---
id: "concept-context-sandwich"
type: "concept"
source_timestamps: ["02:11:12"]
tags: ["prompt-engineering", "tactics", "communication"]
related: ["framework-context-sandwich", "action-use-context-sandwich", "concept-ai-fluency-levels", "entity-igor-pogany"]
definition: "A prompting structure that surrounds a specific task request with deep personal context and clear criteria for what a successful output looks like."
sources: ["day1"]
sourceVaultSlug: "ai-advantage-summit-2026-2026Apr28"
originDay: 1
---
# The Context Sandwich

## Definition

A prompting structure that surrounds a specific task request with deep personal context and clear criteria for what a successful output looks like — taught by [[entity-igor-pogany]].

For the step-by-step framework, see [[framework-context-sandwich-d1]].
For the action prompt, see [[action-use-context-sandwich-d1]].

## The Three Layers

### 🥖 Top Bun — Who I Am
Deep context about your role, your business, your audience, your current situation, your constraints.

### 🥩 The Meat — What I Need
The specific task, question, or problem you want the AI to solve. Crisp, single-intent.

### 🥖 Bottom Bun — What Good Looks Like
The specific format, tone, length, structure, and constraints for the desired output. Examples or anti-examples are powerful here.

## Why It Works

By sandwiching the task between **rich context** and **clear success criteria**, the LLM has its solution space dramatically narrowed. The result: highly relevant, personalized outputs without requiring formal prompt engineering training.

This is the operational engine of Level 2 in [[concept-ai-fluency-levels]].

## Adjacent Research

Structurally related to **Chain-of-Thought + few-shot prompting** (Wei et al., 2022), which boosts LLM accuracy by 20–50% via structure. The Context Sandwich is a layperson-friendly distillation.

## Source

Timestamp 02:11:12 — Igor Pogany.


## Related across days
- [[framework-context-sandwich-d3]]
- [[action-use-context-sandwich-d3]]
- [[arc-context-sandwich-two-authors]]