Prompt Chaining: Big Tasks, Small Steps
Break complex work into a sequence of focused prompts where each output feeds the next — and get far better results than one giant request.
Why One Giant Prompt Falls Short
Ask an AI to “research, outline, write, and polish a 10-slide presentation on our customer retention problem” in one prompt, and you’ll get something — but it will be shallow everywhere. The AI has to make dozens of decisions at once, and you have no chance to steer any of them.
Prompt chaining is the fix: split a big task into a series of smaller prompts, where the output of one step becomes the input to the next, and you review each link before moving on.
It’s how a skilled professional works anyway: nobody writes a report by typing the final version from top to bottom. Chaining just makes those stages explicit.
Three Chaining Patterns
Pattern 1: Sequential Build
Each step builds directly on the last. Great for documents, presentations, and plans.
- “Here are my notes on our customer retention data: [notes]. What are the 3–5 main themes?”
- “Using those themes, propose an outline for a 10-slide presentation to our leadership team. One line per slide.”
- (After you adjust the outline) “Write speaker notes for slides 1–3 using the revised outline.”
- “Review all the speaker notes for consistency and cut any repetition.”
Pattern 2: Generate, Then Filter
Produce many options, then evaluate them in a separate step. You used a version of this in the Module 2 brainstorm.
- “Generate 15 possible names for our internal mentoring program.”
- “Rate each name from 1–5 on: easy to remember, inclusive, not already a common brand. Show a table.”
- “Take the top 3 and write a one-sentence tagline for each.”
Pattern 3: Draft, Critique, Revise
Use separate steps for creating and judging, because AI (like people) evaluates better when it isn’t also trying to create.
- “Draft a response to this customer complaint: [complaint].”
- “Now act as the customer who wrote the complaint. How would you feel reading that response? What’s missing?”
- “Revise the response to address those gaps.”
Tips for Better Chains
- Plan the chain first. You can even ask the AI: “I need to produce X. Break this into 4–6 steps I should do one at a time. Don’t start the work yet.”
- Review at every link. A mistake in step 1 carries through every later step. Catching it early is the whole point.
- Edit the output before passing it on. You don’t have to accept an outline as-is — change it, then say “Use this revised version.”
- Keep one chain in one chat so the AI can see the earlier steps. For very long chains, see the next lesson on carrying context.
Choose a deliverable you actually need, or use this scenario: “Create a one-page onboarding guide for new volunteers at a community food bank.”
- Plan:
I need to create [deliverable] for [audience]. Break this into 4-5 steps I should complete one at a time. Don't do the work yet. - Step through it: complete each step in the same chat. After each response, review it and make at least one change or correction before saying “Next step.”
- Critique: at the end, ask the AI to review the whole thing from the audience’s point of view and list what’s confusing or missing.
- Revise and verify: apply the fixes you agree with, then check every fact yourself.
Reflect: at which step did your review make the biggest difference?
Common mistake: running a chain but typing “looks good, next” at every step without actually reading.
That turns a chain back into one giant prompt with extra clicks. The value of chaining comes from the checkpoints. If you’re not going to review a step, you’ve lost your chance to steer it — and any error compounds through everything that follows.
1. What is the defining feature of prompt chaining?
Chaining is about sequence and checkpoints: focused steps, each building on reviewed output from the previous one.
2. Why separate 'draft' and 'critique' into different steps?
A dedicated critique step — especially from a specific point of view — surfaces problems that are easy to miss during drafting, and lets you choose which changes to make.
Up Next
Long chains and long conversations raise a practical problem: what happens when a chat gets too long, or you need to start fresh without losing everything? The next lesson covers carrying context between chats — and using AI to generate images.