Recommended workflow
User Story / Requirement
↓
Acceptance Criteria
↓
Business / System Context
↓
Generation Prompt
↓
Candidate Test Cases
↓
QA Review & Edit
↓
Approved / Rejected
↓
Test Management / Automation
What context should be provided
- User story
- Acceptance criteria
- Business rules
- Roles/permissions
- Known dependencies
- Data constraints
- Supported platforms
- Relevant non-functional needs
A practical prompt structure
Role
+
Business Context
+
Requirement
+
Testing Objective
+
Constraints
+
Expected Output Format
Example instruction
Act as a senior QA analyst. Review the supplied user story and acceptance criteria. Generate positive, negative, boundary, permission, and integration scenarios. Do not invent unsupported business rules. Include expected result and requirement traceability for each case.
Generate multiple test dimensions
AI tends to produce obvious happy-path cases unless instructed otherwise. Explicitly request coverage dimensions such as:
- Positive
- Negative
- Boundary
- State transition
- Role/permission
- Data
- Integration
- Error handling
- Recovery
Do not over-generate
More cases do not automatically mean more coverage. Large AI-generated sets often contain duplicates and low-value variations.
Generate a manageable candidate set, review it, then expand only where coverage gaps remain.
Preserve traceability
Where possible, keep the relationship between source requirement, prompt or generation request, model/version, reviewer decision, and final retained test case.