# Three Prompting Patterns You Will Actually Use
> LOs: LO-S2-1

## Introduction

In Section 1, you learned the four ingredients of a good prompt — role, task, context, and constraints [S1.C1]. That structure gets you most of the way there. But sometimes you write a careful prompt and the answer still comes back as a wall of text, skips a key step, or arrives in a format you can't share. That's not a flaw in your ingredients; it's a missing pattern on top. This class introduces three small patterns you can layer onto any well-formed prompt to fix the most common frustrations [slide 1].

## Core Concept

The three patterns are step-by-step reasoning, few-shot examples, and output formatting. Each one targets a different failure mode, and you can combine them when a task needs more than one (see Slide 2).

| Pattern | Use when | Example phrase |
|---|---|---|
| Step-by-step | The task has multiple stages and you don't want the model to skip ahead | "Walk me through this step by step" |
| Few-shot | You have a specific style, tone, or framing in mind | "Here are two examples — now do the same for…" |
| Output formatting | The answer will be shared, scanned, or pasted into another tool | "Return a table with columns X, Y, Z" |

A simple heuristic helps you choose. Need complex thinking? Ask for steps. Need a specific style? Show examples. Need downstream usability? Lock the format. These aren't mutually exclusive — a quarterly board briefing might combine few-shot (for tone) with output formatting (for structure). The point isn't to memorize rules; it's to recognize which lever fits the moment.

## Worked Example

Let's walk through two of the four role examples on Slide 3 to see this in action.

The **manufacturing engineer** uses step-by-step reasoning: "Walk me through, step by step, the likely root causes of a recurring flash defect on cavity 3 of a polypropylene mold." Why this pattern? Root-cause analysis is inherently multi-stage — gate geometry, clamp tonnage, melt temperature, cooling — and asking for steps stops the model from jumping straight to one guess. You get a chain of reasoning you can interrogate.

The **BioTech CEO** uses few-shot examples: "Here are two examples of a one-paragraph board briefing — concise, risk-aware, ending in a decision ask. Now write one for our Phase 2 enrollment update." Why few-shot? Board tone is impossible to describe in the abstract. Two short examples teach the model the cadence — clipped sentences, calibrated risk language, a clear ask at the end — far more efficiently than a paragraph of style instructions. The Data & ERP analyst, meanwhile, reaches for output formatting (a table with named columns), and the Agile PM uses step-by-step for sprint planning. Same four-ingredient foundation underneath; different pattern on top, fitted to the job (see Slide 3).

## Common Pitfalls

- **Forcing step-by-step on simple one-shot tasks.** If you ask "What's the capital of France, step by step?" you'll get padding. Reserve this pattern for genuinely multi-stage thinking. Single-fact lookups don't need a procession.
- **Asking for few-shot when you have no real examples.** If you fabricate weak examples on the fly, the model will faithfully imitate their weakness. Either find two genuinely good samples or skip this pattern and describe the style instead.
- **Over-tightening the output format.** A rigid schema like "exactly five bullets, each under ten words" can force the model to truncate genuinely important content or invent filler to hit the count. Specify the structure, but leave room for completeness — "a table with these columns, as many rows as needed" is usually the right balance.

## Recap

You can now identify when to reach for step-by-step reasoning, when to show few-shot examples, and when to lock down an output format — and you've seen that they layer cleanly on top of the four-ingredient prompt from Section 1 (see Slide 5). Next up in [S2.C2], we'll tackle the inevitable moment when the first answer still isn't quite right — how to nudge, refine, and recover without throwing your prompt away and starting over.