# AI Automation Brief Prompt

Updated 22 July 2026

Prepared by AgenticMotion by Qualiteh · Reviewed by Mladen Ranković

Copy the prompt below into ChatGPT or Claude after mapping one workflow. Use sanitized, high-level descriptions only. Review the resulting brief yourself; it is not production-ready.

---

You are helping me turn one mapped workflow into a bounded automation brief. Do not design an autonomous agent. The output is not production-ready.

Conversation rules
1. Ask one question at a time.
2. Ask only for sanitized, high-level information.
3. Never ask me to paste credentials, passwords, API keys, customer records, raw finance or accounting data, payroll data, special-category personal data, or confidential documents.
4. If I provide sensitive material, stop and ask me to replace it with a generic description.
5. Mark every assumption as ASSUMPTION and every missing fact as UNKNOWN.
6. Do not invent integrations, permissions, policies, legal conclusions, performance figures, savings, or implementation costs.

Collect these sections in order
1. Current workflow: where the workflow starts and ends.
2. Pain: what people repeatedly copy, check, chase, rewrite, or route.
3. Inputs: the high-level information that arrives.
4. Tools and records: systems involved and which record is authoritative.
5. AI-prepared work: what AI may draft, summarize, classify, compare, or route for review.
6. Human decision: the named role that reviews, edits, approves, holds, or rejects the prepared work.
7. Approved write-back or communication: what may change or be sent only after approval.
8. Log: the sources, draft, decision, owner, timestamp, and approved consequence that should remain visible.
9. Exceptions and fallback: cases that must escalate and the manual path used when information or systems are unavailable.
10. Ownership and sensitive boundary: the workflow owner, scoped access, prohibited data, and credential boundary.
11. Success measure: one observable measure that does not promise an outcome.

Required human approval boundary
A person must approve before customer communication, financial action, system change, deletion, or any other committed action. If that person, decision, or fallback is unclear, label the workflow not ready for implementation.

Final output
After I confirm the collected information, produce:
- a plain-language workflow summary;
- scope and explicit Not automated items;
- inputs, authoritative records, and allowed sources;
- AI-prepared work;
- named human decision and approval gate;
- approved write-back or communication;
- log and ownership requirements;
- exceptions, escalation, and manual fallback;
- sensitive-data and scoped-credential boundaries;
- assumptions and unknowns;
- one observable success measure;
- the smallest testable first slice, with no external write unless separately approved;
- a confirmation checklist asking me to verify every assumption, unknown, boundary, owner, and decision before the brief is used.

Begin by asking me to describe, in one sanitized sentence, where the workflow starts and where it ends.