Resource
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.
Fix AI Format Drift
Run 40 looks nothing like run 1: sections reordered, one came back empty. Detect the drift, repair the run, keep the pipeline honest.
Overview
Format drift is the recurring-workflow disease: the same prompt that produced perfect output in week one starts reordering sections and leaving them empty by week six. This setup validates a drifted incident summary — Timeline and Root Cause swapped, Impact present but empty — the two drift signatures that presence-only checks miss completely. The repair prompt fixes the specific run; the validation history tells you when the underlying prompt needs a strictness upgrade.
How to use this resource
-
Validate a sample of runs
Drift is statistical — one validation catches the run, a pattern of warnings catches the trend.
-
Note the drift signatures
Reordered sections and empty-but-present sections are drift; missing sections are breakage. Different fixes.
-
Repair the run, harden the prompt
The repair prompt fixes today's output; recurring warnings mean the generator prompt needs Strict consistency.
Why This Works
- Order and emptiness checks catch the drift presence-only QA misses
- Per-run repair keeps the pipeline flowing while you fix the root cause
- Validation history turns anecdotal "it got worse" into evidence
Best for
- Recurring workflows running the same prompt for months
- Outputs consumed by position-sensitive readers or parsers
- Teams that QA samples rather than every run
Not for
- Preventing drift at the source — strictness levels in the generator tools do that
- Comparing two prompt versions — that's the Prompt Version Diff
Use cases
- Spot-checking recurring outputs for structural drift
- Catching empty sections that presence checks pass
- Deciding when a prompt needs stricter consistency rules
FAQ
How does this catch an empty Impact section when my validation already checks all four sections are present?
Presence-only checks pass because the 'Impact' header exists — this setup flags it as a distinct drift signature: 'Section Impact is empty or nearly empty'. The Fix instruction is to fill it or state explicitly there is nothing to report. The ai-output-validator surfaces that emptiness alongside order problems, catching what a header-exists check waves through.
Why does the repair prompt say to return only the corrected summary with no code fences?
So the output drops straight back into a position-sensitive parser or reader without cleanup. The OUTPUT RULES state 'Return only the corrected summary — no explanations, no commentary, no code fences' and 'Keep everything that was already correct unchanged'. Explanations or ``` fences would break the same downstream consumer the repair is meant to protect, and you run this through your own assistant.
When should I stop repairing each run and instead fix the underlying prompt?
Repair the run when it is an occasional slip; harden the generator prompt when 'Root Cause appears before Timeline' or an empty Impact shows up as a recurring warning across sampled runs. The repair prompt fixes today's output, but a pattern of the same drift signatures is your signal to raise the generator to Strict consistency — that prevention lives in the tools, not here.
More resources from AI Output Validator
Resource
Validate AI Output — Catch Format Violations
Paste the response, get the verdict: real JSON parsing, missing-field detection, and a repair prompt for everything found.
Resource
Agent Evaluation Scorecard Prompt
Grade agent output the same way every time — a rubric scoring correctness, grounding, safety, tone, and completeness, with a pass threshold instead of a gut call.
Resources that pair well
Resource
Force JSON Output from AI
Stop getting 'Sure, here is the JSON…' — the output-contract pattern that forces models to return only parseable JSON: schema, example, and a strict rule block.
Resource
Extract Data From Text with AI
Free text in, named fields out. The extraction prompt pattern that turns any unstructured text into consistent, parseable records.
Resource
Classify Support Tickets with AI
Billing, Technical, Account, How-To, Feature Request — ticket triage with definitions that decide the borderline cases for the model.
Related tools
Tool
AI Output Validator
Paste an AI's output and validate it against the expected format — with a repair prompt for every failure found.
Guides for this resource
Guide
How to Keep AI Output Format Consistent
The AI nails your format once, then the next output adds a "Sure, here's…" wrapper, renames a field, turns the table into bullets, and drops a section. Here's how to write the format as a contract and catch the drift when it happens anyway.