How to Format a Prompt Before Using an AI Model

Published March 11, 2026 · Updated March 12, 2026 · 4 min read

A well-formatted prompt isn't necessarily shorter. It's clearer about what you want, in what order, and under what constraints. The same request written as a wall of text versus written with labelled sections can produce meaningfully different output quality — especially for complex, multi-part tasks.

Use labelled sections

Labelled sections help the model identify what each part of your prompt is for. Common conventions:

## Context
You are reviewing pull requests for a Python codebase following PEP 8.

## Task
Review the following code diff and identify any bugs or style violations.

## Format
Return a bulleted list. Include line numbers where relevant.
If no issues found, say "Looks good."

## Code diff
[paste here]

The model can identify where the context ends and the task begins, rather than having to infer structure from a continuous block of text. You don't need to use Markdown headers — even labels like TASK: and FORMAT: in all-caps work. The key is unambiguous separation between different types of content.

Instructions first, examples after

If your prompt includes examples, put the instructions before them. Models generalise from examples — if the examples appear first without context, the model may treat them as definitions rather than illustrations. "Do X — here's what good X looks like" outperforms "here's what X looks like — do X."

Constraints work best stated near the end, just before the content being processed. "Do not include code examples — [user question]" is more reliably followed than "[user question] — and by the way, don't include code examples."

Handle variable parts explicitly

If you reuse a prompt template with different inputs, make the variable part obvious:

## Task
Classify the following customer review as Positive, Negative, or Neutral.
Return only the classification word, nothing else.

## Review
{{REVIEW_TEXT}}

Clear placeholder markers — {{REVIEW_TEXT}}, [USER_INPUT], <document> — make the template maintainable. When you come back to it in two weeks, you immediately know what changes per request and what stays fixed.

For long system prompts: format the system prompt as a document, not as a paragraph. Use headers for major sections (Role, Constraints, Output Format, Examples). This structure also helps when you need to update the prompt — you can locate and modify specific sections without searching through a wall of text.

Cleaning before reuse

Prompts accumulate mess through iteration — extra blank lines, duplicated instructions, clarifications added after failed tests. Run a prompt through a formatter before saving it as a template. Clean whitespace and consistent structure make the template easier to read and modify. A messy template produces unpredictable results when the underlying model updates change how it handles ambiguous formatting.