AntiPattern Logo
AntiPatternAI UX Lab
Back to catalog
Control & Agency

Format Refusal (Lazy Output)

The AI ignores specific formatting instructions (JSON, Table, Markdown) and outputs conversational text.

3 principles3 checklist items2 simulation modes

What you will learn

Respect Constraints
Structured Output
No Chatter
Run the simulation to see the failure in action.

Interactive Lab

Toggle between Bad and Good modes.
Anti-pattern mode
Simulation Status: Idlev2.4.0

Press play to run the simulation

Why it fails

Developers often use AI to generate structured data for other systems. When the AI ignores 'Output as valid JSON' and instead gives a chatty preamble ('Here is the JSON you asked for...') or uses the wrong delimiters, it breaks pipelines and wastes user time.

Root cause

RLHF training prioritizes conversational helpfulness over rigid adherence to formatting constraints. System prompts are not strict enough about 'no chatter'.

How to fix it

Respect Constraints
Structured Output
No Chatter

Implementation checklist

  • Support a 'JSON Mode' or structured output flag
  • Recognize format keywords (JSON, CSV, Markdown) and suppress conversational filler
  • Validate output schema before sending