The Anatomy of a Good Prompt

Six building blocks

Reliable prompts usually contain: (1) a role or perspective ("support analyst"); (2) the task in one clear sentence; (3) the audience and purpose; (4) rules and constraints (length, tone, what to avoid, what to do when unsure); (5) the output format; and (6) the input itself, clearly separated from the instructions. Not every prompt needs all six, but checking this list quickly finds what is missing when output disappoints. Using a template also makes prompts consistent across a team.

A reusable template, run

I ran this plain-Python (standard library only) example. The template fills in role, task, audience, rules, format and input, and prints the final prompt (259 characters, about 65 tokens by the rough 4-characters-per-token rule).

from string import Template

PROMPT = Template("""You are a $role.
Task: $task
Audience: $audience
Rules:
$rules
Output format: $fmt

Input:
$text""")

rules = ["Use only the input text.", "Max 2 sentences.", "If unsure, say 'unknown'."]
p = PROMPT.substitute(
    role="support analyst",
    task="summarise the customer's complaint",
    audience="a busy manager",
    rules="\n".join(f"- {r}" for r in rules),
    fmt="plain text",
    text="My order 4821 arrived late and the box was damaged.",
)
print(p)
print("characters:", len(p), "| rough tokens:", round(len(p) / 4))

Output:

You are a support analyst.
Task: summarise the customer's complaint
Audience: a busy manager
Rules:
- Use only the input text.
- Max 2 sentences.
- If unsure, say 'unknown'.
Output format: plain text

Input:
My order 4821 arrived late and the box was damaged.
characters: 259 | rough tokens: 65

Quick check: Why separate the input from the instructions?

• Because models cannot read inputs
• To hide the prompt
• To save a token
• So the model can tell what to do from what to process