Use this as a working artifact, not a reading assignment. First check whether the problem matches your situation. Then copy the template structure into the tool your team already uses. If you use AI, give it approved source material and keep a human review step before making decisions.
- A clearer version of the problem this template is meant to solve.
- A first draft you can review with the requester, reviewer, leader, or owner.
- A short list of missing information, assumptions, and next actions.
AI experiments stay scattered when the team does not agree where AI fits in the work.
AI becomes more useful when it has a job inside a workflow. The prompt matters, but the source material, review gate, and human decision owner matter more.
- Where can AI remove blank-page work without removing judgment?
- What should AI never be allowed to decide in this workflow?
- What review step would make AI-assisted work safer and more useful?
This sets up the AI workflow layer of the system. The point is not to collect prompts. The point is to decide where AI helps, where people review, and where the final decision belongs.
- People are using AI differently across the same team.
- Prompt sharing is happening without workflow standards.
- A leader wants AI productivity without weakening review quality.
- One workflow where AI might help.
- The approved source material AI is allowed to use.
- The person who owns the final decision before anything goes live.
Start small enough that the work can move today.
- Choose one low-risk workflow instead of trying to govern every AI use case at once.
- Name allowed source material before writing prompts.
- Separate what AI can help draft from what a person must approve.
- Write the review gate before anyone publishes the output.
Use this when you need the words.
- AI can help here, but we need to define the workflow before we define the prompt.
- What source material is allowed, what is AI allowed to do, and where does human judgment take over?
- If we cannot answer those questions, we are not ready to turn this into a team standard.
Use the answers to choose the next move.
AI can assist with first drafts, outlines, summaries, and variants.
AI can draft, but SME and designer review are required before use.
AI can help prepare questions, but owners approve all final wording.
Use only approved tools and follow the organization's data policy.
Do not standardize the AI step yet.
- The team has not agreed which tool is approved for the data involved.
- The workflow includes sensitive data without a clear policy.
- The output would go live without SME, accessibility, risk, or owner review.
Copy this structure into the tool you already use.
Paste this into the tool next to the work.
Blank version
# AI Workflow Brief
| Field | Notes |
| --- | --- |
| Workflow | |
| Allowed source material | |
| Allowed AI tasks | |
| Not allowed | |
| Review gate | |
| Human decision owner | |Completed example
# AI Workflow Brief example
| Field | Example |
| --- | --- |
| Workflow | Drafting scenario questions from approved product documentation. |
| Allowed source material | Published SOPs, approved product notes, existing training examples, and SME-provided constraints. |
| Allowed AI tasks | Generate first-pass scenario questions and distractor options. |
| Not allowed | Do not invent policy details, approve accuracy, or publish final assessment items. |
| Review gate | Designer checks learning fit. SME checks facts and workflow accuracy. |
| Human decision owner | Program owner approves final scenario set after SME review. |Choose one low-risk workflow and mark where AI assists, where humans review, and who owns the final decision.
Use Word, Google Docs, SharePoint, Confluence, or a project brief template to document allowed inputs, source links, review criteria, and publish rules.
Use AI to draft a workflow brief, prompt checklist, review checklist, and risk questions from approved source material.
Use these when AI can help shape the first draft.
Use these as starting points, then adjust them to your approved tool, source material, and review standard.
Draft an AI use boundary
Use this when the team is moving fast with AI and needs one practical boundary before scaling the workflow.
I am defining an AI-supported L&D workflow.
Workflow context:
[paste context here]
Help me draft a practical AI use boundary.
Return:
1. Workflow summary
2. Allowed source material
3. AI can help with
4. AI should not do
5. Required human review gate
6. Risk questions to answer before use
7. Final decision owner
8. One small pilot recommendation
Rules:
- Do not assume the tool is approved for sensitive data.
- Keep human review visible.
- Flag privacy, security, legal, accessibility, and learner-impact concerns.Validation checklist
- Allowed source material is named.
- Not-allowed AI tasks are explicit.
- The review gate names who checks facts, learning fit, tone, accessibility, and risk.
- The final decision owner is a person, not the AI tool.
- Check every fact against an approved source.
- Mark anything AI guessed, inferred, or could not confirm.
- Remove private, sensitive, or customer-specific details that should not be in the working file.
- Confirm the right human owner approves the final decision.
- Review tone, accessibility, and learner impact before anything goes live.
Platform-specific starters
ChatGPT GPT-5 family
Use an outcome-first prompt with the job, source material, constraints, and the exact artifact you want back.
I am using the AI Workflow Brief for an L&D workflow.
Goal: Help me turn the rough notes below into a practical first draft of the template.
Context: AI experiments stay scattered when the team does not agree where AI fits in the work.
Use these template fields: Workflow, Allowed source material, Allowed AI tasks, Not allowed, Review gate, Human decision owner.
Rules:
- Ask clarifying questions if the notes are too thin.
- Do not invent facts, policy details, metrics, or source material.
- Separate what is known from what needs human confirmation.
- Keep the output practical enough to review in a working meeting.
Rough notes:
[paste notes here]
Return:
1. Completed first draft
2. Missing information
3. Risks or assumptions to review
4. One recommended next actionClaude 4 family
Use clear XML-style sections so Claude can keep context, task, constraints, and output format separate.
<context>
I am using the AI Workflow Brief for an L&D workflow.
AI experiments stay scattered when the team does not agree where AI fits in the work.
</context>
<source_notes>
[paste notes here]
</source_notes>
<task>
Turn the source notes into a practical first draft using these fields: Workflow, Allowed source material, Allowed AI tasks, Not allowed, Review gate, Human decision owner.
</task>
<constraints>
Do not invent facts, policy details, metrics, or source material.
Separate what is known from what needs human confirmation.
Flag anything that changes scope, ownership, risk, or decision rights.
</constraints>
<output_format>
1. Completed first draft
2. Missing information
3. Review risks
4. Recommended next action
</output_format>Gemini 3 family
Use a structured task with an example pattern. For Obsidian context, use approved excerpts, Drive exports, Google Docs, or NotebookLM.
Task: Complete a first draft of the AI Workflow Brief from the notes provided.
Template fields:
- Workflow: Which L&D workflow will AI support?
- Allowed source material: What can AI use as input?
- Allowed AI tasks: What can AI help with?
- Not allowed: What should AI not do in this workflow?
- Review gate: Who reviews the output, and what do they check?
- Human decision owner: Who owns the final call before the work goes live?
Example pattern:
Field: Workflow
Good answer: Drafting scenario questions from approved product documentation.
Rules:
- Use only the notes provided.
- If information is missing, write "Needs confirmation".
- Keep the output concise and reviewable.
- End with the next best action.
Notes:
[paste notes here]Microsoft 365 Copilot
Use goal, context, expectations, and source. For Obsidian context, use approved excerpts, Word summaries, OneDrive files, SharePoint pages, Teams context, or Outlook threads.
Goal: Create a first draft of the AI Workflow Brief.
Context: AI experiments stay scattered when the team does not agree where AI fits in the work.
Source: Use the selected document, meeting notes, request thread, or pasted notes as the only source.
Expectations:
- Fill these fields: Workflow, Allowed source material, Allowed AI tasks, Not allowed, Review gate, Human decision owner.
- Keep uncertain items marked as "Needs confirmation".
- Do not add facts that are not in the source.
- Summarize the top three review questions for the team.
Output: Return the completed draft, missing information, and one recommended next action.