โ† All Techniques
๐Ÿฆด

Skeleton of Thought โ€” Prompting Guide & Examples

Skeleton of Thought asks the AI to first generate a high-level outline (skeleton) of the answer, then expand each section in parallel or sequentially. This produces better-structured, more comprehensive responses and can dramatically speed up generation.

How It Works

Two phases: (1) Generate a concise outline with section headers and one-line descriptions, (2) Expand each section into full content. Since sections are independent, they can theoretically be generated in parallel for faster results.

When to Use

Use Skeleton of Thought for long-form content, documentation, articles, reports, and any structured output. Especially useful when you need comprehensive coverage and don't want the model to forget important sections midway through.

Model-Specific Tips

ChatGPT / GPT-4

GPT-4 handles Skeleton of Thought well. Generate the outline first, then expand in follow-up messages for best results. Can also do both in one prompt.

Claude

Claude excels at outline-first approaches. Its long context window allows generating both skeleton and full expansion in a single response without losing coherence.

Gemini

Gemini supports this technique well. Use Gemini 1.5 Pro's large context for comprehensive outlines. Expand sections in follow-up turns for maximum detail.

Pros & Cons

Pros

  • โœ“ Produces well-structured, comprehensive content
  • โœ“ Prevents the model from losing track of topics
  • โœ“ Enables parallel expansion of sections
  • โœ“ Great for long-form content

Cons

  • โœ— Two-step process takes more coordination
  • โœ— Skeleton may need revision before expansion
  • โœ— Can produce somewhat formulaic structure
  • โœ— Adds overhead for short content

Example Prompts

First, generate a skeleton outline for a comprehensive guide on 'Setting up a CI/CD pipeline for a Node.js application.' Include 6-8 main sections with one-line descriptions. Then expand each section with detailed content, code examples, and best practices.

Create a skeleton for a blog post about 'Why startups should adopt AI early.' Outline: 1. [section title]: [one line description] 2. ... Now expand each section into 2-3 paragraphs with examples and data.

Skeleton first approach: Topic: Complete guide to database indexing Phase 1: Generate the skeleton (main sections + key points) Phase 2: I'll ask you to expand one section at a time for maximum detail.

FAQ

What is Skeleton of Thought?
Skeleton of Thought asks the AI to first generate a high-level outline (skeleton) of the answer, then expand each section in parallel or sequentially. This produces better-structured, more comprehensive responses and can dramatically speed up generation.
When should I use Skeleton of Thought?
Use Skeleton of Thought for long-form content, documentation, articles, reports, and any structured output. Especially useful when you need comprehensive coverage and don't want the model to forget important sections midway through.
How does Skeleton of Thought work?
Two phases: (1) Generate a concise outline with section headers and one-line descriptions, (2) Expand each section into full content. Since sections are independent, they can theoretically be generated in parallel for faster results.
Does Skeleton of Thought work with ChatGPT?
GPT-4 handles Skeleton of Thought well. Generate the outline first, then expand in follow-up messages for best results. Can also do both in one prompt.
Does Skeleton of Thought work with Claude?
Claude excels at outline-first approaches. Its long context window allows generating both skeleton and full expansion in a single response without losing coherence.