Fix Invalid JSON from AI
The JSON won't parse and you can't see why. Deterministic cause-sniffing — trailing commas, single quotes, unclosed brackets — and the repair prompt that fixes it.
Overview
Broken JSON from a model fails silently in the worst way: the response looks complete, and the parser just says no. This setup loads the classic case — a contract extraction with a trailing comma before the closing brace — and shows the validator's parse-failure path: JSON.parse fails, the cause-sniffer identifies the likely culprit deterministically (trailing comma, single-quoted keys, or an unclosed bracket), and the repair prompt tells the model exactly what to fix while keeping every extracted value unchanged.
Workflow
-
Paste the unparseable response
The validator attempts a real parse, then sniffs the cause when it fails.
-
Check the likely cause
Trailing comma, single quotes, unclosed bracket — named deterministically, not guessed.
-
Repair, don't regenerate
The repair prompt keeps every correct value and fixes only the syntax — regeneration would re-roll the data too.
Why This Works
- Cause-sniffing turns an opaque parse error into an actionable fix
- Repairing preserves extracted data that regeneration would scramble
- A 50-point parse penalty makes broken syntax unmissable in the score
Best for
- Developers staring at "Unexpected token" errors
- Pipelines that store raw model output for later parsing
- Anyone whose extraction came back 99% right and 100% unparseable
Not for
- Preventing invalid JSON up front — that's the JSON Output Prompt Builder's strictness rules
- Repairing the JSON by hand — the repair prompt makes the model do it
Use cases
- Diagnosing why a model's JSON fails JSON.parse
- Identifying trailing commas and quote problems instantly
- Repairing the response without losing the extracted values