Extract decisions, risks, and actions from a document
Build a traceable register without turning proposals or discussion into commitments
5 min setupTested with:ChatGPTReviewed: 2026-08-28
1
Add your context
Your text stays in this browser. AILesson Prompts does not send it to a model or server.
2
Your prompt
Unfilled fields remain visible as placeholders, so you can still copy and edit the prompt
Extract decisions, risks, and actions from the source.
Document:
[document]
Scope:
[scope]
Interpretation rules:
[rules]
Use only explicit source evidence. Do not convert a proposal, option, question, recommendation, or scheduled discussion into a decision. Do not invent owners, deadlines, probability, or impact. Return three separate tables. Decisions: ID, decision, status, date, decision-maker, evidence locator. Risks: ID, risk statement, cause, consequence, mitigation, owner, evidence locator. Actions: ID, action, owner, due date, dependencies, status, evidence locator. Write Not specified for missing fields, flag contradictions and duplicates, and finish with items that need human confirmation.
Private by defaultPrompt assembly happens locally in your browser. Avoid placing confidential information into any AI service unless your organization allows it.
From input to outcome
A worked example
See how concrete context turns this recipe into a usable result
Actual input
Source document
[Minutes 8 Aug, item 2] The steering group approved moving the launch from 1 October to 15 October. Maya will update the customer notice by 12 August. The team discussed reducing the pilot group but made no decision. [Item 3] If vendor access is not granted by 20 August, integration testing may slip by one week. Leon will ask the vendor for confirmation; no due date was recorded.
Extraction scope
Only launch and integration items in these minutes; include unresolved items.
Interpretation rules
A decision requires words such as approved or agreed. A risk needs an uncertain event and consequence. An action needs an explicit task; preserve missing fields.
Example output
Decisions: D-01 | Launch date changed from 1 October to 15 October | Approved | 8 August | Steering group | Item 2. The smaller pilot group is not a decision and is listed for confirmation.
Risks: R-01 | Vendor access may not be granted by 20 August | Cause: vendor access pending | Consequence: integration testing may slip one week | Mitigation: confirmation request only; broader mitigation not specified | Owner: not specified | Item 3. Probability is not specified.
Actions: A-01 | Update customer notice | Maya | 12 August | Depends on approved launch date | Status not specified | Item 2. A-02 | Ask vendor to confirm access | Leon | Not specified | None stated | Status not specified | Item 3.
Human confirmation: decide whether to reduce the pilot group; assign the risk owner; set Leon's due date; confirm whether the revised launch date changes other milestones.
Why this works
1
Separate registers stop tentative language from being silently promoted to a commitment.
2
Evidence locators and missing-field labels support review and follow-up.
Check the result
Is every decision expressed as final in the source?
Were any owner, date, probability, or mitigation invented?
Does each extracted item include a usable source locator?
ChatGPT