Lesson 8. Prompt chains and multi-step tasks
If you're looking for what a prompt chain is and how to build a multi-step workflow, this lesson shows why one big prompt often loses to a chain of research, outline, draft, and review.
Topic breakdown
A prompt chain is several small prompts connected sequentially. Each step solves one clear subtask, and its output becomes the input for the next stage.
For example, when preparing a landing page, article, or presentation, it's convenient to first gather insights, then build an outline, then write a draft, and only then do a final review.
This approach is especially valuable because errors can be spotted earlier. If research is weak, you fix it before moving on to writing the text.
What you'll learn
- break large tasks into stages
- pass intermediate output to the next prompt
- make review a mandatory part of the chain
- set a separate goal and format for each stage
Lesson plan
When you need a prompt chain
When the task requires multiple thinking stages: research, structure, writing, and review.
How to properly split into steps
Each stage should answer one clear question: what to find, how to build structure, what to write, and how to verify.
How to clean intermediate results
Pass forward only the useful results of the stage, not the entire raw response. This keeps the chain manageable.
Where a human is needed
At strategic points, when working with facts and important decisions, it's useful to insert manual review between stages.
Weak vs strong prompt
Write me a landing page for a new service.
Step 1: find 5 customer pain points for the new service. Step 2: from those insights, build a landing page outline of 6 blocks. Step 3: write the hero and benefits sections. Step 4: check the text for clarity and CTA strength. Present each stage separately.
The strong prompt doesn't jump straight to the final text. It first gathers material and structure, so the final result is deeper and more reliable.
Deep dive
Prompt chaining is a fundamental technique for anyone working with AI on complex tasks. Instead of one big request, you break work into several connected stages where the result of each becomes input for the next. This gives control at every step and lets you fix errors early rather than at the end.
Real business chain examples: creating a marketing campaign — audience analysis, idea generation, selecting the best one, and developing content for each channel. Preparing a commercial proposal — client research, proposal outline, writing sections, review. This exact structure is used in professional AI tools: Make.com, Zapier, Notion AI.
What distinguishes a good chain from a bad one? In a good chain: each step has a clear input and output; intermediate results are reviewed by a human; failure at one step doesn't break the whole chain. In a bad chain: steps are vaguely defined, intermediate context is passed entirely without cleanup, and there are no checkpoints.
Prompt chaining and agentic AI: if a human checks each step in the chain, it's prompt chaining. If the model automatically moves from step to step without human involvement, that's agentic AI. For beginners and high-error-risk tasks, start with manual chaining: it's safer and easier to debug.
Ready prompt template
Copy and adaptStep 1: find 5 key insights on the topic and present them as a list. Step 2: based on those insights, build an outline of 5 blocks. Step 3: from the outline, create a first draft. Step 4: check the draft for clarity, logic, and audience fit. Use the result of each step as input for the next.
Why it works
Splitting the task into stages doesn't overload the model and makes each step clear.
Intermediate output becomes cleaned context for the next prompt.
A separate format at each step helps keep the chain clear and manageable.
Review can be done not only at the end: for complex tasks, checkpoints in the middle of the process are useful too.
Practice
- Choose a topic: landing page, blog post, or presentation.
- Build a chain of 4 stages: research, outline, draft, review.
- For each stage, specify the required output format.
- Note where human review is needed.
Mini-project
Mini-project: 4-step AI workflow
Take a real task and build a prompt chain for it to get a repeatable process, not a one-time answer.
Tasks
- Choose a task: landing page, article, or presentation.
- Build 4 prompts for research, outline, draft, and review.
- Check the result after each step before passing it forward.
- Describe at which stage the chain's value was highest.
Deliverables
- 4 connected prompts
- result of each stage
- brief conclusion on the workflow
Checklist
Common mistakes
- skipping checks between stages
- adding a new goal at each step
- passing too-dirty intermediate output to the next prompt
- collapsing the whole task back into one step after splitting it
Lesson FAQ
Is a chain needed for every task?
No. For quick simple tasks a single prompt is sufficient. A chain is useful where the task is complex, multi-step, or strategically important.
How many steps should a chain have?
Start with the minimum. Often 3-4 stages are enough. Each additional step should genuinely add control or quality.
How does a prompt chain differ from agentic AI?
In a chain, the human checks each intermediate result before moving to the next step. In agentic AI, the model automatically transitions between steps. For most business tasks, a manual chain is more reliable and gives more control.