4 Best AI Prompts for Writing Unit Tests (2026)

Here is the code to test:
[PASTE CODE]

Before writing any tests, list every observable behavior a caller relies on: inputs → outputs, side effects, thrown errors, and ordering guarantees. Format: behavior | trigger | expected result | how likely a refactor breaks it (high/med/low). Include behaviors that are implied but not obvious from signatures.

For each high-risk behavior above, enumerate edge cases: empty/null/undefined, boundaries (0, 1, max, max+1), unicode and whitespace strings, concurrent calls, and invalid types. Output a table: case | input | expected. Mark which cases the current code probably FAILS — be honest.

Write the tests in [FRAMEWORK] covering every behavior and edge case above. Rules: one behavior per test, names that describe the scenario ("rejects_expired_token_with_401" not "test_auth_2"), arrange-act-assert structure, no mocking anything you don't own, and no testing private internals.

Now act as a mutation tester: list 5 specific small bugs you could introduce into the original code (flip a comparison, off-by-one a boundary, swallow an error) and for EACH one say which test would catch it — or admit none would. Write the missing tests for any survivors.