---
id: "framework-context-sandwich"
type: "framework"
source_timestamps: ["01:30:00"]
tags: ["prompt-engineering", "best-practices"]
related: ["concept-context-sandwich", "action-build-context-sandwich", "prereq-basic-prompting", "concept-brand-dna"]
steps: ["Who I am (context and background)", "What I need (the specific task)", "What good looks like (criteria for success and formatting)"]
---
# The Context Sandwich (Framework)

## Purpose
A structural template for writing highly effective AI prompts. The operational form of [[concept-context-sandwich]].

## The Three Layers

### 1. Top Bun — *Who I am*
Provide the AI with:
- Background and credentials
- Business model and target audience
- Unique tone of voice — your [[concept-brand-dna]]

### 2. The Meat — *What I need*
Clearly state the specific task: write an email, draft a proposal, analyze data, generate a content outline.

### 3. Bottom Bun — *What good looks like*
Define exact success criteria:
- Format and length
- Hard constraints (e.g., "no jargon," "under 200 words")
- The specific emotional or strategic outcome
- Optional: examples of good and bad outputs

## Reusability
Build the Top Bun *once* (or store it as a Custom GPT / Claude Project / system prompt). Most users only need to write the Meat + Bottom Bun for each new task.

## Adjacent Techniques
- Chain-of-Thought prompting (Wei et al., 2022)
- Few-shot examples
- Tree-of-Thoughts for branching decisions
- Anthropic-style XML-tagged structured prompts

## Prerequisites & Action
- Prereq: [[prereq-basic-prompting]], [[prereq-llm-access]]
- Action: [[action-build-context-sandwich]]