Prompt Chaining
Also called Chained prompts, Prompt pipeline
Splitting a task into a sequence of smaller prompts, where each step's output becomes the next step's input.
What is prompt chaining?
Prompt chaining breaks one large job into several small, separate requests to an AI model, and passes the output of each step into the next. Rather than asking for "a researched blog post" in one go, you might ask for an outline, then a draft of each section, then an edit for tone, then a list of claims to fact-check.
How does it work?
- Each step has its own focused prompt with one clear job.
- The result is checked or reshaped, often in a fixed format using structured output, so code can pass it on reliably.
- The next prompt includes that result and does the following job.
Steps can also branch: a first prompt classifies a support email, and different follow-up prompts handle refunds, bugs or sales questions. Chains like this are the simplest kind of agentic workflow, and many automation tools let you build them without code.
What is it commonly confused with?
- Chain-of-thought. That's step-by-step reasoning inside a single answer. Prompt chaining uses several separate requests that you, or your code, control.
- An AI agent. In a chain, the steps are fixed in advance. An agent decides its own next step.
Why does it matter?
Small prompts are easier to write, test and fix than one giant prompt, and each step is usually more accurate when it has only one job. You can add checks between steps, use a cheaper model for the simple steps, and see exactly where something went wrong. The trade-off is more requests, which adds time and cost.
