Use this when support tickets need speed but cannot risk invented policies or promises.
| Best for | support agents, founders and operations teams |
|---|---|
| Final output | a support reply with summary, answer, next step, policy boundary and escalation note |
Why this workflow works
This playbook turns one broad AI request into a reviewable sequence. It asks the user to prepare source material first, then uses AI to organize the work, expose missing details and produce a draft that can be checked by a human editor.
For support agents, founders and operations teams, the value is not only speed. The value is a repeatable process: the same inputs, the same review points and the same standard for deciding whether the AI output is ready to use.
What to prepare before running it
- The real business or content goal behind the task.
- Source facts, notes, examples or policies the AI is allowed to use.
- Audience context and channel where the output will appear.
- Boundaries: claims to avoid, facts to verify and details that are unknown.
- The format you want at the end, such as a support reply with summary, answer, next step, policy boundary and escalation note.
Workflow steps
- Summarize the customer issue and emotional state.
- List known policy facts and what is not known.
- Ask AI to draft a reply with empathy and a clear next action.
- Check whether the reply promises anything outside policy.
- Save useful replies as macros with review notes.
Copy-ready prompt
Replace the bracketed fields with your actual source material before using the prompt. If a field is unknown, leave it as unknown and ask the AI to return missing-information questions instead of inventing details.
Example input fields
| Context | Describe the real task, source material and business situation. |
|---|---|
| Audience | Name the reader, buyer, customer, stakeholder or internal team. |
| Constraints | Include claims that must be avoided, facts that must be checked and format limits. |
| Review | Ask for assumptions, missing questions and a checklist before using the output. |
Evaluation rubric
| Clarity | The answer should make the next action obvious without requiring a second explanation. |
|---|---|
| Specificity | The answer should use the provided context and avoid advice that could fit any business. |
| Evidence | Claims, examples and recommendations should be traceable to source material or marked for review. |
| Usability | The final structure should be easy to copy into a document, page, email, ticket or planning tool. |
Common mistakes
- Sounding empathetic while avoiding the real answer.
- Inventing timelines or refund rules.
- Forgetting to ask for missing order details.
- Using a macro without adapting it to the customer issue.
Human review checklist
- Check whether every claim is supported by source material.
- Remove details that the AI guessed.
- Confirm the output matches the intended audience and channel.
- Keep a copy of the source input with the final prompt.
- Revise the prompt when the same issue appears twice.