Think step by step
Lesson 13 of 23 in the course “AI basics”. Free, with animations and a quiz.
Jump to the answer or take steps
The model writes its answer token by token, meaning piece by piece. If you demand the result right away, it has no room to work it out. Example: a $24 pizza for four, plus 3 colas at $2 each, and only three people had cola. The quick answer “$7.50 each” is wrong: the one without cola overpays. Step by step: pizza $24 / 4 = $6, cola +$2 for three people. Result: three pay $8, one pays $6. Check: 8 × 3 + 6 = $30, same as the bill.
The phrase “step by step”
Add to your prompt: “Solve it step by step, show your reasoning, then give the final answer.” Why it works: each written step becomes part of the text, and the next step builds on it. This helps with tasks that take several moves: math, logic, plans, comparing options. For a simple question like “capital of France” you don't need it, the answer just gets longer.
Models that think on their own
There are reasoning models. Before answering, they write a draft of their reasoning on their own, often hidden or collapsed under a label like “thinking”. They need “step by step” less, since they already do it. But they answer more slowly, so for quick everyday questions the regular mode is enough. What helps them more: a clear task, all the conditions, and how to tell a good answer.
Cut a big task into pieces
One huge request like “prepare me for my interview” gives you mush. Better: a prompt chain, meaning several requests in a row where the output of one feeds the next. Example for a barista job: 1. List 10 common interview questions. 2. Help me improve my answers to them. 3. Run a mock interview with me. Check the result after every step. That way you see right away where things went off.
Ask it to check itself
After the answer, write: “Check your answer: recalculate the numbers and look for mistakes.” The model often catches its own slips, especially in calculations. But there's a limit. Steps can look logical and still be wrong. And if you ask “are you sure?”, the model may confidently confirm its own mistake. In short: steps help, but recalculate important numbers yourself.
Check yourself
- Why does “solve it step by step” help on multi-step tasks?
Correct answer: Each step is written down and the next builds on it. The model writes piece by piece. Written steps become part of the text, and the result builds on them instead of being a guess. - You're preparing for an interview with AI's help. What works best?
Correct answer: A prompt chain, checking each step. Small steps are easier to check. If something goes wrong, you see right away where. - The model showed its steps and checked itself. What now?
Correct answer: Still recalculate the important numbers yourself. Steps can look logical and still be wrong, and the model can confirm its own mistake. Check what matters yourself.
← Show an example: few-shot Get the answer in the right format →