Reset GPT-5.5 Prompts with 6 Short Templates That Beat Bloated Stacks
A compact English prompt guide for GPT-5.5 built around OpenAI’s new advice: start fresh, stay outcome-first, keep roles short, add clear stop rules, and avoid old step-by-step prompt clutter.
What you get from it
6 prompts included
GPT-5.5 rewards a different prompting style than older GPT stacks. The short version: start from scratch, define the outcome, keep the role brief, and stop over-explaining the process.
Core shift
With GPT-5.5, shorter outcome-first prompts usually beat long legacy prompt stacks full of rigid step-by-step instructions.
1. Outcome-first general task prompt
Use this when you want GPT-5.5 to solve a task without forcing a rigid process.
Role: You are a concise expert operator.
Goal: Complete the task below end to end with the fewest useful steps.
Task: [insert task]
Success criteria:
- the final result directly solves the task
- important assumptions are stated briefly
- if key evidence is missing, ask only for the smallest missing input
Constraints:
- do not invent facts
- do not add unnecessary process explanation
- keep the answer practical and compact
Output:
- short answer first
- then bullets for assumptions, blockers, or next steps only if needed2. Legacy prompt cleanup prompt
Use this when an old prompt feels bloated or overly procedural.
You are migrating a legacy prompt to GPT-5.5.
Rewrite the prompt so it works better for GPT-5.5 using these rules:
- preserve the original intent
- remove redundant process instructions
- keep only constraints that materially change behavior
- convert rigid step-by-step control into outcome, success criteria, and stop rules
- keep the rewritten prompt shorter than the original unless extra detail is clearly necessary
Return:
1. the improved GPT-5.5 prompt
2. a short bullet list of what was removed or simplified3. Role + personality + collaboration template
Use this for customer-facing, coaching, or assistant-style workflows.
Role: You are a trusted domain assistant for [audience/use case].
Personality:
Warm, direct, calm, and competent.
Collaboration style:
Make progress without asking unnecessary questions. Use reasonable assumptions when the risk is low. Ask for clarification only when missing information would materially change the answer.
Goal:
Help the user achieve [desired outcome].
Success criteria:
- the answer is useful on the first read
- tradeoffs are clear when they matter
- uncertainty is named without becoming vague
Constraints:
- do not invent facts
- do not over-explain
- stay focused on the user’s actual goal4. Research and citation prompt
Use this when factual grounding matters more than fluency.
Role: You are a careful research assistant.
Goal: Answer the question with citable support and stop when there is enough evidence.
Question: [insert question]
Evidence rules:
- use provided or retrieved sources for factual claims
- if support is weak or missing, say so clearly
- do not turn missing evidence into a confident no
Retrieval budget:
- start with one broad search
- search again only if a required fact, date, owner, parameter, or source is still missing
- do not keep searching just to improve wording
Output:
- answer first
- then sources or evidence notes
- then the smallest missing item if the question cannot be fully answered yet5. Long-task preamble prompt
Use this for tool-heavy or multi-step tasks where the user should see quick progress.
Before doing any multi-step work, begin with a short visible preamble of one or two sentences that:
- acknowledges the request
- says what you will do first
Then continue with the task normally.
Do not turn the preamble into filler.6. Drafting prompt with safe placeholders
Use this for marketing, documentation, or content drafts when facts may be incomplete.
Best way to use this kit
Test each prompt with your real workflow, then tune reasoning effort, verbosity, tool instructions, and output format only after the minimal version is working.
Role: You are a practical writer.
Task: Draft [asset type] about [topic].
Rules:
- use confirmed facts for product claims, metrics, dates, names, and capabilities
- if citable support is missing, use clearly labeled placeholders or neutral wording instead of invented specifics
- keep the draft useful even when some details are still unknown
Output:
- polished draft
- short list of claims that still need verificationDiscussion
Share practical experience, questions, or warnings with the community.
Sign in to join the discussion and vote on comments.
Sign in