Lesson 006Field guide · Prompting 101
Prompt Chains
Link prompts together so one output feeds the next step.
You'll learn
- Why cramming five asks into one prompt weakens all five
- How a chain works: one prompt's output becomes the next prompt's input
- When to chain and when a single prompt is enough
- How review points between steps stop errors from compounding
- How to design your own three-step chain
01Section
The five-ask prompt
Way back in this path's first lesson, one warning read: "five asks in one prompt — split big requests into steps instead." This capstone lesson is that split, done properly.
Here's what the five-ask prompt looks like in the wild:
Five jobs, one prompt
Come up with topic ideas for my newsletter, pick the best one, write an outline for it, draft the introduction, and suggest some subject lines.
The AI will do all five — that's the trap. Each job gets a shallow pass, the AI makes the one decision that mattered most (picking the topic) without you, and by the end it's drafting an introduction for an idea you never approved. The fix isn't a longer prompt. It's a sequence.
Plain-language definition
Prompt chain — a series of prompts where each step's output becomes the next step's input — with you reviewing and steering between the steps.
Checkpoint
02Section
One output feeds the next
Here's that same newsletter job as a three-step chain. Notice what happens between the prompts — that's where you live:
Diverge: "Give me 10 newsletter topic ideas for [my audience], one sentence each." Then you choose — the decision that matters most stays yours, not the AI's.
Develop: "Take this topic: [your pick, pasted in]. Build an outline with 4 sections, one line on what each covers." You review the outline and cut what doesn't belong.
Produce: "Draft the introduction from this outline: [the fixed outline, pasted in]. Under 100 words, plain language." Now the draft is built on decisions you made.
The paste is the link
You carry work forward by pasting the reviewed version — your pick, your trimmed outline — into the next prompt. In a single conversation the AI does remember earlier messages, but pasting the approved version forward means each step builds on what you chose, not on the AI's first draft.
Checkpoint
03Section
Chain or single prompt?
Chains cost more of your attention than single prompts, so they have to earn it. The signals are reliable:
Chain: a choice sits mid-way
Brainstorm, then develop — someone has to pick, and it should be you.
Chain: different kinds of thinking
Generating ideas and polishing prose are different jobs; separated, each gets full attention.
Chain: raw material to refine
Messy notes → summary → action list → email. Each stage feeds the next.
Single prompt: one job, one shape
Rewrite a paragraph, draft one email, make a list — if you can judge the result at a glance, don't chain it.
Key takeaway
If you'd want to check the work halfway through, that halfway point is a chain link.
Checkpoint
04Section
The review between links
Chains multiply quality in both directions. Fix step 1's output and every later step inherits the fix. Wave a mistake through and every later step builds on it — by step 3 the error is woven in so neatly it looks intentional.
A review point doesn't need to be a deep audit. Thirty seconds and three questions: Is anything here wrong? Is this what I would have chosen? What should I cut before it feeds forward?
An error you don't catch at step 1 doesn't stay step-1-sized. Step 2 builds on it, step 3 polishes it, and the final output presents it with full confidence. The 30-second review between links is the cheapest quality control you'll ever do.
Privacy note: Every link is a paste — so the usual rule applies at every step, not just the first. Check what you're carrying forward: summaries and drafts can pick up names and details that shouldn't travel further.
Checkpoint
05Section
Design your own chain
Time to build one. The capstone exercise below gives you a fill-in plan for a three-step chain: the goal, a prompt for each step, and a review point after each. Most useful chains follow the same rhythm you've now seen — gather or generate, then shape, then produce.
A chain plan you keep is more than a prompt — it's a small system: a repeatable recipe for a whole task. That's the door into the next path, which turns recipes like this into full workflows.
Can't the AI just run all three steps itself?
Increasingly, yes — some tools can execute multi-step work on their own. But the review skill you're building here is exactly what makes that safe to use later: if you can't judge each step's output by hand, you can't judge it when it's automated either. Learn chains by being the link first.
Checkpoint
Prompt exercise
Plan your own three-step chain
This is the path capstone: copy the plan below, replace each [BRACKET] for a real recurring task of yours, then run it step by step in ChatGPT, Claude, Gemini, Copilot, or whichever AI tool you have access to — the website doesn't run AI itself. Pause at every review point before pasting forward. Keep the filled-in plan with your templates: it's a reusable recipe you can rerun whenever the task comes back.
MY THREE-STEP CHAIN: [TASK NAME] Goal: [what the finished result is, and who it's for] Step 1 — [job of this step, e.g. generate options / summarize raw material] Prompt: [your step 1 prompt] Review point: [what I check or choose before moving on] Step 2 — [job of this step, e.g. develop my pick / structure the material] Prompt: Here is the reviewed result of step 1: [PASTE]. [your step 2 prompt] Review point: [what I fix or cut before moving on] Step 3 — [job of this step, e.g. produce the final draft] Prompt: Here is the reviewed result of step 2: [PASTE]. [your step 3 prompt] Final review: [what I verify before actually using the result]
Reflection: After running your chain: which review point earned its keep — and honestly, would you have caught that problem inside one five-ask mega-prompt?
Quick check
4 quick questions — no pressure
There's no pass or fail here. Answer them all, and we'll show you the answers either way.