When the AI gets it wrong

AI responses fail in predictable patterns. When yours misses, run through these in order:

1. Add context. Did you assume the model knew something it couldn't? Tell it what's missing.
2. Get more specific. Was your prompt vague? Say exactly what you want.
3. Specify the output format. Did the shape come out wrong? State the shape.
4. Break it into smaller steps. Big tasks fail more than small ones. Ask for step 1 only.
5. Correct it directly. "That's not quite right — the issue is X. Try again." Iteration is the work.
6. Check for hallucination. If the model cited something specific, verify it exists before trusting it.
7. Start fresh. Sometimes the conversation has drifted. A new chat with a cleaner first prompt beats fighting the old one.