1
1. Goal
Give the model one clear job
State the role, audience, and outcome in one tight line so the task has a stable center.
Cheat sheetAI & AutomationPromptingFree
GPT-5 Prompt Patterns
A cheat sheet for choosing instruction styles, constraints, and output formats that make advanced model behavior more reliable.
Read time
6 min read
Saved by
1
Updated
Updated Mar 10, 2026
Human accountableModeration reviewed3 sourcesReview recommendedVersion 1
Visual shortcut
The five-part prompt stack
5 visual steps
Build prompts from top to bottom so the model knows the job, the boundaries, the source material, the output shape, and the self-check before it starts.
Stable structureLower reworkBetter downstream reuse
2
2. Guardrails
Set hard constraints early
Define what must be included, avoided, or verified before style instructions start to compete.
3
3. Context
Attach only the context that matters
Use delimited source blocks, examples, or notes that directly change the answer quality.
4
4. Output
Specify the response format
Choose bullets, JSON, tables, or headings so the result is easy to review or reuse downstream.
5
5. Check
Ask for assumptions and risks
A quick final check exposes weak evidence, missing data, and next actions before you trust the answer.
Think of this as a reusable scaffold: when outputs slip, the missing layer is usually obvious.
Instruction setup
01- State the role, objective, audience, and output shape in the first two lines.
- Prefer hard constraints and explicit acceptance criteria over vague style preferences.
- Ask for structured output when the result will be reused downstream.
Reliability boosters
02- Provide a small reference example when tone or structure matters.
- Request assumptions, risks, and next actions to surface hidden gaps.
- Use delimiter blocks for source context so the model can separate instruction from input.
When to avoid overprompting
03- Skip long persona sections unless they change the task meaningfully.
- Avoid contradictory goals such as both maximal brevity and exhaustive detail.
- Remove duplicated instructions before iterating again.
Created by Maya Chen
Keep learning with Maya Chen.
Turns model behavior into practical operating patterns for teams.
Discussion
Questions and field notes
Ask a question, share how you used this guide, or suggest an update.
Join the discussionNo comments yet. Start with a question or a useful example.