---
id: "framework-context-sandwich"
type: "framework"
source_timestamps: ["02:11:12"]
tags: ["prompt-engineering", "tactics"]
related: ["concept-context-sandwich", "action-use-context-sandwich", "entity-igor-pogany"]
steps: ["Top Bun — Who I am: deep context (role, business, audience, situation).", "Meat — What I need: the specific task, problem, or question.", "Bottom Bun — What good looks like: format, tone, length, constraints."]
---
# The Context Sandwich Prompting Framework

## Purpose

A three-step framework for structuring prompts to get **high-quality, personalized outputs** from Large Language Models without needing advanced technical skills. Concept note: [[concept-context-sandwich]]. Author: [[entity-igor-pogany]].

## The Three Steps

### 1. 🥖 Top Bun — Who I Am
> Provide deep context about your **role**, your **business**, your **audience**, and your **current situation**.

Example: *"I am the founder of a 7-person B2B SaaS company selling HR analytics to mid-market firms. Our buyer is a VP of People at a 500–2,000 person company. We're preparing for a Series A in Q1."*

### 2. 🥩 The Meat — What I Need
> Clearly state the specific task, problem, or question you want the AI to address.

Example: *"Draft a cold email opener that earns a meeting with the VP of People."*

### 3. 🥖 Bottom Bun — What Good Looks Like
> Define the success criteria: **format, tone, length, and any specific constraints**.

Example: *"Output 3 variants. Each ≤90 words. Conversational, slightly irreverent. No buzzwords like 'leverage' or 'synergy'. Open with a specific observation, not a compliment."*

## Why It Works

- Top Bun grounds the model in *your* world.
- Meat focuses computation on a single, precise intent.
- Bottom Bun collapses the variance space of acceptable outputs.

This is the core day-to-day technique for Level 2 fluency in [[concept-ai-fluency-levels]].

## Practice It

See [[action-use-context-sandwich]].

## Adjacent Research

A layperson distillation of **Chain-of-Thought + few-shot prompting** (Wei et al., 2022).

## Source

Timestamp 02:11:12.