Use Good Answers As Templates

Use Past Winning Answers as Templates (The Self-Exemplar Clone) (AI Skill)

Overview

Attempting to describe a complex formatting structure in descriptive prose ("Make it have a bold header, then a 3-column table, then two bullet points with emojis, and a code block at the bottom...") is tedious and leaves room for model hallucination.

The Self-Exemplar Clone Protocol leverages the fundamental superpower of Large Language Models: in-context pattern completion. By pasting a past response that you loved and telling the AI to "replicate this exact schema for Topic B", you achieve 100% structural fidelity on Turn 1.

Descriptive Prose vs. Self-Exemplar Cloning

┌─────────────────────────────────────────────────────────────┐
│                 Format Consistency Mechanics                │
│                                                             │
│  Descriptive Prose Specification (High Variance):           │
│  "Write a feature doc like the one you did yesterday with   │
│   a title, bullet points, parameters, and a code block."    │
│  ↳ Inconsistent header sizes, omitted parameter types       │
│                                                             │
│  Self-Exemplar Pattern Injection (Zero Variance):           │
│  <exemplar_format> [PASTE EXACT PAST WINNING POST] </exemplar>│
│  "Generate the entry for [NEW TOPIC] following this schema."│
│  ↳ 100% Identical Visual Hierarchy, Tone, & Depth           │
└─────────────────────────────────────────────────────────────┘

Master Self-Exemplar Prompt Templates

Pattern 1: The Exact Structural Clone Directive

Use when standardizing documentation, case studies, or employee reviews:

I want to write an entry for: **[NEW TOPIC / FEATURE NAME]**.

Below is a gold-standard reference example that I approved earlier. 
Follow its **exact visual layout, heading hierarchy, tone, and depth**:

<exemplar_format>
[PASTE YOUR FAVORITE PAST RESPONSE HERE]
</exemplar_format>

Input Data for [NEW TOPIC]:
[PASTE RAW NOTES / DATA FOR THE NEW ITEM]

Output the new entry strictly matching the exemplar structure.

Real-World Case Study

Scenario: Writing API Documentation for 10 Microservices

The Problem

Writing prose instructions for each API doc resulted in service A having a Markdown table, service B having a bulleted list, and service C omitting error codes entirely.

The Self-Exemplar Fix

The team took the perfectly formatted doc for the UserService and used it as an exemplar template:

Generate the API documentation for `PaymentService`:

<exemplar_format>
### `POST /v1/users`
**Description**: Creates a new user profile and triggers welcome email.

#### Request Headers
| Header | Type | Required | Description |
| :--- | :--- | :--- | :--- |
| `Authorization` | Bearer Token | Yes | Scoped OAuth2 token |

#### Example Payload (JSON)
```json
{ "email": "user@example.com", "name": "Jane" }

Raw details for PaymentService: Endpoint POST /v1/charges, requires Authorization and Idempotency-Key, payload takes amount in cents and currency.

**Outcome**: AI produced a clone of the `PaymentService` documentation with identical tables, formatting, and JSON fences in under 5 seconds.

---

## Summary Best Practices
- **Store winning outputs**: Save 2 or 3 of your all-time favorite AI responses into an `exemplars.md` file for reuse.
- **Enclose in `<exemplar>` tags**: Cleanly isolates the reference pattern from the new prompt data.