Recursive summarization

A summarization pattern that compresses a document or conversation in multiple passes, summarizing summaries until the result fits a target token budget.

What is Recursive summarization?

Recursive summarization is a multi-pass summarization pattern that compresses a document or conversation until it fits a target token budget. In practice, the model summarizes chunks first, then summarizes those summaries again, which is a common approach in long-document workflows and OpenAI’s summarization examples. (cookbook.openai.com)

Understanding Recursive summarization

The basic idea is simple: if one pass cannot fit the full source material into the available context window, break the input into smaller pieces and summarize each piece. Those intermediate summaries become the new source material for the next pass, letting you steadily reduce length while preserving the most important points. OpenAI’s long-document cookbook explicitly shows this “previous summaries + next chunk” pattern, and its book-summarization work describes recursive task decomposition as a way to handle much longer inputs. (cookbook.openai.com)

This pattern is especially useful when you need a compact representation of a very large source, such as meeting transcripts, research reports, or support conversations. It is related to hierarchical summarization, where higher-level summaries are built from lower-level summaries. The tradeoff is that each pass can remove nuance, so the workflow works best when you define the summary goal clearly up front and keep the output structure consistent across passes. (red.anthropic.com)

Key aspects of Recursive summarization include:

  1. Chunking: split long input into manageable sections before summarizing.
  2. Intermediate summaries: use the output of one pass as the input to the next.
  3. Token budgeting: stop when the result fits the target context window or output limit.
  4. Structure preservation: keep headings, bullets, or schema stable across passes so meaning survives compression.
  5. Goal alignment: tune the prompt for the exact summary you need, such as executive notes, issue extraction, or action items.

Advantages of Recursive summarization

  1. Handles long inputs: it makes very large documents workable in models with finite context windows.
  2. Improves scalability: you can summarize books, transcripts, and multi-hour conversations without manual trimming.
  3. Supports reuse: intermediate summaries can be stored, searched, or reused in later workflows.
  4. Fits automation: it is easy to implement in pipelines that process content in batches.
  5. Creates tiered views: teams can keep both detailed and condensed versions of the same source.

Challenges in Recursive summarization

  1. Information loss: each pass can drop details that matter later.
  2. Summary drift: repeated compression can slowly shift emphasis away from the original source.
  3. Evaluation difficulty: it can be hard to tell whether a final summary is faithful without a reference standard.
  4. Prompt sensitivity: small changes in the instruction can produce very different results across passes.
  5. Format consistency: if intermediate outputs vary too much, the next pass becomes less reliable.

Example of Recursive summarization in action

Scenario: a product team has a 90-minute customer interview transcript and wants a one-page summary for leadership.

First, the transcript is split into sections by topic. Each section is summarized into a short note with the same fields, such as pain points, feature requests, and quotes. Those section notes are then combined into a second-pass summary that removes repetition and surfaces the strongest themes.

The final output is short enough to paste into a weekly update, but it still reflects the main user concerns from the original conversation. In a PromptLayer workflow, you can version the prompt, compare passes, and inspect where the summary changed as it moved from raw text to condensed output.

How PromptLayer helps with Recursive summarization

PromptLayer helps teams manage recursive summarization prompts, track how each pass changes the output, and compare summary quality across different chunking strategies or model settings. That makes it easier to tune for accuracy, brevity, and consistency as your documents get longer.

Ready to try it yourself? Sign up for PromptLayer and start managing your prompts in minutes.

Related Terms

Socials
PromptLayer
Company
All services online
Location IconPromptLayer is located in the heart of New York City
PromptLayer © 2026