Step-by-step prompting
Explicitly asking the model to work through a task step-by-step, slowing down reasoning and reducing mistakes.
Think of it like
A parent asking a child to explain their math work step-by-step, not just give the answer.
Example
"Solve step-by-step: 5×(3+2) = ?" Model: "Step 1: Solve inside parens: 3+2=5. Step 2: Multiply: 5×5=25."
How it actually works
Step-by-step reasoning (sometimes called "scratchpad") forces intermediate outputs, making reasoning visible and checkable. It works for logic puzzles, math, and fact-based tasks. Less effective for open-ended generation. Chain-of-thought (CoT) prompting is a formalization of this.
For product teams
Boosts accuracy on reasoning tasks with minimal prompt changes.
For engineers
Add "Let's think step-by-step" or similar phrasing. Parse intermediate outputs if you need to validate.
Related
- Chain of thought — Related to chain-of-thought and decomposition.
- Decomposition prompting — Decomposes tasks into substeps.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome