Use File > Print or the button above to save as PDF.
Write prompts that actually work.
Five patterns that separate useful AI output from generic filler. Each prompt is annotated with the techniques that make it work.
Write a short follow-up email to a marketing director named Sarah, after a discovery call about her team's Q3 content strategy. Our tone is warm but direct - end with one clear next step.
Output format
Specifying length and purpose prevents generic, overlong responses.
Context
The AI only knows what you tell it. Name specifics.
Constraint
One clear constraint sharpens the output and prevents vague endings.
The first output is not the final output.
I need to create a creative brief for a new client campaign. Ask me questions one at a time until you have everything you need to write it. Start with your first question.
Flipped interaction pattern
Let the AI figure out what it needs to know from you.
One at a time
Prevents a wall of questions. Keeps the exchange conversational.
Stopping condition
Tells the AI when to stop asking and start producing output.
You don't have to type everything out.
Summarize this meeting using this exact format: Client: <name> Key concerns: <2-3 bullets> What we committed to: <1-2 bullets> Next step: <single action with owner and date> Do not include anything that wasn't explicitly discussed.
Output format
Locks the structure so the AI matches your exact layout every time.
Template pattern
Angle brackets define what goes where and how much to include.
Constraint / hallucination guardrail
Stops the AI from filling gaps with invented information.
The template is yours to modify.
Act as a skeptical creative director with 15 years of brand experience working with marketing agencies. Review this tagline: [tagline] Tell me what's weak about it first, then what's working. Be direct - don't soften the critique. Use plain language, not marketing jargon.
Persona pattern
A role changes vocabulary, confidence, and what the AI notices.
Sycophancy
Flipping order to weaknesses-first overrides the default praise.
Explicit voice instructions
Explicit voice instructions override the AI's polite defaults.
When you assign a persona, you're setting four things.
You help [Company] turn client meeting notes into proposal outlines. [Company] is a marketing agency that embeds into client teams. Our tone is professional but direct. Never use jargon. Always ask one clarifying question before you start. Never invent details not in the transcript - flag gaps instead. Special commands: /outline - produce the proposal outline /alternatives - give me two different framings
Persistent instructions
The intern briefing, written once and applied to every conversation.
Embedded patterns
Baked-in behaviors that apply automatically, regardless of who types.
Menu commands
Shortcuts your team can use without knowing how to prompt.
Everything before this was a conversation. This is a tool.
The pattern behind the patterns
Every example on this page runs on the same engine: context the AI cannot guess, the shape you want back, and what counts as good. The difference between useful and generic is not clever phrasing. It is what you put in. Pick the pattern closest to your task, copy the prompt structure, and replace the specifics. When the output is close but not right, say what is wrong and run it again. That is the whole skill.