Use this practical rubric to identify what someone can demonstrate, what support they need and what to practise next. It is a learning conversation based on a short task, not a confidence survey.
Purpose: employee learning and training needs. Not for: hiring, promotion, performance ranking, certification, or assessing whether a business is ready for AI implementation. This is a facilitator-created practice rubric, not a validated psychometric test.
How to use the rubric
- Explain the learning purpose, who can see the record and how long it will be kept. Agree the activity and any accessibility support.
- Use a short, comparable task with fictional or approved public material. The companion meeting-notes exercise provides a ready-made example.
- Observe the work and ask the participant to explain their choices. Keep the prompt, marked-up output or a short evidence note where appropriate.
- Assign a level separately for each skill using observed behaviour. If there is no opportunity to observe a skill, record N/O — not observed, not zero.
- Agree one practice action and a review date. Revisit with a new task of similar difficulty. Record any change in tool, task or support that affects comparison.
Allow about 45 minutes for the companion exercise plus a short learning conversation. Adapt timing to the participant; speed is not a score.
Level descriptions
- 0 — Not yet demonstrated: The task gave a fair opportunity to show the skill, but it was not demonstrated, even with support. Plan guided practice.
- 1 — With guidance: Demonstrates the skill with specific prompts, reminders or a worked example. Record the support provided.
- 2 — Independently on this task: Demonstrates the skill without task-specific prompting in this activity. Do not generalise this to every task or permission level.
- 3 — Explains and adapts: Demonstrates the skill, explains the reasoning and adapts it to a second example or variation. A confident explanation alone is insufficient.
- N/O — Not observed: The task or circumstances did not supply adequate evidence. Provide another opportunity rather than making a judgement.
Skill 1 — Choose a suitable task and permitted information
0: Uses restricted material or cannot recognise the task boundary after support.
1: Selects a suitable task and removes disallowed information after a specific reminder.
2: Independently checks the approved-tool and information boundaries, uses permitted material and recognises when to ask for help.
3: Also explains the boundary and applies it correctly to a second scenario, including when not to use the tool.
Evidence to look for: A stated permitted use, a data-boundary check, and an example of information that must stay out of the tool. Use fictional examples; do not ask someone to reveal a real confidential record as evidence.
Skill 2 — Frame a clear request
0: Does not communicate the task or required output clearly enough to complete the activity, even with support.
1: Uses a template with help to include the task, audience, source and format.
2: Independently specifies a clear task, audience, source boundaries and useful output format.
3: Also explains how those choices affect the output and adapts the request for a different audience or constraint.
Evidence to look for: The actual prompt and an explanation of what the output should contain. Do not reward verbosity by itself.
Skill 3 — Check factual claims against the source
0: Treats a plausible draft as correct despite a guided opportunity to check it.
1: Finds an unsupported claim when directed to a specific sentence or source passage.
2: Independently compares important claims with the source, corrects errors and keeps unknowns visible.
3: Also explains a repeatable checking approach and applies it to a new output or scenario.
Evidence to look for: Marked-up claims, source references, corrections and unresolved questions. For the companion exercise, look for capacity versus bookings, proposed versus approved budget, and unassigned owners.
Skill 4 — Improve the output with specific feedback
0: Cannot make a useful correction or revision during the activity, even with support.
1: Improves the draft using a suggested revision or worked example.
2: Independently identifies a concrete weakness, revises the prompt or draft and checks that the change helped.
3: Also compares alternatives, explains the trade-off and adapts the approach to a second case without losing source accuracy.
Evidence to look for: Before-and-after versions with a reason for the change. The revision should solve an identified problem, not merely sound more polished.
Skill 5 — Apply audience and human-review boundaries
0: Treats a draft as ready to share despite a guided opportunity to identify review or permission requirements.
1: Identifies the audience and review point after a reminder.
2: Independently keeps internal material internal, names a human reviewer and flags missing approvals before use.
3: Also explains how the review route changes for a higher-risk or external audience and applies that distinction in a second scenario.
Evidence to look for: A clear internal/external distinction, the name or role of a reviewer, and a decision to stop when approval is absent. Do not require actual external sharing to demonstrate this skill.
Skill 6 — Explain the work and identify a next step
0: Cannot describe the approach or a useful next step after support in this activity.
1: Describes a useful action with guided questions.
2: Independently explains the task, checks and limitations, and chooses a specific practice goal.
3: Also teaches the approach using an example and adapts the practice goal to a new learning need.
Evidence to look for: A short explanation in the participant's preferred accessible format. Communication style, accent or confidence is not a proxy for AI judgement.
Copyable observation worksheet
Participant or voluntary identifier: [Write here]
Facilitator and date: [Write here]
Task, source material and approved tool: [Write here]
Access arrangements, support provided and relevant context: [Write here]
| Skill | Level (0–3 or N/O) | Observed evidence | Next practice action |
|---|---|---|---|
| Suitable task and permitted information | [Write here] | [Write here] | [Write here] |
| Clear request | [Write here] | [Write here] | [Write here] |
| Source-based checking | [Write here] | [Write here] | [Write here] |
| Specific revision | [Write here] | [Write here] | [Write here] |
| Audience and human review | [Write here] | [Write here] | [Write here] |
| Explanation and next step | [Write here] | [Write here] | [Write here] |
Participant's reflection: [Write here]
One agreed practice priority: [Write here]
Support person and review date: [Write here]
Who can access the record, and deletion/review date: [Write here]
Turn observations into a learning plan
Do not add the levels into a single readiness score or invent a pass mark. The skill profile is more useful than a total, and a total can hide a serious checking or information-handling gap.
- If safe information handling is uncertain, return to approved-tool and data-boundary practice before practising with real work information.
- If prompting is adequate but checking is weak, practise finding unsupported claims rather than collecting more prompt templates.
- If a participant succeeds with support, document what support worked and offer another comparable practice opportunity.
- If the task was inaccessible or unfamiliar, fix the activity design before drawing conclusions about ability.
- If someone performs well in this exercise, retain normal workplace approval, supervision and policy requirements. Training completion does not authorise unrestricted AI use.
Worked interpretation — illustrative, not a result
A participant writes a clear request independently but only notices the fictional printing budget is unapproved after a facilitator points to that sentence. You might record level 2 for clear request and level 1 for source-based checking, with the specific evidence and support noted. A useful next activity is checking a new draft for approval-status errors without that prompt.
This example explains how to use the rubric. It is not a score or outcome from a real person, and it does not establish the rubric's reliability.
Continue with practical learning
Use the 45-minute AI workshop exercise to collect observable evidence, then adapt the four-week team AI training plan around the needs you find.
This resource assesses human learning needs only. For business implementation support rather than a people-skills diagnostic, visit AI Agent Agency.