ToolsPromptsSkillsWorkflowsBlogAbout

Chain-of-Density — progressively denser, entity-rich summarization

Generate summaries that iteratively incorporate more entities and details while staying concise, producing information-dense summaries without fluff.

Prompting4.2
Chain-of-Density — progressively denser, entity-rich summarization preview

Prompt template

You will generate increasingly entity-dense summaries.

Article: {article}

Step 1. Identify 1-3 important entities from the article.
Step 2. Write a summary (under 120 words) that includes those entities.
Step 3. Identify 1-3 more entities missing from the previous summary.
Step 4. Rewrite the summary from scratch to include all previous entities plus the new ones, staying under 120 words.
Step 5. Repeat steps 3-4 for a total of 4 iterations.

Final summary (iteration 4):

How to use

  1. 1Provide the source article and set a target word count (typically 80-150 words)
  2. 2First pass: identify 1-3 core entities and write an initial summary covering them
  3. 3Second pass: identify 1-3 additional entities missing from the first summary
  4. 4Rewrite the summary from scratch to include all entities so far, staying within the word limit
  5. 5Repeat the identify-missing-then-rewrite cycle for 3-5 total iterations
  6. 6The final iteration produces the most information-dense summary — use that as the output

Pro tips

  • Chain-of-Density consistently outperforms standard summarization on information coverage metrics
  • Always rewrite from scratch in each iteration rather than appending — this forces better compression
  • 4 iterations is the sweet spot; beyond that summaries become cramped and less readable
  • Works best for news, research papers, and meeting notes where entity coverage matters

Works best with

ChatGPTClaudePerplexity

Ratings & Reviews

Loading reviews…

Share your experience

Select a rating

Related Skills

Persona Prompting — get consistent, expert-toned answers
Prompting

Persona Prompting — get consistent, expert-toned answers

Give the AI a role, audience and constraints so every answer comes back in a consistent, professional voice.

PromptingWritingchatgpt
4.7
Few-Shot Prompting — teach by example
Prompting

Few-Shot Prompting — teach by example

Give the AI 2-3 input/output examples before your real request to lock in format and style.

PromptingFormat ControlAutomation
4.5
RAG Prompting — retrieval-augmented generation for grounded answers
Prompting

RAG Prompting — retrieval-augmented generation for grounded answers

Combine retrieved context with targeted prompts so the AI answers from your documents instead of hallucinating.

RAGRetrievalGrounding
4.5