Facilitator delivery guide
Learning materials only. This guide supports facilitators who deliver the kit for a hosting organization. It does not replace your organization’s policies, safeguarding procedures, or legal counsel.
Each session stays 90 minutes (clinic: 60 minutes; short activity: 15 minutes). Detailed timings are in each run sheet. Use the cut lists below if you run long. Never extend a session beyond its scheduled end.
Before you start
- Complete the facilitator preparation checklist.
- Send the sample invitation and agenda email at least five business days before Session 1.
- Distribute the participant handbook and technology and accessibility checklist.
- Confirm safeguarding focal points with the hosting organization (see escalation path below).
- Open a facilitator reflection log for the series (one entry per session).
Safeguarding escalation path (all sessions)
If a participant shares real beneficiary data, a disclosure of harm, or child protection or GBV-related information:
| Step | Facilitator action |
|---|---|
| 1. Pause | Stop the exercise. Do not ask for more detail in plenary. |
| 2. Protect | Ask co-host to remove identifying content from chat or recording per platform rules. |
| 3. Refer | Direct the participant privately to the hosting organization’s safeguarding, PSEA, or protection focal point. Use contacts named in your pre-session briefing. |
| 4. Do not investigate | You are not a counsellor or investigator. Do not promise outcomes. |
| 5. Document | Notify the hosting organization’s designated contact after the session per their policy. Do not copy sensitive content into your personal notes. |
Primary sources: Core Humanitarian Standard (Commitment 5 on complaints and feedback); IASC Operational Guidance on Data Responsibility (2023) (do-no-harm and dignity). This guide is not legal advice.
Session 3 includes a safeguarding and protection module for teaching content.
Session 1: Responsible AI foundations (90 minutes)
Run sheet: session-01-foundations/run-sheet.md
Minute-by-minute cues
| Time | Cue | Run sheet row |
|---|---|---|
| 0:00 | Welcome, ground rules, diagnostic | Opening |
| 0:10 | Teach AI types and accountability | Teach block |
| 0:25 | Worked example walk-through | Source check demo |
| 0:45 | Assign group roles; start Case 01 | Group exercise |
| 1:10 | Debrief: one protection issue and G/A/R per group | Debrief |
| 1:25 | One-rule recall and action plan | Close |
Expected participant responses
- Most groups identify red for the evacuation SMS case when they compare draft to bulletin.
- Protection issues named: wrong accessibility information, route errors, dropped negations in translation.
- Human control: local communications lead or equivalent approver, not the worksheet reviewer.
Common misconceptions
| Misconception | Response |
|---|---|
| “AI is always dangerous” | Some green uses exist with checks. Proficiency is discrimination, not blanket refusal. |
| “Urgent tone means send” | Speed does not replace verification or approval authority. |
| “Deleting the chat fixes a data breach” | Retention and harm may already exist. Escalate per org policy. |
| “Our org has no AI policy” | Default to amber. Session 3 covers governance. |
Difficult questions
- “Can I use ChatGPT for my real caseload?” Redirect to data-responsibility checklist and org approval. No tool endorsement in the classroom.
- “What if leadership ignores red decisions?” Name accountability gap; suggest pilot charter and policy ask in Session 3. Do not advise whistleblowing paths without org contact.
- Participant shares real case data: Follow safeguarding escalation path above.
Running long: cut list (Session 1)
Cut in this order until back on schedule:
- Shorten teach block by 5 minutes (slides 5 to 6 only; skip optional predictive-AI example).
- Reduce debrief to one poll question per group instead of two verbal reports.
- Skip optional pair share at 1:15 if ahead buffer was used.
- Do not cut: group exercise, contradiction identification, or safeguarding pause if triggered.
Session 2: Verified work by role (90 minutes)
Run sheet: session-02-verified-work/run-sheet.md
Minute-by-minute cues
| Time | Cue | Run sheet row |
|---|---|---|
| 0:00 | Recall Session 1 rule | Recall |
| 0:10 | Teach structured prompts | Prompt teach |
| 0:25 | Pairs: prompt lab Part A | Pairs |
| 0:48 | Groups: source check Part B | Source check |
| 1:10 | Debrief: one correction and one approver | Debrief |
| 1:25 | Close and action plan | Close |
Optional: Organizations with an approved tool may use the live-tool lab route instead of pre-supplied outputs. Default remains no-account.
Expected participant responses
- Prompt lab: improved prompts name task, audience, source, format, and missing-fact rule.
- Source check: 78% district prevalence marked unsupported or contradicted vs 31% in source.
- Managers often classify donor summarization as amber without org approval.
Common misconceptions
| Misconception | Response |
|---|---|
| “I’ll fix the prompt and rerun at work” | Requires approved tool and data rules first. |
| “Rounding 31% to 30% is the main error” | Inventing 78% is a different failure mode. |
| “Pre-supplied AI output is unrealistic” | Many staff receive drafts from colleagues who used unapproved tools. Verification skill still applies. |
Difficult questions
- “Why not practice in the real tool?” Default protects data and avoids account barriers. Live route is optional with synthetic data only.
- “Verification takes too long” Name which claims are consequential vs cosmetic. Session goal is selective verification, not line-by-line proofreading of everything.
Running long: cut list (Session 2)
- Shorten prompt pair work by 5 minutes (provide one partially improved prompt as starter).
- Use Case 2 only; skip Case 3 variant.
- Debrief via chat poll: one correction, one approver, no verbal round.
- Do not cut: source-check table, screen break, or live-tool safety briefing if using optional route.
Session 3: Governance and adoption (90 minutes)
Run sheet: session-03-governance/run-sheet.md
Safeguarding module: safeguarding-module.md (10 minutes within teach block).
Minute-by-minute cues
| Time | Cue | Run sheet row |
|---|---|---|
| 0:00 | Recall pathway Sessions 1 to 2 | Recall |
| 0:10 | Teach policy vs convenience; safeguarding module | Teach |
| 0:25 | Decision block: fictional brief | Decision block |
| 0:55 | Transparency card exercise in pairs (15 min) | Transparency card |
| 1:10 | Challenge block: one written perspective (shortened) | Challenge |
| 1:25 | Post-assessment reminder; 30-day follow-up preview; close | Close |
Timing note: The transparency card exercise uses 15 minutes permanently taken from the challenge block (was 30 minutes; now 15). The challenge block is the least essential activity shortened to keep the session at 90 minutes.
Expected participant responses
- Initial decision often option 2 or 3 (more evidence or limited pilot), rarely option 1 alone without rationale.
- Charter drafts name consultation still required, stop thresholds, and manual fallback.
- Challenge block revises decision toward more consultation or narrower pilot scope.
Common misconceptions
| Misconception | Response |
|---|---|
| “Classroom role play equals consultation” | Real pilots require real engagement. See AAP adaptation design note. |
| “Vendor promises mean green” | Vendor claims are not verification. Amber until framework exists. |
| “Uploading de-identified registration data is safe” | Re-identification risk and partner agreements still apply. |
Difficult questions
- “Our executive already signed the contract” Name governance gap; charter and exit plan still required before operational upload.
- “Indigenous data sovereignty” Direct to org policy and Session 3 worksheet consultation row. Do not simulate nation-specific consent in plenary.
Running long: cut list (Session 3)
The standard plan already shortens the challenge block by 15 minutes for the transparency card exercise. If you still run long:
- Shorten safeguarding teach to 5 minutes (facilitator summary only; skip pair reflection).
- Reduce decision block by 10 minutes (assign one case in advance; groups start at 0:25).
- Skip verbal debrief of transparency cards; collect one field per pair in chat.
- One group presents charter headings; others submit worksheets asynchronously within 48 hours.
- Do not cut: fictional-data rule, consultation reminder, transparency card exercise, or post-assessment announcement.
Follow-up clinic (60 minutes)
Run sheet: clinic/run-sheet.md
Minute-by-minute cues
| Time | Cue |
|---|---|
| 0:00 | Welcome; de-identified questions only |
| 0:05 | Pulse poll |
| 0:15 | Q&A queue (3 to 5 pre-submitted questions) |
| 0:45 | Peer exchange |
| 0:55 | Close and evaluation |
Expected participant responses
- Pulse poll: mix of “tried screening at work” and “blocked by no policy.”
- Q&A: practical questions about approval paths, not requests to review live casework.
Common misconceptions
| Misconception | Response |
|---|---|
| “Clinic can troubleshoot my real dataset” | De-identified questions only. Redirect casework to org channels. |
| “No questions means we failed” | Use backup scenarios from run sheet. |
Difficult questions
Apply the seven-step screen aloud with the group. If a question contains PII, pause and move to private channel with org contact.
Running long: cut list (clinic)
- Limit Q&A to three questions.
- Skip peer exchange; send written prompt after clinic.
- Do not cut: opening rules on de-identification or safeguarding escalation.
15-minute source-check activity
Activity sheet: short-source-check/activity.md
Cues
Follow the activity run-of-show table (minutes 0 to 15). Pre-assign rooms before starting the clock.
Expected responses
- Pairs flag 08:00 vs 09:00, invented shuttle, missing live-notice check.
- Decision: stop publication pending verification with responsible authority.
Running long: cut list
- Skip room comparisons; go straight from pair work to plenary poll.
- Shorten one-line rule to optional async submission.
- Do not cut: fictional-data warning or answer key reveal.
After the series
- Score performance scenarios and worksheets with the skills rubric.
- Complete the facilitator reflection log end-of-series synthesis.
- Schedule the 30-day transfer follow-up (one reminder only; anonymous option).
- Send hosting organization aggregate themes only (no individual ranking).
- Feed transfer summary into the sample final-report page.
- Remind participants that materials are CC BY 4.0; adaptation requires separate AAP process per design note.