Tool 01 — live
PROMPT GENERATOR Free · CO-STAR, few-shot & chain-of-thought
A simple structure for turning vague instructions into useful prompts. Instead of "write about diabetes" — tell the model who you are, what you need, how it should sound, who is reading, and what format you want.
The framework
CO-STAR — six fields, one good prompt
The most complete of the five. Use it when the output has to land right first time — a brief, a report, anything going to other people.
THE CO-STAR FRAMEWORK
LIVE CO-STAR BUILDER — FILL EACH FIELD AND WATCH YOUR PROMPT ASSEMBLE
→ YOUR ASSEMBLED PROMPT
The framework
Zero-shot — ask once, no examples
Just the instruction. Fastest to write and the right default for common tasks a model already understands — summarise, translate, rewrite, classify. If the output drifts, add an example and move to one-shot.
ANATOMY OF A ZERO-SHOT PROMPT
LIVE ZERO-SHOT BUILDER
→ YOUR ASSEMBLED PROMPT
The framework
One-shot — show it exactly one example
Describing a format in words is harder than demonstrating it. One worked example pins down structure, length and voice far more reliably than another paragraph of instructions.
ANATOMY OF A ONE-SHOT PROMPT
LIVE ONE-SHOT BUILDER
→ YOUR ASSEMBLED PROMPT
The framework
Multi-shot — a pattern, not a one-off
Two to five examples let the model infer the rule rather than copy a single case. Worth the extra typing for classification, structured extraction, and any format with edge cases. Cover a tricky case in one of them — that is where the value is.
ANATOMY OF A MULTI-SHOT PROMPT
LIVE MULTI-SHOT BUILDER
EXAMPLES
→ YOUR ASSEMBLED PROMPT
The framework
Chain-of-Thought — make the reasoning visible
For problems with steps: sums, logic, diagnosis, anything where a confident wrong answer is expensive. Asking for the working makes errors visible and usually makes the final answer better too.
ANATOMY OF A CHAIN-OF-THOUGHT PROMPT
LIVE CHAIN-OF-THOUGHT BUILDER
→ YOUR ASSEMBLED PROMPT
WHICH ONE SHOULD I USE?
- Start zero-shot. It is the cheapest to write. Only add machinery when the output is actually wrong.
- Wrong format? Move to one-shot — showing beats describing.
- Inconsistent across inputs? Move to multi-shot so the model infers the rule.
- Wrong reasoning? Use chain-of-thought and read the working.
- High-stakes writing? Use CO-STAR, which pins down audience and tone as well as the task.