P

Chain of Abstraction: Definition and Examples

A prompting technique that breaks down complex reasoning into successive levels of abstraction, allowing the model to progress gradually from the general concept to specific details.

Full definition

Chain of Abstraction (CoA) is an advanced prompting method that structures a language model's reasoning into hierarchical abstraction layers. Instead of asking for a direct answer to a complex problem, the model is guided to process information starting from a macro view before gradually descending into concrete details.

This approach draws inspiration from software engineering principles and systems thinking, where concerns are separated into distinct levels. At each abstraction level, the model focuses on a specific aspect of the problem: first broad categories, then subcategories, and finally specific elements. This reduces the cognitive load imposed on the model at each step.

Concretely, Chain of Abstraction works by asking the model to first identify the abstract concepts or placeholders needed for resolution, then progressively fill them with concrete information. This separation between abstract reasoning and concrete instantiation allows for more structured and reliable answers, especially on tasks requiring factual knowledge or calls to external sources.

CoA differs from classic Chain of Thought in that it does not simply unfold a linear step-by-step reasoning: it organizes the reasoning vertically, into levels of increasing granularity. This technique is particularly effective for planning tasks, multi-criteria analyses, and structured content generation.

Etymology

The term "Chain of Abstraction" was formalized in AI research in 2024, notably in works by Meta AI. It combines "chain", by analogy with Chain of Thought, and "abstraction", a fundamental concept in computer science referring to the simplification of a complex system by hiding its implementation details.

Concrete examples

Complex project planning

Plan the launch of a mobile app. Level 1: identify the 4 main phases. Level 2: for each phase, list the key steps. Level 3: for each step, detail the concrete actions with responsible parties and deadlines.

Multi-layer technical problem analysis

My API is slow. Reason by abstraction levels: first, the broad categories of possible causes (network, database, code, infrastructure), then for each category the sub-causes, and finally the specific diagnostics to run.

Structured content writing

Write a guide on machine learning. Start by defining the 3 major paradigms (high abstraction), then break each down into main techniques (medium abstraction), and finally give a concrete implementation example for each technique (low abstraction).

Practical usage

To apply Chain of Abstraction, structure your prompts into explicit levels: first ask the model to identify high-level concepts or categories, then progressively detail each element. Use markers like "Level 1", "Level 2" or "Macro view / Detailed view" to guide the descent in abstraction. This technique is especially powerful for analysis, planning, and problem-solving tasks where a direct approach would yield incomplete or disorganized results.

Related concepts

Chain of ThoughtTree of ThoughtsTask DecompositionHierarchical Reasoning

FAQ

What is the difference between Chain of Abstraction and Chain of Thought?
Chain of Thought unfolds a linear reasoning step by step, while Chain of Abstraction organizes reasoning into hierarchical levels, from the most general to the most specific. CoT is horizontal (sequential), CoA is vertical (by layers of granularity).
When to use Chain of Abstraction instead of a simple prompt?
Chain of Abstraction is particularly useful for complex problems with multiple dimensions or nested sub-problems: project planning, multi-criteria analyses, generating long structured content, or system diagnostics. For simple, straightforward questions, a classic prompt remains more efficient.
Does Chain of Abstraction work with all language models?
The most advanced models (GPT-4, Claude, Gemini) benefit the most from this technique thanks to their structured reasoning ability. Smaller models can follow the structure if the abstraction levels are explicitly defined in the prompt, but the results will be less rich at each level.

See also

How to use this prompt

  1. Copy the prompt with the button above.
  2. Paste it into ChatGPT, Claude or your favorite AI assistant.
  3. Replace the bracketed variables with your details, then refine the result.

About Prompt Guide

Prompt Guide is a free library of 2500+ ready-to-use prompts for ChatGPT, Claude and other AIs, with guides to learn prompting and tools to build and optimize your own prompts.

More definitions

Get new prompts every week

Join our newsletter.