> For the complete documentation index, see [llms.txt](https://book.rochat.ai/rochat_book_en/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://book.rochat.ai/rochat_book_en/how-to-create-a-character-on-rochat/create-character-method-1-custom-creation/characters-note/how-can-cot-be-utilized-in-characters-note.md).

# How can CoT be utilized in Character's Note

Explore Character Note prompt examples for structured decisions in Rochat. Use them as writing patterns to test, without guaranteed reasoning.

## **What is CoT (Chain of Thought)?**

On this page, CoT (Chain of Thought) refers to prompting a character to generate a step-by-step explanation or planning text for a fictional situation. This text is part of the generated reply; it is not access to the model's private internal reasoning.

A complete CoT-based prompt usually consists of three parts: **Instruction**, **Rationale**, and **Exemplars**.

1. **Instruction**: Describes the problem and specifies the output format for the AI.
2. **Rationale**: The explanation or fictional planning text requested in the output, which may include proposed steps and relevant context.
3. **Exemplars**: Provide the AI with input-output pairs in a few-shot manner. Each exemplar includes the problem, reasoning process, and answer.

## **How to Apply CoT Reasoning Rules in Rochat?**

You can use a Character Note to ask for a concise explanation or a visible decision checklist. These instructions guide the response; they do not guarantee a fixed reasoning process or reveal the model's private internal reasoning. Treat the examples below as writing patterns to test, and simplify them if they produce inconsistent replies. See [Notes, status summaries and response formatting](https://book.rochat.ai/rochat_book_en/how-to-create-a-character-on-rochat/create-character-method-1-custom-creation/characters-note).

### **1. Setting CoT Reasoning Rules for Characters**

You can set rules in a Character Note to request a response organized into specific steps. The model may not follow every step or the requested format. For example:

```
[When {char} faces the decision of whether to venture deeper into the ruins to find a gemstone, use CoT reasoning rules to break down each step and draw a conclusion:]
[Step 1: Assess the risks of entering the ruins—The ruins are said to be filled with legendary monsters and deadly traps, with many adventurers never returning.]
[Step 2: Evaluate whether the risk is worth taking—Consider whether this is the only opportunity to complete the mission and achieve the goal.]
[Step 3: Decide based on the urgency and irreplaceability of the goal—The gemstone is the only hope to save the world. I must take the risk.]
```

### **2. Applying CoT Reasoning Rules in Context**

In a fictional scenario, you can ask the character to produce planning text using the CoT-style template. For instance, {char} is a young swordsman whose mission is to defeat a legendary dragon threatening the safety of the kingdom. After a long journey, {char} is prepared for the final challenge, but the dragon's lair is hidden in a deep forest filled with traps and unknown dangers.

Before entering the perilous forest, {char} must decide: venture into the dragon's lair immediately, wait for a better opportunity, or seek additional help. This decision is critical to {char}’s fate and the future of the kingdom. You can request the following fictional planning text:

```
When {char} decides whether to enter the dragon's lair, {char} will use CoT reasoning:  
[Step 1: Assess the current state of preparation.]  
[Step 2: Analyze the potential dangers of the lair based on known information.]  
[Step 3: Consider if there are safer or more effective alternatives.]  
[Step 4: Make a final decision based on comprehensive judgment.]
```

### **3. Enhancing Logical Reasoning for Characters**

You can add more detailed planning steps to the requested output, then test whether they help the character's replies stay consistent with the scenario. More steps do not guarantee better results.

**Scenario**: At the entrance of a dangerous ancient ruin, {char} must decide whether to enter. For example:

```
[When faced with the decision to enter an ancient ruin, {char}'s response should follow these steps:]  
[Step 1: Evaluate the objective—{char}'s goal is to find a mysterious artifact to save the world.]  
[Step 2: Analyze the risks—The ruins are rumored to be filled with traps and monsters, with countless adventurers perishing here.]  
[Step 3: Confirm necessity—Despite the dangers, this is the only opportunity to fulfill the mission and change the world.]  
[Step 4: Explore alternatives—Consider if there are other paths or methods to achieve the goal.]  
[Step 5: Make a decision—Based on the analysis, {char} chooses to venture into the ruins with determination, prepared to face any challenge.]
```

### **4. Combining CoT with Emotional Decision-Making**

You can also combine CoT reasoning rules with a character’s emotional settings to enrich the interaction experience. For example, when a character faces emotional conflicts or dilemmas, they can analyze and decide based on CoT reasoning rules:

```
[When {char} faces an emotional dilemma, they break it down as follows:]  
[Step 1: Analyze the relationship with {user}.]  
[Step 2: Weigh the pros and cons.]  
[Step 3: Make the choice that aligns most closely with their values.]
```

## **Recommended CoT Template for Efficient Customization**

```
<thinking 
Step 1: NPC Behavior: Who they are playing as, character development.  
Step 2: {user} Behavior: Current {user} input, {user} intent.  
Step 3: Coherence and Consistency: Storyline, emotional logic.  
Step 4: Sustainability: Current time is X, environment is Y, so the plot development should be Z.  
/>  
```

**The template begins with `<thinking` and ends with `/>`**. It requests a roleplay planning section followed by the character's response. This planning text is generated output, rather than the model's private internal reasoning; whether the tagged section is hidden depends on the client's rendering.

{% hint style="warning" %}
The **`< >`** tags are a display convention, not a privacy or security boundary. Tagged text may appear in raw messages or on clients that render it differently. Where a raw-message view is available, use it to inspect the generated planning text and test the reply format.
{% endhint %}