Prompt Chaining — Prompting Guide & Examples
Prompt chaining breaks a complex task into a sequence of simpler prompts, where each prompt's output becomes the next prompt's input. Like an assembly line, each step handles one specific sub-task, producing higher quality results than a single monolithic prompt.
How It Works
Design a pipeline: Prompt A produces output, which feeds into Prompt B, which feeds into Prompt C, etc. Each prompt is optimized for one specific task. You can add validation, branching, and error handling between steps.
When to Use
Use chaining for complex multi-step workflows: research→outline→draft→edit, data extraction→analysis→visualization, code generation→review→testing. Essential when a single prompt can't handle the full complexity.
Model-Specific Tips
ChatGPT / GPT-4
Use the Assistants API or LangChain for automated chaining. Multi-turn conversations work for manual chains. GPT-3.5 for simple steps, GPT-4 for complex ones.
Claude
Claude handles prompt chains well in multi-turn conversations. For programmatic chains, use the Messages API with sequential calls. Claude's consistency helps prevent error propagation.
Gemini
Gemini supports chaining via the API. Use Vertex AI pipelines for production workflows. Gemini's function calling can automate chain orchestration.
Pros & Cons
Pros
- ✓ Handles complex tasks that defeat single prompts
- ✓ Each step is simple and debuggable
- ✓ Can mix models — cheap for simple steps, powerful for hard ones
- ✓ Easy to add validation between steps
Cons
- ✗ Error propagation — mistakes compound
- ✗ Higher total latency (sequential steps)
- ✗ Requires orchestration logic
- ✗ More complex to build and maintain
Example Prompts
STEP 1: Extract all company names, revenue figures, and growth rates from this earnings report. STEP 2: [Feed step 1 output] Organize this data into a comparison table sorted by growth rate. STEP 3: [Feed step 2 output] Write a 200-word executive summary highlighting the top 3 performers and key trends.
Chain: (1) Generate 10 blog post title ideas about AI productivity → (2) Select the best 3 and explain why → (3) Create a detailed outline for the #1 pick → (4) Write the introduction paragraph
Pipeline: 1. 'Analyze this codebase and list all API endpoints with their methods and parameters' 2. 'Generate OpenAPI spec from this endpoint list' 3. 'Create example request/response pairs for each endpoint' 4. 'Write developer documentation from the spec and examples'