Train teachers and staff on top-line guardrails

Rules of the road, not driver's education

Nobody learns to drive in one session, but everyone learns which side of the road to be on before they get in the car. Action two is that: the small number of things every staff member must know before they open an AI tool, delivered to everyone, quickly.

The temptation is to skip ahead to the interesting part — prompt techniques, lesson design, the demo that gets applause. Do the guardrails first. They are what make the interesting part safe to teach.

Questions to answer

• What information must never be entered into an AI tool? Tie the answer to FERPA and to your own data classification scheme, so it is a rule rather than a vibe.
• Can AI output be trusted? What does a staff member do to check it, and what is the standard before something reaches a student or a family?
• What oversight responsibility does a person keep when they use AI to do part of their job? Who is accountable when the output is wrong — the tool, or the person who sent it?
• How will training reach people who miss the first offering? One session at one time reaches one shift.

What "done" looks like

1. An intro learning session for all AI users

   Offered more than one way — live, recorded, in a staff meeting, in a newsletter — because the goal is coverage, not attendance.

2. Clear do's and don'ts

   Short enough that people remember them without looking. Specific enough to cover the cases that actually come up in your buildings.

3. Data privacy expectations

   What never goes in, stated plainly. Student names, IDs, grades, discipline records, IEP content, health information, anything you would not email to a stranger.

4. Output evaluation basics

   How to check for accuracy, fabricated citations, bias, and instructional fit — and the habit of assuming polished output can still be wrong.