Prompt Chaining

Prompt chaining lets you break complex tasks into sequential steps, where each step builds on the results of previous steps. This enables sophisticated, multi-step workflows that can accomplish complex goals.

What is Prompt Chaining?

Prompt chaining connects multiple prompts together using the ~ (tilde) separator. Each section separated by ~ is executed as a separate step, with results from earlier steps available to later steps.

Simple Example:

Research [Topic]~
Summarize the findings~
Create a Google Doc with the summary

This workflow:

First researches the topic
Then summarizes the findings (using results from step 1)
Finally creates a Google Doc (using results from step 2)

Basic Syntax

Separating Steps

Use ~ (tilde) to separate steps:

Step 1 description~
Step 2 description~
Step 3 description

Step Structure

Each step should be:

Complete: A full instruction, not a fragment
Action-oriented: Tell the agent what to do
Specific: Clear about expected output

Good:

Research the top 5 competitors in [Industry]~
Analyze their pricing strategies~
Create a comparison document

Avoid:

Research~
Analyze~
Create

How Prompt Chaining Works

Sequential Execution

Steps execute one after another:

Step 1 runs and produces results
Step 2 receives Step 1's results as context
Step 3 receives both Step 1 and Step 2's results
And so on...

Context Passing

Each step has access to:

Results from all previous steps
Variables filled in at the start
Tool outputs from previous steps

Example Flow

Research [Topic] and find the top 5 key points~
Create a summary document with those key points~
Format the summary for [Format]~
Share the formatted summary via Slack

Execution:

Step 1 researches and finds 5 key points
Step 2 creates summary using those 5 points
Step 3 formats using the summary from step 2
Step 4 shares the formatted content

Building Effective Chains

Step 1: Break Down the Task

Identify logical steps:

Task: "Create a competitor analysis report"

Steps:

Research competitors
Analyze their strategies
Compare features
Create report document
Share report

Step 2: Write Each Step

Make each step clear and actionable:

Research the top 5 competitors in [Industry]~
Analyze each competitor's pricing, features, and market position~
Create a comparison table of key features~
Write a comprehensive analysis report~
Create a Google Doc with the report

Step 3: Add Variables

Include variables where needed:

[Industry]=[Industry to analyze]
[Format]=[Report format: detailed / summary]

Research the top 5 competitors in [Industry]~
Analyze their strategies~
Create a [Format] report~
Format for [Format]~
Share the report

Step 4: Tell Agent to Use Tools

Simply describe what tools to use in natural language:

Research [Topic] using web search~
Summarize findings~
Create a Google Doc with the summary~
Share the document via Slack to #research

Your agent will automatically use Perplexity, Google Docs, and Slack if those tools are enabled.

Best Practices

1. Limit Chain Length

Recommended: 3-5 steps maximum
Too many steps can be hard to follow
Break very long chains into separate workflows

2. Make Steps Independent

Each step should be understandable on its own:

Good:

Research [Topic]~
Summarize the research findings~
Create a document with the summary

Avoid:

Research [Topic]~
Summarize it~
Create it

("it" is unclear without context)

3. Build Progressively

Each step should build on previous steps:

Step 1: Gather data
Step 2: Analyze the data from step 1
Step 3: Create output using analysis from step 2

4. Use Clear Instructions

Be specific about what each step should do:

Good:

Research [Topic] and identify the top 5 key points~
Create a summary document with those 5 key points~
Format the summary as a blog post

Avoid:

Research [Topic]~
Do something with it~
Make it better

Advanced Techniques

Conditional Logic

While templates don't have explicit conditionals, you can structure steps to handle different scenarios:

Research [Topic]~
If the research finds [Condition], then [Action A], otherwise [Action B]~
Proceed with the appropriate action

Data Transformation

Transform data between steps:

Step 1: Gather raw data~
Step 2: Clean and structure the data~
Step 3: Analyze the structured data~
Step 4: Visualize the analysis

Iterative Processing

Process multiple items:

For each item in [List]:
  Research the item~
  Summarize findings~
  Add to master document

Examples

Content Creation Chain

[Topic]=[Topic to write about]
[Audience]=[Target audience]
[Format]=[Blog post / Email / Social media]

Research [Topic] for [Audience]~
Create an outline in [Format] format~
Write the full content based on the outline~
Review and refine the content~
Format the content for [Format]

Research & Analysis Chain

[Research Topic]=[What to research]

Search for information about [Research Topic] using web search~
Summarize the top 5 findings~
Identify key insights and trends~
Create a comprehensive analysis~
Create a Google Doc with the analysis~
Share the document via Slack to #research

Automation Chain

[Customer Name]=[Customer name]
[Order Number]=[Order ID]

Retrieve order [Order Number] for [Customer Name]~
Generate a status update email~
Format the email professionally~
Send email to [Customer Name] via Gmail~
Log the interaction in HubSpot CRM

Troubleshooting

Steps Not Executing

Check ~ placement: Should be between steps, not in the middle
Verify step syntax: Each step should be complete
Review error messages in activity drawer

Context Not Passing

Ensure steps are properly separated with ~
Check that previous steps completed successfully
Review step outputs in activity drawer

Chain Too Long

Break into smaller chains
Create separate workflows for different phases
Use background tasks for long chains

Next Steps

Templates Overview - Learn more about template syntax
Variables - Use variables in your workflows
Available Tools - See all available integrations

Prompt Chaining

On this page