For School Leaders

You are helping me analyze diagnostic interviews from approximately [number] employees of [school name], [one-line description of your school]. Each submission captures a snapshot of an employee's engagement with AI tools.

Your work has two parts: aggregate analysis (for an in-service follow-up) and individual analysis (for personalized follow-up to each employee).

FRAMEWORK — The Engagement-Discernment Grid

Engagement (mode of use):
1. Peripheral — AI sits at the edge of the work. Use is occasional or exploratory, and no regular task depends on it.
2. Functional — AI handles discrete tasks on demand, such as drafting a message or looking something up, but it is not built into how the work gets done.
3. Integrated — AI is part of recurring workflows. It is a routine partner in core work like communication, grading, and content preparation.
4. Generative — AI is something the person builds with: custom tools, configured context, and new workflows that extend what they can produce.

Discernment (how wisely AI use is judged):
A. Receptive — Output is trusted by default. Little scrutiny is applied, and understanding of AI's limitations is thin.
B. Critical — Output is scrutinized after it is produced: compared against a source and checked before use. Discernment as hygiene.
C. Prudent — Judgment comes first: whether AI belongs in the task, how it should be used, and what must stay human. Discernment as architecture.

The columns differ in when judgment happens, not in how much effort it takes. Two people can check output the same way and still sit in different columns.

Place each employee in one cell. Base placement on behavioral evidence in the structured fields and narrative — what they actually do — not on their self-claims or aspirations. Be conservative; when ambiguous, place at the lower stage. Some submissions will include data the interviewer LLM pulled from saved memory or past chats; weight that more heavily than self-report, since it reflects actual usage.

PART 1 — AGGREGATE ANALYSIS

Produce:
- Distribution across the 12 cells (count and percentage per cell)
- Distribution by role or department
- Top 5 use patterns: what people actually do with AI, with example phrasings drawn from the data
- Top 5 stated aspirations
- Top 5 stated blockers
- Discernment red flags worth surfacing — places where receptive use is institutionally risky (e.g., grading, parent communications, theological writing)
- Outliers in either direction. Especially: anyone placed at Generative. These are people I should learn from. Quote them.
- Three patterns I should highlight that I might not have anticipated. Surprise me.

Write Part 1 as a memo of about 1,200 words. Prose with selective bullets where they genuinely serve clarity. Do not pad. Cite specific quotes from submissions where they make the point better than your summary would. The audience is me — [your role] — preparing for an in-service follow-up presentation.

PART 2 — PER-EMPLOYEE OUTPUT

Do not generate Part 2 until I ask for it, and do it in batches of 10 employees at a time. For each employee, produce two required artifacts and one conditional artifact.

---
Artifact 1 — Email to employee (~250 words):
- Subject line that is specific, not generic
- One sentence naming where they are on the grid, in plain language (describe the behavior; do not name the cell directly)
- One paragraph reflecting what stood out from their interview, specific to them
- One paragraph on the single most leveraged next step for them, with a concrete example tailored to their actual work
- For employees at stages 1 or 2: include a closing line directing them to their custom instructions draft (Artifact 3)
- For employees at stages 3 or 4: close with a pointer to the coaching prompt
- Tone: direct, warm, specific. Do not flatter. Do not list "areas for growth" — pick one thing.

---
Artifact 2 — Coaching prompt (~300 words):
A prompt the employee pastes into their LLM going forward to keep developing. It should:
- Establish a coaching persona calibrated to the employee's current stage. Do not push a Functional user to think about MCP servers. Do not patronize a Generative user.
- Reference their specific role and the work they actually do, drawn from the submission
- Set up a periodic check-in cadence ("every Friday, ask me what I tried this week and what I noticed")
- Push them diagonally — toward both more sophisticated engagement and more discernment
- For Receptive users: build in a verification reflex they have to perform out loud before leaving any chat where output will be used
- For high-engagement + Receptive users: emphasize discernment exercises (catching errors, naming tasks they should refuse AI for)
- For Prudent + low-engagement users: nudge toward experimentation with a specific, low-stakes first step drawn from their stated role and work

---
Artifact 3 — Draft custom instructions (STAGES 1–2 ONLY; omit for stages 3–4):
Generate a ready-to-use custom instructions block the employee can paste directly into Claude (Settings → Profile → Preferences) or ChatGPT (Settings → Personalization → Custom Instructions). Format it as two sections matching each platform's standard fields:
  "What would you like Claude/ChatGPT to know about you?"
  "How would you like Claude/ChatGPT to respond?"
Ground the draft in:
- Their actual role and stated primary work activities
- Any tools, communication patterns, or tasks named in their interview
- Their current discernment level — keep verification prompts and guardrails visible in the instructions if they are Receptive, lighter-touch if Critical
Keep the draft short enough to be non-intimidating (under 200 words total) but specific enough to be immediately useful. Do not include aspirational features they have not demonstrated readiness for.

---
Begin Part 1 when I paste or upload the submissions.