Prompt Chaining: Split Big AI Tasks Into Small Steps

When you ask an AI to do a big job in one go, such as "write a complete business proposal" or "analyse this data and make a report", the result is usually long, generic and hard to fix. Experienced users do something different: they break the job into a chain of small prompts and check each step before moving on.

Why one giant prompt struggles

The AI has to make many decisions at once, and you only see the final result.
If one early decision is wrong (the wrong audience, a misread number), everything after it is built on that mistake.
A long answer is tiring to check, so mistakes slip through.

Small steps fix all three problems: fewer decisions per step, mistakes caught early, and short answers that are easy to review.

The basic chain

Most work tasks follow the same four steps:

Understand: ask the AI to read the material and summarise what it sees.
Plan: ask for an outline or approach. Correct it.
Draft: ask for one section at a time, following the plan.
Polish: ask for a final review, formatting and a short summary.

Example: a monthly sales report

Step 1: Understand

Here is our sales data for October (CSV below). Before any analysis,
tell me: how many rows, which columns, the date range, and anything that
looks wrong or missing. Do not analyse yet.
"""
[paste data]
"""

Check the answer. If it says 230 rows and you know there are 240, stop and fix it now.

Step 2: Plan

Suggest an outline for a one-page report for our director. She cares
about total sales, top 5 products, which region fell behind, and one
recommendation. Outline only.

Edit the outline in your reply: "Drop section 4, add a comparison with September."

Step 3: Draft, one section at a time

Write section 1 (Total sales) using the data above. Show the calculation
so I can check it. Maximum 80 words.

Check the numbers in Excel before asking for section 2.

Step 4: Polish

Combine the approved sections into one report. Keep it under one page,
add a 3-line summary at the top, and do not change any numbers.

Keep the chain in one chat

Run the whole chain in one conversation so the AI remembers the data and your corrections. Start a new chat for the next job.

If a chat gets very long and the AI starts forgetting earlier details, ask it to summarise everything agreed so far, then paste that summary into a fresh chat and continue.

Chains you can reuse

Job	Chain
Proposal	Client needs → Outline → Section by section → Executive summary
Training material	Learning goals → Lesson outline → Content → Quiz questions
Difficult email	Facts and goal → Key points → Draft → Tone check
Data analysis	Check data → Questions to answer → Calculations → Findings

Practice task

Choose one bigger task from your work this month: a report, a proposal or a presentation. Do it as a four-step chain: understand, plan, draft section by section, polish. Note one mistake you caught early that a single giant prompt would have hidden.

Key takeaways

Big jobs work better as a chain of small prompts than one giant prompt.
The basic chain: understand, plan, draft section by section, polish.
Check each step before moving on, especially numbers.
Keep one chain in one chat, and summarise into a new chat if it gets too long.