ChatGPT Prompting
How To Prompt ChatGPT 5.5
If you have spent the last year perfecting a detailed prompt library, some of it may now be working against you. ChatGPT 5.5 works best when you describe the outcome, not every step.
Back to resourcesThe core shift is simple: ChatGPT 5.5 is outcome-first.
Describe what done looks like. Stop describing every step the model should take to get there.
Start With The Outcome
Older prompt libraries often tried to control the model tightly: role, steps, constraints, formatting rules, examples, warnings and edge cases all stacked into one long instruction.
For ChatGPT 5.5, that can add noise. Instead, tell it what successful completion looks like.
Ask yourself: what should be true when this task is done?
One
Start Fresh
Do not automatically carry over a detailed prompt stack from an older model.
Long inherited prompts can narrow the model's thinking, especially when they contain old workarounds that the newer model no longer needs.
Start with the outcome, test the response, then add only the instructions that clearly improve the result.
Two
Use Decision Rules
Drop the heavy ALWAYS and NEVER language.
Use decision rules instead.
For example:
If the task is strategic, challenge assumptions before giving recommendations.
If the task is executional, prioritise clear next actions.
If confidence is low, say what needs to be verified.This gives the model room to choose the right behaviour for the situation.
Three
Separate Personality From Collaboration Style
Personality is how it sounds.
Collaboration style is how it works.
Write them as two labelled sections instead of one mixed paragraph.
Personality:
Direct, calm and practical. Avoid hype.
Collaboration style:
Challenge weak assumptions, explain trade-offs and ask clarifying questions when the brief is missing something important.Four
Default To Plain Paragraphs
Do not force heavy formatting into every answer.
If your old prompts mandate bullet points, tables, headings and frameworks for everything, you may be working against the model.
Use plain paragraphs as the default. Ask for bullets, tables or sections only when the output actually needs them.
Five
Test Medium Reasoning First
Medium reasoning is the recommended default for most work.
Do not max out reasoning for every task. More is not always better if the job is simple, fast or clearly scoped.
Test medium first. Increase reasoning only when the task genuinely needs deeper planning, complex trade-offs or careful analysis.
Final Thought
The better ChatGPT gets at understanding intent, the less you need to micromanage every step.
Write the outcome clearly. Add decision rules where they matter. Keep the format simple. Then let the model do more of the work.