GPT-5.5 Prompt Guidance Primitives

Source: https://developers.openai.com/api/docs/guides/prompt-guidance, accessed 2026-07-09.

Use these as checks, not a required structure.

Check

• Outcome: states the user-visible result.
• Success: defines done.
• Constraints: gives priority when rules conflict.
• Context: says what to use, ignore, retrieve, or treat as stale.
• Evidence: requires sources or placeholders for factual claims.
• Output: defines sections, length, tone, and format.
• Validation: says what to check, or how to explain skipped checks.
• Stop rules: says when to ask, retry, fallback, abstain, escalate, or stop.
• Tools: states allowed tools, side effects, approvals, and completion checks.
• Responses workflow: preserves preambles, phase, and assistant-item replay rules only when relevant.

No-Op Detection

Flag a line as a no-op when removing it would not change behavior, output, validation, tool use, evidence, or rule priority.

Common no-ops:

• broad quality words: good, best, high quality, robust
• intensity words: very, extremely, thoroughly, carefully
• duplicated rules
• motivational filler
• process rules with no output or decision effect
• legacy rules copied from older prompts without a GPT-5.5 reason

Do not flag short rules that change behavior, such as a required format, tool, source, constraint, or stop rule.

Reduce

• Put outcome before process.
• Keep detail only when it changes behavior.
• Merge overlapping rules.
• Replace vague terms with observable checks.
• Delete unsupported product facts, metrics, tools, and model claims.
• Add Role, Personality, Goal, Success, Constraints, Output, or Stop rules sections only when they change behavior.