Planning and step-by-step reasoning explained simply

What is it?

For multi-step problems, models usually do better when they work through the problem in stages instead of jumping to the answer. You can ask for a plan first, then the steps, then a check of the result. This is often called step-by-step or chain-of-thought reasoning.

Some newer models are trained to reason internally before replying and are often called reasoning or "thinking" models. They tend to be stronger on maths, logic, coding and planning, but are usually slower and cost more per answer.

Which models use it?

OpenAI offers reasoning-focused models within its product line.
Claude models can use extended thinking for harder problems.
Gemini models offer thinking modes on some versions.
Many chat assistants show a "thinking" or "reasoning" option you can switch on.

How to use it

Describe the goal and the constraints clearly.
Ask the model to propose a short plan before doing the work.
Review the plan and correct anything wrong or missing.
Ask it to carry out the plan one step at a time.
Ask it to check its final answer against the original goal and constraints.
For important results, verify key steps yourself.

Benefits

More reliable answers on multi-step problems.
A visible plan lets you catch mistakes early.
Breaks big tasks into manageable parts.
Makes the model's assumptions easier to see.

Limits and risks

Written reasoning can look convincing even when a step is wrong.
Reasoning modes are slower and can cost more.
Not needed for simple tasks, where it only adds length.
The explanation a model gives may not fully reflect how it actually reached the answer.

Prompt template

Copy it, replace the parts in [brackets] and paste it into any AI chat.

Goal: [what you want to achieve].
Constraints: [time, budget, rules, tools].
First, write a numbered plan with no more than [number] steps and list any assumptions. Wait for my approval. After I approve, carry out the plan one step at a time, and at the end check the result against the goal and constraints and point out anything uncertain.

Planning and step-by-step reasoning