Guide
Prompt chaining
Prompt chaining is a sequence of prompts where the output of one step becomes the input of the next. Each prompt has one job. The chain is how you get a draft, a critique, and a revision without stuffing every instruction into a single message.
Updated 2026-09-23
Before and after
Before
Research our refund policy, decide if this customer qualifies, and write the reply.
After
Step 1 — Extract: From the policy below, list the refund rules that mention timing or condition. Output a bullet list. Do not write the reply. Step 2 — Decide: Using only that list, say whether this customer qualifies and which rule applies. One paragraph. Step 3 — Write: Using only the decision, write the customer reply in under 80 words.
How to do it
1. Write the last artifact first
Decide what the final message must contain. Then walk backward to the smallest step that can feed it.
2. Give each step one verb
Step 1 extracts. Step 2 chooses. Step 3 writes. If a step has two verbs, split it.
3. Paste only what the next step needs
Do not paste the entire chat. Pass the outline, the selected facts, or the critique bullets.
4. Lock the format between steps
If step 2 expects a list of five claims, say that in step 1. A chain breaks when the handoff shape changes.
When to use this
Chain prompts when a single answer would have to research, decide, and polish at once. Typical chains are outline then draft, extract then classify, or draft then critique then revise. Do not chain when one clear prompt already produces the artifact. Extra steps add chances for the model to drift.
Chaining versus chain of thought
Chain of thought is one prompt that asks for steps and an answer together. Prompt chaining is several prompts. Use chain of thought when you want the reasoning in a single response. Use a chain when each stage should be checked by a person before the next stage runs.
In a chat window you can run the chain by hand: send step 1, read it, send step 2 with that text attached. In an agent or API workflow the same handoff is code. The prompting pattern is identical.
Sources
Questions
What is prompt chaining?
Prompt chaining is a series of prompts where each answer is passed into the next prompt. Each step does one job.
How is prompt chaining different from chain of thought?
Chain of thought is one prompt that asks for reasoning and a final answer. Prompt chaining is multiple prompts, with a handoff between them.
When is one prompt better than a chain?
When the task has a single deliverable and you can state the constraints up front. A chain is for jobs that need a checkpoint between stages.