Guide8 min readUpdated Feb 2025
How to Write Better AI Prompts: 7-Step Guide
A simple framework to get reliable outputs from Claude, ChatGPT, and Gemini—plus examples you can copy and tools to polish them.
Role + tone + structureExamples & variablesSuccess criteria
Speed up formatting
Polish outputs with microtools.
Step 1
Define the goal and audience
State the exact outcome, audience, and constraints. Example: “Summarize this sales call for execs with 3 action items.”
Step 2
Set the role and tone
Assign a role (support rep, PM, copywriter) and tone (professional, friendly). Tone tools help keep it consistent.
Step 3
Provide structure
Ask for bullets, JSON, or sections. Structure reduces variance and makes outputs scannable.
Step 4
Add examples or variables
Few-shot examples or placeholders ({{audience}}, {{cta}}) make outputs repeatable.
Step 5
Constrain length and focus
Word counts, bullet limits, or time windows (30-60s scripts) keep outputs usable.
Step 6
Include success criteria
State what “good” looks like (1 CTA, no jargon, cite source). Helps the model self-check.
Step 7
Review and iterate
Run, skim, tighten. Use a tone converter or formatter tool to finalize quickly.
Example prompt template
You are a {{role}} writing for {{audience}}.
Goal: {{goal}}
Tone: {{tone}}
Format: {{format}} (bullets/json/sections)
Length: {{length}} words max
Include: {{must_haves}}
Exclude: jargon, multiple CTAs
Provide: 2 options and a 1-line CTA.Common mistakes
- Vague goals (“write an email”) instead of specific outcomes.
- No structure—forgetting bullets/JSON/sections leads to rambling outputs.
- Multiple CTAs; keep one action per prompt.
- Skipping review—always skim and tighten tone before sending.
FAQ
Which model should I start with?
Start with a fast, cost-effective model (Claude Haiku or GPT-4o mini), then upgrade to Claude Sonnet/GPT-4o for nuance.
Do I need examples?
Examples help. Even one short “good output” improves consistency. Use variables to keep them reusable.
How do I control length?
Specify word count, bullets, or sections. The more specific, the more consistent the output.
What about safety/tone?
Set tone and audience explicitly, and review outputs. Use a tone converter to normalize voice before sending.