anything. |
| 0:10–0:30 | Generation and evidence | What a language model produces, and why a fluent draft is not proof. |
| 0:30–0:50 | When AI is wrong | Kinds of failure and the check that fits each. AI drafts, you verify, you own. |
| 0:50–1:10 | Prompt like an Operator | Context, Task, Standards, Evidence. Two prompts for the same task, side by side. |
| 1:10–1:35 | Tool demo | Skill Translator with code 68W. Then Benefits Finder if many in the room share a state. Read each data date aloud. |
| 1:35–1:55 | Worksheet rep | Everyone fills in the worksheet for a real task they will do this week. You circulate. |
| 1:55–2:00 | Next move | One thing to do tomorrow. Where the free lessons and tools are. |
Block by block
0:00 Try it first
Do not open with what AI can do. Open with a draft that is wrong. Use the case from Foundations 01. It is fictional, it needs no AI account and it is the same every time. Put the source on screen first.
Source R1
Riley coordinates six people processing 40 requests weekly at about 18 minutes per request. A two-week trial of AI-assisted request summaries is proposed. No trial has run; savings are unknown.
Draft to judge
“Our 20-person team has reduced processing time by 40% with AI. The system is ready for rollout across the organization. We recommend immediate adoption.”
Ask the room to mark each claim: supported (the source says it), contradicted (the source says otherwise) or unsupported (the source does not say). Take answers before you give any.
- 20-person team: contradicted. The team size contradicts R1: six, not twenty.
- 40% reduction: unsupported. The 40% reduction is unsupported: the source reports no trial.
- Ready for rollout: unsupported. Readiness for rollout is also unsupported.
- Immediate adoption: unsupported. Immediate adoption is a recommendation without trial evidence; a proposal can still be poorly justified.
Say this
“I am not here to sell you on AI. I am here so you can use it and catch what it gets wrong. This draft reads fine. Tell me which parts the source backs up.”
You can also project the lesson page and use its “Try it first” step with the room. It is in the chapter “Find the mistake”.
0:10 Generation and evidence
- A language model generates text from patterns learned in training and from the context you give it.
- A fluent draft is useful. It is not proof. A model can work through a task and still give a wrong answer.
- A statement needs evidence: a source, a calculation or an observed test that supports that statement.
- A chat product may add search, tools or memory. A confident answer does not show a source was checked. Ask what was available and what was checked.
Ask the room
“A draft says a trial saved 40%. What would establish that claim?” Answer: evidence of the trial and its measurement, including what was compared. A polished sentence does not supply a trial that never ran.
0:30 When AI is wrong
Match the check to the failure:
- An unsupported or made-up claim: find the source, or remove the claim.
- A misread source or a dropped condition: open the exact passage. Watch for “may” turned into “must”.
- A number: recompute it.
- A current rule, rate or deadline: check a current authoritative source.
These are not independent evidence: the model sounding sure, asking the same question again, or a second model agreeing. Check numbers, dates, quotations, citations, current information and legal or technical claims first. That is a checking priority, not a ranking of which errors happen most.
AI drafts. You verify. You own. Your name goes on the work.
0:50 Prompt like an Operator
Four parts, in order. The lessons call it CTSE.
| Part | What it answers |
|---|---|
| Context | Who you are, what is going on, what has already happened. |
| Task | What you want produced. Use a verb. Be specific. |
| Standards | Format, length, audience, tone, what to include and what to avoid. |
| Evidence | Which supplied facts it may use, how claims are checked, and what to do when support is missing. |
Show two prompts for the same task. The profile is fictional: managed 47 personnel and a $12 million fleet during a one-year deployment, with zero accidents and 96% mission completion.
Prompt 2: “Using only the supplied fictional profile, write three active-voice bullets for a civilian supply-chain manager. Remove military jargon; use at most 25 words per bullet. Preserve all supplied figures and add no achievements.”
Prompt 2 says which facts may be used, who will read the result and how it will be judged. Those are things you can inspect. Have the room name three checks for its output: every number matches the profile, no achievement is added, the audience and format are respected. This comparison is not a measured improvement in model performance and not a promise about hiring.
1:10 Tool demo
Run Demo 1 (Skill Translator, 68W) from the tool demo scripts. If many in the room share one covered state, add Demo 2 (Benefits Finder) for that state. Read the AI notice at the Skill Translator description box before you type in it, and read the “Data current as of” line on each tool.
1:35 Worksheet rep
Hand out the worksheet on the next page. Each person picks a real task for this week and writes. Circulate and keep it moving.
1:55 Next move
Say this
“One move tomorrow. Take a task you would do anyway. Write the four-part prompt, then check the output against your source before you use it. The lessons are free at vetslearn.com and the tools are free at tools.vetslearn.com. Next lesson for you is Foundations 02, ‘Five useful prompts and how to check them’.”
Participant worksheet: 2-hour workshop
Print one per participant. It stays with them. Do not collect it.
Name: Date:
1. The task I will run this week
A real task you would do anyway: an email, a summary, a first draft, a plan. Nothing classified or controlled.
2. My Operator prompt
Context (who I am, what is going on, who it is for):
Task (what I want produced):
Standards (format, length, tone, what to avoid):
Evidence (which facts it may use, and what to do when support is missing):
3. What I will check before I use it
List each name, number, date or claim you would confirm against a real source.
4. Three questions
| Would I put my name on this as written? | YES / NOT YET |
| Does the output contain a number I did not supply? | YES / NO |
| Did I read the data date on any tool I used? | YES / NO |
5. My one move tomorrow
Keep going, free: lessons at vetslearn.com/lessons/ · tools at tools.vetslearn.com · the ADAPT framework and its worksheets at vetslearn.com/framework/
Worksheet from VetsLearn (vetslearn.com), Soldier Data LLC. Facilitation guide v2, 2026-10-09.