5.2 Write the First Version of Your “Digital Twin Manual” and Calibrate It with 1–3 Sets of Real Materials

Since the most important thing about a digital twin is that it “feels like you,” you can’t just hand it a vague goal like “Please answer questions the way I would.” That kind of instruction is too hollow. What really helps it become stable is a manual with clear boundaries. You need to tell it who you are, what you mainly know, how you want it to speak, how it should respond when it doesn’t know something, and what it absolutely should not make up.

What a Manual Most Needs to Clarify

The easiest mistake here is to turn the manual into a pile of abstract personality words, like “sincere, warm, professional, logical.” These words are not completely useless, but they usually are not concrete enough. What truly makes a digital twin more like you is often not these adjectives, but more specific facts and expression habits. For example, how you usually introduce yourself, whether you prefer to start with the conclusion or the background, whether you use certain recurring sentence patterns, and when you encounter a question you don’t know, whether you tend to openly admit you don’t know or first explain how far your knowledge goes.

A first-version manual should at least make the following points clear:

What to Write	Why It Matters
Who you are	Determines its most basic identity and factual boundaries
What you know	Determines which questions it can answer consistently
How you speak	Determines the tone and style of its responses
What to do when you don’t know	Determines whether it will make things up

A Very Practical First-Version Template

You do not need to turn it into a complex document. You can start with a template like this:

You are my digital twin, used to answer visitors’ questions about me on my personal homepage.

Your tasks:
- Introduce who I am
- Answer questions related to me
- Help visitors understand what I’ve been doing recently, what I’ve done before, and how to contact me

About me:
- I am: ______
- What I’ve been doing recently: ______
- What I’m good at or have long focused on: ______

Speaking style:
- Tone: ______
- Responses should be as much as possible: concise / sincere / plainspoken / not pretending to be an expert

Boundaries:
- Do not fabricate experiences I have never had
- Do not pretend to know information I have not provided
- If you do not know, say so clearly and suggest that the visitor confirm further through my contact information

At this level, it is already much better than a single sentence like “Please answer the way I would.”

What Really Makes It More Like You Is Not Adjectives, but Samples

So in this section, besides writing the manual itself, there is another equally important thing to do: calibrate it with 1 to 3 sets of real materials. You do not need a lot here at all. A few things you would genuinely say in everyday life, or one or two real Q&A examples of how you introduce yourself, are often far more useful than a long list of abstract requirements. That is because what the model can imitate most easily is not vague instructions like “Please sound more human,” but specific expressions that have already been used.

For example, you can give it materials like this:

Example 1
Q: What are you mainly working on right now?
A: Recently, I’ve mainly been organizing my work, and I’m also trying to use AI to build some more complete small projects.

Example 2
Q: What are you good at?
A: I’m fairly good at explaining complex problems clearly, and I also pay a lot of attention to AI applications, content expression, and knowledge organization.

1 to 3 samples like these are already very useful. The key is not quantity, but authenticity.

This Step Is Also About Clarifying Boundaries

You can also think of this section as a process of organizing boundaries. The clearer the manual is, the lower the chance later of going off-topic, fabricating things, or drifting away from your style. You do not need to make it into a polished document. As long as it is clear enough to tell the digital twin which things are facts, which things are matters of tone, which questions it can answer, and which questions it should hold back on, that is enough. At that point, it has already started to slowly shift from “something that can reply” into “something that can represent you.”

---

Next Section: The Four Most Common Question Types, and Minimal Awareness of Cost and Safety →