Prompt optimizer for ChatGPT
Rules traceable to official prompting guidance
GPT models follow instructions literally and punish contradictions: OpenAI's own guide shows conflicting instructions burn reasoning quality, and that three specific prompt reminders lifted SWE-bench scores by close to 20%. Reasoning and standard GPT models need opposite prompt styles.
What the rewrite does
- Branches reasoning models (GPT-5, o-series) from standard ones: brief and goal-led versus explicit steps
- Audits your prompt for contradictions, the #1 documented failure mode
- Converts vague length asks into numeric caps, which GPT-5.1 reliably honours
- Uses OpenAI's canonical structure: role, instructions, format, examples, context
Nothing is invented: missing details become clearly marked placeholders, and we don’t store your prompts. 10,000 free words a month.