AI assistants are excellent debugging partners, but the quality of help depends on the information you give. Treat the AI like a colleague who can't see your screen.
Give the full picture
- The exact error message and stack trace
- The relevant code (not the whole project, not just one line)
- What you expected and what actually happened
- Language, framework and versions
- What you've already tried
A debugging prompt template
I'm using [language/framework + version].
Expected: [behavior]. Actual: [behavior].
Error: [paste error and stack trace]
Code: [paste relevant code]
I've tried: [attempts].
Explain the likely cause first, then suggest a minimal fix.Ask for hypotheses, not just fixes
For tricky bugs, ask: "List the 3 most likely causes and how I can test each one." This turns the AI into a guide for your own investigation.
Add logging
Ask the AI where to add log statements or breakpoints to narrow down the problem. Paste the output back for the next round.
Verify the fix
- Run the code and reproduce the original scenario.
- Ask for a test that would have caught the bug.
- Make sure the fix doesn't hide the error (like an empty try/catch).
Protect secrets
Remove API keys, passwords and customer data before pasting code or logs.