The Magic of the Phrase 'Think Step by Step'
A legendary prompt engineering technique (Zero-Shot Chain-of-Thought). Adding the phrase 'Let's think step by step' compels the neural network to unfold a chain of intermediate calculations, reducing logical errors by 50–70%.
1. Concept Overview & Systemic Problem
In 2022, researchers from the University of Tokyo and Google made a remarkable discovery that became a sensation in the scientific community. They tested a large language model on complex mathematical problems: the model provided correct answers only 17.7% of the time.
However, once the scientists added a simple magic phrase at the end of the prompt:
"Let's think step by step"
...the accuracy of the same model skyrocketed to 78.7% without any retraining!
Mental Model: the ultimate free hack that instantly transforms the model from a careless word generator into a meticulous analyst.
2. Why AI Needs a "Draft"
Recall how you solved a complex problem like (124 * 7) - 348 in school. If the teacher demands an answer in one second without paper, you are almost certain to make a mistake. But if you are allowed to write down steps 1 and 2 on scratch paper, you will solve the problem easily and correctly.
┌─────────────────────────────────────────────────────────────┐
│ HOW STEP-BY-STEP REASONING WORKS │
├─────────────────────────────────────────────────────────────┤
│ ❌ Request without the phrase: │
│ "A farmer has 15 apples. He sold a third and then ate 2.│
│ What is the final answer?" │
│ The model rushes to give a number: ➔ "Answer: 7" [ERROR]│
├─────────────────────────────────────────────────────────────┤
│ ✅ Request with the phrase "Think step by step": │
│ The model writes intermediate conclusions: │
│ Step 1: There were 15 apples. │
│ Step 2: A third of 15 is 15 / 3 = 5 apples sold. │
│ Step 3: Remaining: 15 - 5 = 10 apples. │
│ Step 4: Ate 2 apples: 10 - 2 = 8 apples. │
│ ➔ "Final answer: 8 apples" [100% CORRECT] │
└─────────────────────────────────────────────────────────────┘
3. How to Use This Trick in Everyday Work
You don't have to use just one specific English phrase. Equivalent phrases work excellently in other languages:
- For calculations: "Before giving the final number, outline all intermediate formulas and calculation steps."
- For complex solutions: "Break down the solution to this task into 4 sequential logical stages. Do not proceed to the conclusion until you justify each stage."
- For code verification: "Trace the variable x step by step from initialization to exit from the function."
4. Production Engineering Scenarios
If you notice that the AI is making an annoying logical error in calculations or reasoning — do not blame the model. Simply hit the edit button on the prompt and add the magic sentence at the end: "Think through your reasoning step by step." In 9 out of 10 cases, the error will resolve itself!
FAQ: The Magic of the Phrase 'Think Step by Step'
Related terms
Chain of Thought (CoT)
A methodology that prompts a language model to generate sequential intermediate reasoning steps before producing a final answer, converting additional tokens (Test-Time Compute) into quality and accuracy of the output.
Zero-Shot CoT vs Dynamic Reasoning
The historical and practical evolution of reasoning techniques: from the simple phrase 'Let's think step by step' (Zero-Shot Chain-of-Thought) to native computational budgeting of reasoning in modern models.
OpenAI o-Series / Reasoning (Advanced Reasoning Models)
The new generation of artificial intelligence from OpenAI (o1, o3 series) is optimized for multi-step hidden reasoning, advanced mathematics, quantum physics, and complex algorithmic coding.
Related Ecosystem Resources
Explore practical workflows, core terminology, and engineering tools
Agent Scratchpad
A temporary working memory area within the model's context window where the agent records its intermediate thoughts, plans, doubts, and conclusions before invoking tools. Hidden from the end user to maintain the cleanliness of the final interface.
Chain of Thought (CoT)
A methodology that prompts a language model to generate sequential intermediate reasoning steps before producing a final answer, converting additional tokens (Test-Time Compute) into quality and accuracy of the output.
Agentic AI Roadmap: Step-by-step implementation of agent systems in business
Agent AI is no longer an experiment. 79% of top managers are implementing agents right now, and most of them are making the same mistakes at the start.
Prompt Guide: How to Build a Website with ChatGPT — Steps, Prompts, and Limits
Comprehensive step-by-step guide to building websites with ChatGPT: proven prompts for structure, HTML/CSS generation, copy polishing, design rules, and ChatGPT Sites overview.
ultrathink returns to Claude Code CLI
Claude Code has brought back the ultrathink command, allowing users to increase reasoning depth during the current turn when a task turns out to be more complex than expected.
Claude Code cuts background subagent status update costs by 99%
In the CLI version of Claude Code, status updates for background subagents now use Haiku 5.5 based on short notes, reducing limit consumption.