Use ChatGPT to draft tests by pasting the unit under test, the behaviors that must hold, and the test framework you run. Ask for focused cases, including failure paths. Run them. Delete assertions that only mirror implementation trivia. ChatGPT cannot see the rest of your suite unless you paste it. Prefer characterization tests before risky refactors. Keep humans responsible for what production must guarantee. Keep the workflow practical for ChatGPT for Writing Tests: paste evidence, ask for a minimal patch, and verify with tests. ChatGPT only knows what you provide unless a coding tool is connected. An editor assistant can read the workspace for multi-file edits. Separate chat billing from
Prompt for runnable tests
Name the framework: pytest, JUnit, PHPUnit, Jest, or another you actually run. Paste the function or class public API. List behaviors as bullets. Ask for tests that would fail if those behaviors broke. Request factories or fixtures only if your project already uses them; paste a sample if so. Forbid new testing libraries unless you approve them. Keep the advice concrete for ChatGPT for Writing Tests. State constraints, verify locally, and reject invented APIs. Prefer minimal patches you can review. Move multi-file work to an editor assistant that can read the workspace when paste fatigue starts.
- Framework named explicitly.
- Behaviors listed as bullets.
- No surprise test libraries.
Improve AI tests before merging
Watch for tests that assert mocks were called and nothing about outputs. Watch for duplicated setup. Watch for missing edge cases you stated. Ask ChatGPT to add negative cases. Prefer table-driven styles when your stack supports them. Move multi-file fixture work into an editor assistant if paths get messy. Green tests that encode the wrong behavior are worse than no tests. Keep the advice concrete for ChatGPT for Writing Tests. State constraints, verify locally, and reject invented APIs. Prefer minimal patches you can review. Move multi-file work to an editor assistant that can read the workspace when paste fatigue starts.
Keep flaky timing tests out unless you own the infrastructure to stabilize them.
Pair tests with refactors and reviews
Before a ChatGPT-guided refactor, generate characterization tests around current behavior. After code review findings, ask for regression tests that lock the fix. Paste the failing assertion when a new test is red and you want help. Do not let the model weaken the assertion to make green happen. Fix the code or fix the expectation deliberately. Keep the advice concrete for ChatGPT for Writing Tests. State constraints, verify locally, and reject invented APIs. Prefer minimal patches you can review. Move multi-file work to an editor assistant that can read the workspace when paste fatigue starts.
- Characterization before refactors.
- Regression tests after bug fixes.
- Never weaken asserts to pass.
Related on this site
Frequently asked questions
-
Can ChatGPT replace a test strategy?
No. It drafts cases from what you describe. You still choose critical behaviors, risk areas, and coverage goals. Use it to accelerate writing, not to invent product requirements. Keep the answer grounded in what you paste and what you can verify for ChatGPT for Writing Tests. Check official pricing or docs when plans matter. Use an editor assistant when the workspace must be read. Review every generated change before you
-
Why are AI tests sometimes useless?
They often assert implementation details or mocks instead of outcomes. Demand behavior-focused cases. Delete brittle tests that break on every rename without catching bugs. Keep the answer grounded in what you paste and what you can verify for ChatGPT for Writing Tests. Check official pricing or docs when plans matter. Use an editor assistant when the workspace must be read. Review every generated change before you merge it.
-
Should I generate an entire suite at once?
Better to generate for one module or one bug class. Huge dumps hide gaps and create review fatigue. Grow the suite with the riskiest paths first. Keep the answer grounded in what you paste and what you can verify for ChatGPT for Writing Tests. Check official pricing or docs when plans matter. Use an editor assistant when the workspace must be read. Review every generated change before you merge it.
-
How do I handle missing fixtures?
Paste an existing fixture style from your repo. Tell ChatGPT to match it. If fixtures span many files, use an editor assistant to place them correctly while chat shapes the cases. Keep the answer grounded in what you paste and what you can verify for ChatGPT for Writing Tests. Check official pricing or docs when plans matter. Use an editor assistant when the workspace must be read. Review every generated
Sources
Bdeb Technology builds websites, WordPress systems, and tools on top of models like these. A written quote comes back within 24 hours.