Build an AI Use rubric
In this unit you build an AI Use rubric that supports consistent decisions in classroom and staff contexts. First, you walk through a worked example and understand why each row choice keeps responsibility visible. Next, you build one rubric row yourself using a guided template. Finally, you draft your own complete rubric with at least four rows from your real work.
Role pathways
Choose the pathway that reflects your professional role.
For teachers
Your rubric work centers on learning, feedback, student agency, and what counts as student work. Build checkpoints that protect accuracy, learning quality, and your transparency with students.
For coaches
Your rubric work centers on tasks that shape shared practice: planning supports, meeting summaries, and staff-facing resources. Build checkpoints that protect accuracy and shared understanding while keeping workload realistic.
For administrators
Your rubric work centers on tasks that affect trust: staff communication, family messaging, and official records. Build checkpoints that verify high-impact claims and keep sensitive information protected.
Worked example - build a row, understand the logic
In this teacher rubric row examples each cell includes an annotation explaining the decision logic behind it. The common pitfall appears at the end.
Read the vague prompt. When you explored all four, review the upgraded prompt for a revised version that applies each element.
Rubric row-teacher
Example: Draft feedback comment starters aligned to a writing rubric
| Task Type | What is the task? Draft feedback comment starters for each rubric level. Use these as a starting bank, then adapt each comment to the specific student work. |
Decision callout The task description names exactly what AI is being used to produce—comment starters, not final comments. This precision matters because it keeps the boundary between AI drafting and human judgment visible from the start. |
| Category | What is the fit? Appropriate with human oversight |
Decision callout The category isn't a rating of the AI tool. It's a decision about this specific task. Comment drafting is appropriate because the educator applies the final judgment—but it requires oversight because the output directly shapes how students understand their work. |
| Must stay human | What can't be delegated? The educator verifies each comment against the student's actual work and decides the specific next learning step for each student. |
Decision callout This column protects professional judgment. AI can draft a comment about paragraph structure in the abstract. Only the teacher knows whether this particular student's paragraph structure problem is a skill gap, a comprehension issue, or a confidence barrier. That distinction must stay human. |
| Privacy default | What boundary is set before you start? Don't paste student names, identifying details, or full student paragraphs into the AI prompt. Describe the task and the rubric level instead. |
Decision callout A privacy default is written as a before-you-start boundary, not an after-the-fact fix. If student data ends up in an AI prompt, the privacy risk has already occurred. The default prevents the exposure before the session starts. |
| Required checkpoint | What must be verified before use? Spot-check every comment against the draft it refers to. Revise any comment that uses generic phrasing or could apply to any student equally. |
Common Pitfall A rubric row that says "review before sharing" isn't enough. Strong checkpoints name what to review, against what source, and what counts as a problem. "Generic phrasing that could apply to any student" is a specific, testable standard. |
| Transparency line | What will you say if asked? I used AI to draft comment starters, then reviewed and revised each one based on your specific work. |
Decision callout The transparency line is short and calm. It doesn't over-explain or apologize. It describes the review step, which is the most important part—because the review is what makes the output trustworthy, not the drafting. |
| Documentation | What will you keep? Final comments and the checklist or criteria used for the spot-check review. |
Why this matters? Documentation supports accountability when someone asks how a decision was made. It also helps you repeat the process consistently and share it with colleagues. |
Independent practice
Open Copilot Chat and practice writing, revising, and evaluating prompts in a real environment. The goal isn't to learn the tool. The goal is to apply the prompting habits and judgment you developed in the Explore pages and the guided practice above.
Work through these steps:
- Choose the role-based starting prompt that best fits your work.
- Open the Copilot Chat.
- Enter the original vague prompt exactly as written and review the output you receive.
- Revise the prompt to include a clear purpose, audience, real constraints, and at least one quality criterion.
- Enter your revised prompt into the same tool and review the new output.
- Apply all four evaluation checks before deciding what to keep, revise, or set aside.
- Record these in your notebook: the original prompt, the revised prompt, one concrete improvement you noticed in the output, and one place where your human judgment must lead.
Document your learnings: Create a OneNote page called prompt upgrade log. Each time you use AI for a professional task, add an entry with four fields:
- The original prompt
- The revised prompt
- One improvement you noticed
- One thing that still needed human judgment.
Over a school year, this becomes a personal reference for what works in your specific role and context.