What is prompt chaining and when should I use it?

Prompt chaining breaks a complex task into sequential LLM calls, where each step's output feeds into the next. Use it when a single prompt cannot reliably handle the full task — for example, research → outline → draft → edit for content creation.

How do I pass context between chained prompts?

Include the previous step's output directly in the next prompt, often with a clear label like 'Based on the following research: [output]'. Keep only relevant parts to avoid hitting context limits. Summarize long outputs before passing them forward.

What are the disadvantages of prompt chaining?

Main downsides: higher latency (multiple API calls), higher cost (more tokens), error propagation (mistakes in early steps compound), and complexity in error handling. Use single prompts when they produce acceptable results.

How many steps should a prompt chain have?

Most effective chains have 2-5 steps. Each step should have a clear, single purpose. If a chain exceeds 5 steps, consider whether some steps can be combined or if you need a different architecture like an agent loop.

TechniqueNew

Prompt Chaining

Sequential task decomposition

The Problem: Some tasks are too complex for a single prompt. How can we break them into stages and pass results between steps?

The Solution: Build a Production Line

Prompt Chaining connects multiple prompts where the output of one becomes the input for the next. It's like an assembly line in a factory — each station does one job well, and the product moves through stages. Using structured output between steps keeps data clean, and function calling can automate the handoffs.

Think of it like a factory assembly line:

1. Station 1: Extract key entities from the document
2. Station 2: Analyze sentiment for each entity
3. Station 3: Generate summary with sentiment context
4. Final product: Rich, structured analysis

Where Is This Used?

Content Pipelines: Research → Draft → Edit → Format
Data Processing: Extract → Transform → Validate → Load
Analysis Workflows: Classify → Summarize → Recommend
Code Generation: Plan → Implement → Review → Test

Fun Fact: Prompt chaining lets you use different models or temperatures for different stages! You might use a creative model for brainstorming and a precise model for fact-checking in the same pipeline.

Try It Yourself!

Use the interactive example below to see how chaining multiple prompts creates more sophisticated outputs than a single prompt could achieve.

Prompt Chaining

Break complex tasks into a sequence of simpler steps

Breaking down content creation into research → outline → draft → edit

Research

○ Pending

Outline

○ Pending

Draft

○ Pending

Edit

○ Pending

Benefits of Prompt Chaining

• Better quality control at each step
• Easier to debug and improve
• Ability to use different models per step
• Overcome context limitations

Code Example

// Prompt Chaining Example
async function createBlogPost(topic: string) {
  // Step 1: Research
  const research = await llm.complete({
    prompt: `List 5 key facts about ${topic}`
  });

  // Step 2: Create outline using research
  const outline = await llm.complete({
    prompt: `Create a blog outline using: ${research}`
  });

  // Step 3: Write draft based on outline
  const draft = await llm.complete({
    prompt: `Write introduction based on: ${outline}`
  });

  // Step 4: Edit and polish
  const final = await llm.complete({
    prompt: `Improve for clarity: ${draft}`
  });

  return final;
}

Use Cases

• Content creation: research → outline → draft → edit
• Code review: security → performance → style → report
• Data analysis: collection → cleaning → analysis → visualization
• Documentation: structure → sections → examples → final version

Try it yourself

Interactive demo of this technique

Technique Comparison

Demo Mode

Pre-recorded responses

TaskBeginnerWriting

Write a blog post about the benefits of morning walks

Without technique

Prompt

Write a blog post about the benefits of morning walks.

Response

Morning walks are good for health. They help you wake up, improve mood, and boost immunity. Walking in the morning is good for the heart. Fresh air fills the body with oxygen. Walks also help with weight control. Try starting with 15 minutes a day.

Tokens:18/68

Time:420ms

Quality:

With Prompt Chaining

With technique

Prompt

--- Prompt chain --- Step 1 (outline): Create an outline for a blog post about morning walk benefits. 5 sections with key points. → Step 1 result: 1. Intro: why we ignore the simplest habit 2. Science: what happens to your body in the first 30 minutes 3. Mental health: walking as an antidepressant 4. Practical tips: how to build the habit 5. Conclusion: a 7-day challenge for the reader Step 2 (draft): Write the post following this outline. Use facts, examples, engaging language. → Step 2 result: [800-word draft] Step 3 (polish): Edit the post — remove filler, add a compelling intro and a strong ending.

Response

The Simplest Life Hack You Keep Ignoring

Every morning you have a free tool that reduces stress by 25%, improves sleep, and burns 150 calories. You keep ignoring it — it is a simple walk.

What Happens in 30 Minutes

Within 5 minutes of walking, cortisol levels start dropping. By minute 15, your brain releases endorphins. By minute 30, blood flow to the brain increases by 15% — a Stanford study showed this directly boosts creativity.

Walking as an Antidepressant

A 2023 meta-analysis (British Journal of Sports Medicine) confirmed that 30 minutes of walking reduces anxiety symptoms on par with medication. Morning sunlight calibrates circadian rhythms, improving sleep quality by 42%.

How to Start (and Not Quit)

Place your sneakers by the door the night before
Start with 10 minutes, add 5 minutes each week
Anchor to an existing habit: coffee → walk

Your Challenge: 7 Days

Tomorrow morning — 10 minutes. No excuses. In a week, your body will thank you.

👁️Step 1: create structure — article skeleton with key points

🧠Step 2: write to the plan — each section develops its thesis

🧠Step 3: edit — remove filler, strengthen opening and closing

✅Each step builds on the output of the previous one

Tokens:165/310

Time:1550ms

Quality:

Why this works

The "outline → draft → edit" chain produces results impossible to achieve with a single prompt. Each step focuses on one task, passing the result forward — like an assembly line in a factory.

1 / 2

Practice Challenges

Create a free account to solve challenges

3 AI-verified challenges for this lesson

Related lessons:Least To Most Chain Of Thought

This lesson is part of a structured LLM course.

My Learning Path

TechniqueNew

Prompt Chaining

Sequential task decomposition

The Problem: Some tasks are too complex for a single prompt. How can we break them into stages and pass results between steps?

The Solution: Build a Production Line

Think of it like a factory assembly line:

1. Station 1: Extract key entities from the document
2. Station 2: Analyze sentiment for each entity
3. Station 3: Generate summary with sentiment context
4. Final product: Rich, structured analysis

Where Is This Used?

Content Pipelines: Research → Draft → Edit → Format
Data Processing: Extract → Transform → Validate → Load
Analysis Workflows: Classify → Summarize → Recommend
Code Generation: Plan → Implement → Review → Test

Try It Yourself!

Use the interactive example below to see how chaining multiple prompts creates more sophisticated outputs than a single prompt could achieve.

Prompt Chaining

Break complex tasks into a sequence of simpler steps

Breaking down content creation into research → outline → draft → edit

Research

○ Pending

Outline

○ Pending

Draft

○ Pending

Edit

○ Pending

Benefits of Prompt Chaining

• Better quality control at each step
• Easier to debug and improve
• Ability to use different models per step
• Overcome context limitations

Code Example

// Prompt Chaining Example
async function createBlogPost(topic: string) {
  // Step 1: Research
  const research = await llm.complete({
    prompt: `List 5 key facts about ${topic}`
  });

  // Step 2: Create outline using research
  const outline = await llm.complete({
    prompt: `Create a blog outline using: ${research}`
  });

  // Step 3: Write draft based on outline
  const draft = await llm.complete({
    prompt: `Write introduction based on: ${outline}`
  });

  // Step 4: Edit and polish
  const final = await llm.complete({
    prompt: `Improve for clarity: ${draft}`
  });

  return final;
}

Use Cases

• Content creation: research → outline → draft → edit
• Code review: security → performance → style → report
• Data analysis: collection → cleaning → analysis → visualization
• Documentation: structure → sections → examples → final version

Try it yourself

Interactive demo of this technique

Technique Comparison

Demo Mode

Pre-recorded responses

TaskBeginnerWriting

Write a blog post about the benefits of morning walks

Without technique

Prompt

Write a blog post about the benefits of morning walks.

Response

Tokens:18/68

Time:420ms

Quality:

With Prompt Chaining

With technique

Prompt

Response

The Simplest Life Hack You Keep Ignoring

Every morning you have a free tool that reduces stress by 25%, improves sleep, and burns 150 calories. You keep ignoring it — it is a simple walk.

What Happens in 30 Minutes

Walking as an Antidepressant

How to Start (and Not Quit)

Place your sneakers by the door the night before
Start with 10 minutes, add 5 minutes each week
Anchor to an existing habit: coffee → walk

Your Challenge: 7 Days

Tomorrow morning — 10 minutes. No excuses. In a week, your body will thank you.

👁️Step 1: create structure — article skeleton with key points

🧠Step 2: write to the plan — each section develops its thesis

🧠Step 3: edit — remove filler, strengthen opening and closing

✅Each step builds on the output of the previous one

Tokens:165/310

Time:1550ms

Quality:

Why this works

The "outline → draft → edit" chain produces results impossible to achieve with a single prompt. Each step focuses on one task, passing the result forward — like an assembly line in a factory.

1 / 2

Practice Challenges

Create a free account to solve challenges

3 AI-verified challenges for this lesson

Related lessons:Least To Most Chain Of Thought

This lesson is part of a structured LLM course.

My Learning Path