Chain of Density Summarization

Expert

Iteratively pack more information into summaries without increasing length.

01 The Concept

Chain of Density (CoD) is an iterative prompting technique designed to write highly dense, informative summaries.

The prompt instructs the model to start with a sparse summary, identify 1-3 entity-dense details missing from the draft, and rewrite the summary to fuse these details into the same word limit, repeating this density cycle 5 times.

02 Weak vs. Strong

EX 01Chain of Density Summary Loop

You will generate a summary of the article. Instructions: - Start with a sparse, 80-word summary. - Perform 5 cycles of optimization. - In each cycle: identify 2 critical entities missing from the current draft, then rewrite the summary to fuse those entities while keeping it strictly under 150 words.

→ Why it works

Systematically forces information density. Instructing the model to iteratively find and merge key terms within a fixed word budget produces dense, high-yield summaries.

03 Key Points

01Entity Fusion: Merging missing details into the existing summary structure without increasing length.

02Density Cycles: Repeating the identification and fusion process over 5 cycles to maximize information.

03Word limit constraints: Enforcing a strict word budget (e.g. 150 words) to compel condensation.

04Sparse Start: Beginning with a generic outline to leave room for dense insertions.

05Refinement Strategy: Fusing information using compound sentences and removing filler adjectives.

04 Model-Specific Notes

Claude executes CoD summarize instructions with high semantic accuracy, adjusting sentence grammar dynamically to fit compound phrases.

05 For Your Role

It's like packing a suitcase: start with the main clothes, then slide socks and accessories into the small gaps.

Next: Prompt Versioning & Git Workflows →