What Is Prompt Chaining? Steps and Practical Examples

Prompt chaining means splitting a task into separate, connected AI requests. The output of one request becomes input for the next. You can check a result, supply missing information or decide whether to continue before starting another step.

For example, a business proposal can begin with extracting customer requirements, continue with an approved outline and end with a draft. This gives you specific material to review along the way. New to the terminology? Start with what an AI prompt means.


When is prompt chaining useful?

Create a separate step when its output supports a real check or decision. A list of requirements, an agreed outline and a review against written criteria are useful handoffs. Splitting a simple sentence translation into several requests will often add work without a clear benefit.

  • Documents: extract findings → check sources → prepare a summary.
  • Content: define the brief → agree an outline → draft → revise.
  • Customer messages: identify the issue → check available facts → propose a response.

Worked example: turn meeting notes into a proposal draft

The following sequence is an original practice example. Use approved material and leave missing prices or commercial terms for a person to complete.

1. Extract facts and unanswered questions

From these notes, create a table with customer requirement, supporting note and missing information. Do not add services, prices or deadlines. Turn uncertainty into questions to resolve. Notes: {text}.

Compare the table with the notes. Correct misunderstandings and add answers that you can confirm before moving on.

2. Build an outline from checked inputs

Use this checked requirements table: {table}. Propose a business proposal outline covering the goal, scope, deliverables, assumptions and open questions. Leave placeholders for any missing price or deadline. Do not write the full proposal yet.

3. Draft from the approved outline

Use approved outline {outline} and confirmed source material {sources} to draft the proposal. Keep the agreed scope and do not introduce unconfirmed commitments. Finish with a list of details a person must check before sending.

The result is a draft for review. Sending it to a customer is a separate action that this sequence does not perform.


How to pass information between steps

A loose summary is a poor handoff when the next step needs exact figures or quotations. Pass the relevant source, the checked intermediate result and the constraints that still apply. Name approved versions clearly, such as “proposal outline v2”.

Earlier messages may remain available in the same chat. In a new conversation or an automated workflow, supply the required material explicitly. Begin manually, then automate only after you can describe each step’s input, output and stop condition.

For a reusable brief with a goal, audience and output format, use our practical prompt-writing guide.


Prompt chaining vs Chain of Thought and few-shot prompting

Prompt chaining organises separate requests and their visible outputs. Chain of Thought concerns a reasoning approach; it is not another name for this workflow. Request useful results, supporting material and concise explanations that you can inspect.

Few-shot prompting supplies examples within a request. You can use examples in any step of a chain. Role prompting sets the perspective for handling the task.


Common failure points

  • An early factual error passes through every later step.
  • A summary drops an important constraint or source reference.
  • Extra steps produce more text without supporting a decision.
  • The AI reviews its own draft without checking the original evidence.

Ask four questions before using a sequence: What must this step return? How can it be checked? Who approves it? When should the workflow stop? More requests can add time and cost; they do not guarantee a correct answer.


Frequently asked questions

Do I need to write code?

No. You can run a short sequence manually in a chat and review the outputs yourself. Code or a workflow tool becomes relevant when you want the steps to run automatically.

How many prompts should a chain contain?

Use the smallest number that supports the work and the checks you need. Add a step for a clear purpose, not to make the process appear more advanced.


Try the method on your own task

Start with one task supported by reliable source material. Use our practical guide and AI prompt templates. To discuss a workflow for your team, tell us about your use case.


Related techniques and sources

Role prompting · Few-shot prompting · Chain of Thought

Further reading: Anthropic: Prompting best practices.

The examples in this article are original practice scenarios, not model performance measurements.

← Back to the AI Glossary