AI in testing does not replace the tester, but helps complete some everyday tasks more effectively.
Do not start with "let AI test the feature for me." Take a small, safe piece of work: an anonymized requirement, a checklist, an API response example, or an old bug report. The goal of the first attempt is to get a draft you can review, not a finished solution.
A safe first prompt
A useful answer does not require a long prompt. Four clear parts are usually enough.
Built prompt example
An AI answer is a draft, not a finished artifact. First check that it answers the task, then verify facts, coverage, and data.
Check the task
Make sure the answer matches your request: the right format, the right level of detail, and the artifact you asked for.
Verify the source
Compare facts with the requirement, API contract, log, or bug report. Roles, statuses, fields, and rules should be supported by the source material.
Mark assumptions
Keep new ideas that are not in the source as hypotheses, risks, or questions for the team. Do not turn model guesses into requirements.
Check testability
Every item should be testable: the action, expected result, data, and condition should be clear. Rewrite items that are too broad.
Complete coverage
Check whether there are enough negative scenarios, boundary cases, access rights, errors, user states, and important risks.
Check sensitive data
Make sure the prompt and answer do not contain passwords, tokens, personal data, internal links, or private documents.
