Prompt Chaining
One prompt rarely gets you all the way there. Here's how to break a big task into a short sequence of small, sharp steps.
When you ask AI to do something complex in a single prompt—'research our competitors and write a strategy memo'—you usually get a shallow, generic result. Not because it can't help, but because you've asked for five things at once. The fix is chaining: do it in steps, where each builds on the last.
Why one prompt isn't enough
Think back to the hallway. A big, vague request sends the AI down a generic path because there's too much to cover and no focus. Breaking the work into stages keeps each step tight, lets you check the output before moving on, and steers the whole thing toward what you actually want.
The pattern
Most chains follow the same shape: gather, structure, draft, refine. You run one prompt, glance at the result, then feed it into the next.
'List the 5 things a strategy memo on X should cover.' Get the raw material first.
'Turn those into an outline with a one-line note under each heading.' Now you have a skeleton.
'Write the pricing section from that outline, about 150 words.' Tackle it piece by piece.
'Tighten this and flag anything that needs a real number.' Polish at the end.
A worked example
Say you're building a workshop. Prompt one: brainstorm 10 possible topics. Prompt two: pick the 5 strongest and arrange them into a half-day agenda. Prompt three: write the description for session two. Each step is small enough to get right, and you're in control the whole way.
When to chain vs. one-shot
For quick, single jobs—a short email, a definition, a tidy-up—one prompt is fine. Reach for a chain when the task has parts, when quality matters, or when the first attempt came back vague. If a request feels too big, that's your signal to break it up.
Gather, structure, draft, refine. Complex work goes better as a short chain of small steps than one giant ask.