Master AI Testingwith Better Checks
Learn how to define focused checks that produce useful results with proven templates and best practices.
Writing Useful Checks
Clear objectives and setup details help the QA agent return useful results:
Be Specific and Clear
Provide detailed instructions, including the goal of the test, the environment URL, user roles, and any necessary setup data.
Good Prompt:
"As a logged-in user, I should be able to add a product to my cart and see the correct item and price in the cart."Bad Prompt:
"Test the shopping cart"Define Explicit Pass/Fail Criteria
Clearly state what success and failure look like. This helps the agent determine the test outcome accurately.
Good Oracle:
"The user should see a confirmation message: 'Product added to cart'. The cart total should be $29.99."Bad Oracle:
"The cart should be updated."Specify Artifacts to Collect
Tell the agent what evidence to gather, such as screenshots, DOM snapshots, or console logs.
"Verify the login page UI. Collect a full-page screenshot and any console errors."Do's
- ✓Specify environment, user role, and entry URL
- ✓Provide deterministic data or setup steps
- ✓Define pass/fail oracles and artifacts to collect
Don'ts
- ✕Combine unrelated flows in a single check
- ✕Omit authentication details
- ✕Request massive coverage with no timebox
Proven Templates
Copy and customize these templates for your specific testing scenarios
Generate and run an end-to-end browser test
Create and execute a stable E2E test for user flows
API test for a critical endpoint with auth
Validate API endpoints with authentication and error handling
Regression suite update for a changed feature
Update regression tests after feature changes
Bug reproduction and triage
Reproduce and minimize bugs for development
Release smoke checklist (pre-prod)
Produce and execute smoke tests for releases
Exploratory testing charter
Run exploratory testing sessions on new features
Accessibility sanity pass
Check critical pages for WCAG AA compliance
Localization checks
Validate translations and internationalization
Core Principles
Follow these principles to create focused, actionable checks
Be Goal-Driven
State the business outcome first (e.g., "validate checkout for logged-in users")
Scope Tightly
One flow or risk area per check to reduce ambiguity and flakiness
Ground with Context
Include environment, auth, data setup, and dependencies
Define Oracles
Specify what a "pass" looks like (UI text, API status, side-effects)
Request Artifacts
Ask for screenshots, DOM snapshots, logs, network traces, and reports
Iterate Quickly
Use focused checks per task and attach failure details for follow-ups
Troubleshooting
Common issues and how to resolve them
Tests are flaky
Reduce scope, fix preconditions, and specify stable selectors. Ask the agent to prefer role/label-based selectors.
Agent misunderstands
Add example data and a single golden pass criterion to clarify intent.
Results are incomplete
Request a second pass limited to the missing sections only.