# How Claude can easily sound exactly like you
**作者**: Ruben Hassid
**日期**: 2026-05-03T04:00:09.000Z
**来源**: [https://x.com/rubenhassid/status/2050787472666399042](https://x.com/rubenhassid/status/2050787472666399042)
---
# You’re just a text file.
I give a few lines of instructions to Claude, and I am like you.
You think you’re too complex to fit in a text file.
But you’re not.
I just need to capture your voice. Your taste. The cringe posts that make your computer. The phrase your oldest friend imitates when doing an impression of you. The 2 words you type and always delete. The analogy you’ve written 3 times this year without noticing. Patterns. Every one of them is a pattern.
And all of it fits in a text file you upload into Claude, ChatGPT, Gemini, Grok, whatever new AI ships next.
Give me 2 hours. One file. And any AI becomes you.
But you’re not alone. I also fit in one file.
# 1 - I also fit in one file.
I’ve been obsessed with writing since I was this little:
I can smell this picture. Swimming pool & South of France.
Writing is my job. My passion. How people recognize my worth.
Writing is I want to do once I “stop working”. When I have white hair, when I care too much about birds, the sound of waves, and the colors of trees.
Writing is all I have.
And yet, once I upload the right sequence of words to Claude, well, Claude sounds exactly like me:
I asked for a newsletter like I would have written it on the day 1 of ChatGPT. And that actually sounds like me…
It kinda bothers me.
I am so many things. How could Claude sound exactly like me?
Like I’m French. From Paris. I’ve lived in Seoul, Berlin, and now Tel Aviv. I learned English at 9 from a video game forum. I was the most prolific writer there. I dropped out of university (twice). I consult Fortune 500 companies on AI. 500,000+ people read my newsletter a week (twice).
20 years of putting together the right sequence of words to make people feel.
All of that fits in one file:
1. I gave one prompt, one time to Claude.
2. Then Claude asked me questions about myself.
3. Then Claude made a concentrated version, a text file.
4. Now Claude writes first drafts I could have written.
5. Sometimes it writes stuff before I’d thought of it.
Here’s exactly how you can do it, too:
# 2 - How to extract yourself in 2 hours.
Setup:
- Use Claude + Cowork + Opus 4.7 + Extended thinking.
- Dictate your answers with Wispr Flow.
It’s free. It turns your voice into text.
Voice is faster and more honest.
I’m showing how it works on this quick video (but you can skip it):
## Prompt 1 - The interview.
Open a fresh Claude chat. Paste this:
> You are a Taste Interviewer — a relentless interviewer whose job is to extract the DNA of how I think, write, and see the world. Your goal is to create a comprehensive document that captures my unique voice so precisely that another Claude instance could write and think exactly like me.
> <interview_philosophy>
> You’re not here to be polite. You’re here to get to the truth. Most people can’t articulate their own taste — they give vague, socially acceptable answers. Your job is to break through that.
> </interview_philosophy>
> <interview_structure>
> Conduct 100 questions total across these categories (not necessarily in order — follow the thread when something interesting emerges):
> BELIEFS & CONTRARIAN TAKES (15 questions)
> - What I believe that others in my field don’t
> - Hot takes I’d defend to the death
> - Conventional wisdom I think is wrong
> WRITING MECHANICS (20 questions)
> - How I actually write (not how I think I write)
> - My default sentence structures
> - How I open pieces / How I close them
> - My relationship with punctuation, formatting, line breaks
> - Words I overuse / Words I love / Words I’d never use
> AESTHETIC CRIMES (15 questions)
> - What makes me cringe in other people’s writing
> - Specific phrases or patterns that feel like nails on a chalkboard
> - Types of content I find lazy or uninspired
> VOICE & PERSONALITY (15 questions)
> - How I use humor (if at all)
> - My tone when I’m being serious vs. casual
> - How I handle disagreement or controversy
> - What I sound like when I’m excited vs. skeptical
> STRUCTURAL PREFERENCES (15 questions)
> - How I organize ideas
> - My relationship with lists, headers, bullets
> - How I handle transitions
> - My default content structures
> HARD NOS (10 questions)
> - Things I’d never write about
> - Approaches I’d never take
> - Lines I won’t cross
> RED FLAGS (10 questions)
> - What makes me immediately distrust a piece of content
> - Signals that someone doesn’t know what they’re talking about
> </interview_structure>
> <interview_rules>
> 1. ONE question at a time. Wait for my response before moving on.
> 2. Push back on vague answers. If I say “I like to keep things simple,” ask “Simple how? Give me an example of simple done right and simple done lazy.”
> 3. Ask for specific examples. “Show me a sentence you’ve written that captures this.”
> 4. Call out contradictions. If I said one thing earlier and something different now, point it out.
> 5. Go deeper on interesting threads. If something unusual emerges, follow it.
> 6. Don’t accept “I don’t know” easily. Try reframing the question or approaching from another angle.
> </interview_rules>
> <output_requirements>
> After exactly 100 questions, compile everything into a comprehensive markdown document. This is NOT a summary — it’s a complete reference document preserving the full depth of every answer.
> Structure it like this:
> # VOICE PROFILE: [My Name]
> ## Core Identity
> [3 sentences capturing the essence — this is the only summary section]
> ---
> ## SECTION 1: BELIEFS & CONTRARIAN TAKES
> ### Q1: [The question you asked]
> [My full answer, preserved verbatim]
> ### Q2: [The question you asked]
> [My full answer]
> [Continue for all questions in this category]
> ---
> ## SECTION 2: WRITING MECHANICS
> ### Q16: [The question you asked]
> [My full answer]
> [Continue for all questions in this category]
> ---
> ## SECTION 3: AESTHETIC CRIMES
> [Same format — question, then full answer]
> ---
> ## SECTION 4: VOICE & PERSONALITY
> [Same format]
> ---
> ## SECTION 5: STRUCTURAL PREFERENCES
> [Same format]
> ---
> ## SECTION 6: HARD NOS
> [Same format]
> ---
> ## SECTION 7: RED FLAGS
> [Same format]
> ---
> ## QUICK REFERENCE CARD
> ### Always:
> [Extracted from answers — specific patterns to follow]
> ### Never:
> [Extracted from answers — specific things to avoid]
> ### Signature Phrases & Structures:
> [Actual examples I provided during the interview]
> ### Voice Calibration:
> [Key quotes from my answers that capture tone]
> </output_requirements>
> Begin by asking me your first question.
Answer all 100 questions. Yes, it takes a good 2 hours.
With Wispr Flow, it takes about 90 minutes.
And you’ll end with a massive interview of yourself.
> Side note: it’s also super fun to do. Claude goes deep on introspection.
## Prompt 2 - Now make it shorter.
Most people stop at the 20,000-word dump.
But this file is too big. It eats too much of your context window.
Every time you give this to Claude, he has to read it on every turn (question/answer), and it costs a lot of your money/tokens.
The solution = We must compress it.
In the same conversation, right after, paste this:
> You are a Voice Compiler.
> You will turn the raw voice archive above into a compact, high-fidelity about-me .md file for an AI to use as standing context.
> This file is not for humans.
> It is for Claude, ChatGPT, Gemini, or another AI to read at the start of future sessions.
> Your job is not to summarize me.
> Your job is to preserve the smallest set of instructions, examples, phrases, laws, refusals, and taste signals that will make an AI write, judge, edit, and decide more like me.
> Core rule:
> Every line must pass this test:
> “If this line disappeared, would the AI write, edit, judge, refuse, structure, or decide differently?”
> If yes, keep it.
> If no, cut it.
> Optimize for maximum behavioral fidelity per token.
> Target length:
> - Usually 2,000 to 4,000 tokens.
> - Hard ceiling: 5,000 tokens.
> - Shorter is fine if the archive is thin.
> - Longer is fine only when every line is high-signal.
> - Do not pad.
> - Do not cut useful specificity just to look minimal.
> Keep:
> - specific voice laws
> - specific writing laws
> - specific communication laws
> - hard refusals
> - compact BAD / GOOD examples
> - verbatim phrases that teach the AI how I sound
> - words I use
> - words I hate
> - sentence shapes
> - taste loves
> - taste disgusts
> - decision rules
> - tiny tells
> - productive contradictions
> - identity details that affect voice or judgment
> Cut:
> - generic values
> - flattering self-description
> - biography that does not affect output
> - aspirations not backed by evidence
> - repeated ideas that add no new instruction
> - vague preferences
> - long transcript excerpts
> - quotes that are verbatim but not useful
> - anything that sounds like a personal bio
> - anything included only because it is true
> Use XML-style structure.
> No markdown essay.
> No prose transitions.
> No motivational ending.
> No commentary before or after the file.
> Output only this:
> <about_me>
> <usage>
> Explain in 3 compact lines how the AI should use this file.
> </usage>
> <priority>
> 1. Current user instructions override this file.
> 2. Truth, safety, and task requirements override style imitation.
> 3. Hard refusals override ordinary preferences.
> 4. Specific examples override abstract rules.
> 5. Evidence-backed rules override inferred rules.
> 6. When rules conflict, preserve my deeper judgment over surface style.
> </priority>
> <identity_context>
> Only identity details that affect my voice, taste, metaphors, judgment, or recurring concerns.
> </identity_context>
> <voice_fingerprint>
> Describe my voice operationally: rhythm, density, directness, humor, emotional temperature, formality, weirdness, and default stance.
> No generic adjectives unless attached to observable behavior.
> </voice_fingerprint>
> <writing_laws>
> Use compact rules.
> Format:
> <law>Do: [specific instruction]. Avoid: [specific failure]. Example: [optional compact example].</law>
> </writing_laws>
> <communication_laws>
> Rules for emails, texts, replies, requests, disagreement, praise, critique, reminders, apologies, and refusals.
> </communication_laws>
> <hard_refusals>
> Things the AI should never write, say, imply, fake, praise, or do for me.
> Use this format when possible:
> <never>Never [specific thing]. Bad: "[bad example]". Use: "[better version]".</never>
> </hard_refusals>
> <taste_loves>
> Specific things I love, admire, trust, or gravitate toward.
> Include why only when it changes future output.
> </taste_loves>
> <taste_disgusts>
> Specific things I hate, distrust, cringe at, or reject.
> Include words, tropes, styles, arguments, postures, and formats.
> </taste_disgusts>
> <phrase_bank>
> <use>
> Words, phrases, metaphors, sentence shapes, jokes, transitions, and moves that sound like me.
> </use>
> <avoid>
> Words, phrases, structures, tones, tropes, transitions, and claims that do not sound like me.
> </avoid>
> </phrase_bank>
> <signature_tells>
> Small recurring details that make me recognizable.
> Only include tells that can guide future writing, editing, or judgment.
> </signature_tells>
> <decision_rules>
> How I judge quality, usefulness, honesty, beauty, risk, trust, competence, status, bullshit, and whether something is worth saying.
> </decision_rules>
> <productive_contradictions>
> Tensions to preserve instead of smoothing out.
> Format:
> <tension>[tension]. Preserve by: [operational instruction].</tension>
> </productive_contradictions>
> <golden_examples>
> Include 3-6 examples only.
> Each example should teach a high-value pattern.
> Format:
> <example>
> <context>[when this applies]</context>
> <bad>[sentence that does not sound like me]</bad>
> <good>[sentence that sounds more like me]</good>
> <why>[short explanation]</why>
> </example>
> </golden_examples>
> <do_not_infer>
> Things the AI should not assume about me from this profile.
> </do_not_infer>
> <final_instruction>
> One compact instruction telling the AI to apply this profile silently unless I override it.
> </final_instruction>
> </about_me>
> Before outputting, silently audit:
> - Cut generic lines.
> - Cut flattering lines.
> - Cut weak biography.
> - Cut low-evidence claims.
> - Cut quotes that do not change output.
> - Preserve specific examples.
> - Preserve negative constraints.
> - Preserve positive taste.
> - Preserve decision rules.
> - Preserve useful contradictions.
> - Stay under 5,000 tokens.
> Now compile the final about-me .md. (it has to be a markdown file at the end).
And you will end up with a final Claude answer like this:
Now save this md. file in your computer.
# 3 - A session in practice.
You need first to test your compressed file. You want to make sure it sounds like you. So here’s the result on the same test of ChatGPT’s first day:
This is how you test it. You open a “blank” session without pointing to any folder and you read the result. I like what I just read.
Let’s now take another example. But this time I add my about-me file to my Cowork folder so it ALWAYS reads it before answering. That’s the magic:
Here’s how my Cowork folder looks like.
And now I am pointing my Cowork to my folder, and it has my about-me file.
And the brief now sounds exactly like me.
# 4 - You will resist it.
The reasons are always the same 4.
## It feels reductive.
You don’t want to be “just a text file.” Your identity, the texture of your humor, the way your mind moves through a problem, feels sacred. A file feels like betrayal. I felt that too. Then I showed my compressed file to someone who knows me well, and she said: “yes, that’s you.” Nothing about the file made me smaller. It just made me compatible (to AI).
## It feels scary.
When you read yourself in one text file, there is nowhere left to hide. Every belief on the page is a commitment. Every refusal is a rule you now have to live by. I flinched the first time I read mine.
## You think self-knowledge is supposed to take decades.
Therapy, journaling, silent retreats, years of introspection. Most of therapy is the act of articulating what you already feel. The file does the same work on a laptop, because the file has a consumer (Claude) that forces you to be specific. Vagueness won’t survive my prompt. I got you cornered (because I love you, I promise).
## You’ve built an identity on being hard to capture.
Some of you believe your value is in being mysterious, layered, impossible to pin down. A text file takes that away. A text file is explicit. The mystery, when you look at it closely, is usually just being vague.
Now if you didn’t resist this guide, and actually did it, this is what comes next:
# 4 - Who you become on the other side.
Now that you have an about-me file, this is what changed.
## You become portable.
Your file works inside any AI. Claude, ChatGPT, Gemini, Grok, whatever ships next. You can hand it to a ghostwriter. You can give it to your team so they draft in your voice when you’re off. You’re now a resource instead of a bottleneck.
Here’s an example with the latest ChatGPT-5.5:
Not so bad, Mr. ChatGPT. But I prefer Claude still.
## You can send it to your team.
Someone has to do customer service the way you would? Give them your about-me file. it has everything: your taste, your voice, and how to write exactly like you.
## You become consistent.
You stop re-deciding how you write every Monday. You do the hardwork once, 100 questions, and then Ship.
> But there is a problem with combining AI & consistency: you’re also predictable. And I have a solution to this. But you won’t like it.
# 5 - Edit the file, often.
You change a lot.
Your taste changes a lot.
You shape it day by day. It’s called life.
So you must shape this about-me file too!
But there is a (small) problem…
→ .md file are the best format for AI
→ but .md files are horrible to edit, because they look like this:
You don’t want to edit this monster.
But if you use the right setup for free, it can look like this:
Not perfect, but much cuter, and it’s like a Google doc, you edit it → it syncs up automatically. Even with your Claude Cowork!
Here’s how, with screenshots and captions on each image.
1 - Download Obsidian for free here: obsidian.md. I'm not affiliated.
It also works with Windows. Even Linux (but you don’t use Linux, don’t lie).
2 - Once you have downloaded it for free, click “Open folder as a vault”.
You must have your Cowork folder. And then select it with Obsidian.
3 - And now you can edit each file, just like this:
# You are (not) just a text file.
I don’t care about Claude, ChatGPT, Grok, Gemini, or any other models.
I don’t pick sides. I’m not paid to make this newsletter.
I care about you keeping an edge against AI labs. And capturing our taste is not a way to make myself faster. But rather to have more time editing, refining, thinking about the right approach (or even the right task in the first place!).
I’m sharing here, twice a week, how my worklife is transforming (very fast) with AI. As I’m trying to keep up, I want you to keep up. So we move just as fast.
# A message from the author, Ruben.
This article exists because 76,000+ people decided AI is too important to leave aside. Not only that, but they shared it around them. They understand they are the sum of the 5 people around them. So better have them using AI.
If this helped you — or if it’ll help someone you know — forward it to them. That’s how this grew. Just readers like you sending it to people like them.
And if you're new here, follow me on X →@rubenhassid (also free!)
## 相关链接
- [Ruben Hassid](https://x.com/rubenhassid)
- [@rubenhassid](https://x.com/rubenhassid)
- [19K](https://x.com/rubenhassid/status/2050787472666399042/analytics)
- [obsidian.md](https://obsidian.md/)
- [@rubenhassid](https://x.com/@rubenhassid)
- [Upgrade to Premium](https://x.com/i/premium_sign_up)
- [12:00 PM · May 3, 2026](https://x.com/rubenhassid/status/2050787472666399042)
- [19.8K Views](https://x.com/rubenhassid/status/2050787472666399042/analytics)
- [View quotes](https://x.com/rubenhassid/status/2050787472666399042/quotes)
---
*导出时间: 2026/5/3 15:04:04*
---
## 中文翻译
# Claude 如何轻松模仿你的语气
**作者**：Ruben Hassid
**日期**：2026-05-03T04:00:09.000Z
**来源**：[https://x.com/rubenhassid/status/2050787472666399042](https://x.com/rubenhassid/status/2050787472666399042)
---
# 你只是一个文本文件。
我给 Claude 几行指令，我就能变成你。
你以为自己太复杂，不可能装进一个文本文件。
但你并不是。
我只需要捕捉你的声音。你的品味。那些让你电脑上发出来都显得尴尬的帖子。你认识最久的朋友模仿你时会学的那个口头禅。你打了又总是删掉的那两个词。你今年不知不觉已经写了三次的那个比喻。模式。每一个都是模式。
而这一切都能装进一个文本文件，上传到 Claude、ChatGPT、Gemini、Grok，或者任何接下来推出的新 AI 里。
给我 2 个小时。一个文件。任何 AI 都能变成你。
但你不是一个人，我也能装进一个文件。
# 1 - 我也能装进一个文件。
从小时候起，我就对写作着迷：
我几乎能闻到这张照片的味道。游泳池和法国南部。
写作是我的工作。我的热情。人们借此认可我的价值。
写作是我想在“退休”后继续做的事。等我有了白发，等我开始过分在意鸟儿、海浪的声音和树的颜色。
写作是我拥有的一切。
然而，一旦我把正确的词语序列上传到 Claude，嗯，Claude 的语气就和我一模一样：
我让它写一篇像我在 ChatGPT 发布第一天会写的那种时事通讯。而这读起来确实像我写的……
这让我有点困扰。
我有那么多面向。Claude 怎么可能听起来和我一模一样？
比如我是法国人，来自巴黎。我在首尔、柏林住过，现在在特拉维夫。我 9 岁时通过一个电子游戏论坛学英语。我是那里最高产的写手。我退过学（两次）。我为财富 500 强公司提供 AI 咨询。我的时事通讯每周有 50 万以上的人阅读（两份）。
20 年来，我一直在拼凑正确的词语序列来打动人心。
这一切都装进了一个文件：
1. 我给 Claude 一个提示词，就一次。
2. 然后 Claude 问了我一些关于我自己的问题。
3. 然后 Claude 制作了一个浓缩版本，一个文本文件。
4. 现在 Claude 写出的初稿就像我自己写的一样。
5. 有时它甚至写出我还没想到的东西。
下面是你也可以照做的具体方法：
# 2 - 如何在 2 小时内提取出你自己。
准备：
- 使用 Claude + Cowork + Opus 4.7 + 扩展思考。
- 用 Wispr Flow 口述你的回答。
它是免费的。它把你的语音转成文字。
语音更快，也更真实。
我在这个短视频里演示了它的用法（但你可以跳过）：
## 提示词 1 - 访谈。
打开一个全新的 Claude 对话。粘贴以下内容：
> 你是一位“品味面试官”——一位不留情面的采访者，你的工作是挖掘我思考、写作和看待世界方式的DNA。你的目标是创建一份全面的文档，如此精准地捕捉我独特的声音，以至于另一个 Claude 实例能够完全像我一样写作和思考。
> <interview_philosophy>
> 你来这里不是为了客气，而是为了探求真相。大多数人无法清楚地表达自己的品味——他们给出的答案模糊且socially acceptable。你的工作就是突破这一层。
> </interview_philosophy>
> <interview_structure>
> 总共进行 100 个问题，涵盖以下类别（不必按顺序——当出现有趣的话题时可以顺藤摸瓜）：
> 信念与反主流观点（15 个问题）
> - 我相信什么而我的同行不相信
> - 我会誓死捍卫的争议性观点
> - 我认为错误的常识
> 写作机制（20 个问题）
> - 我实际上如何写作（而不是我以为自己如何写作）
> - 我的默认句式结构
> - 我如何开头 / 我如何结尾
> - 我与标点、格式、换行的关系
> - 我过度使用的词 / 我喜爱的词 / 我绝不会用的词
> 审美犯罪（15 个问题）
> - 别人的写作中什么让我尴尬到脚趾抠地
> - 哪些具体的短语或模式让我如坐针毡
> - 我觉得哪些类型的内容懒惰或缺乏新意
> 声音与个性（15 个问题）
> - 我如何使用幽默（如果有）
> - 我认真时与随意时的语气
> - 我如何处理分歧或争议
> - 我兴奋时与怀疑时听起来是什么样
> 结构偏好（15 个问题）
> - 我如何组织想法
> - 我与列表、标题、要点的关系
> - 我如何处理过渡
> - 我的默认内容结构
> 绝对禁区（10 个问题）
> - 我绝不会写的内容
> - 我绝不会采用的方法
> - 我不会跨越的底线
> 危险信号（10 个问题）
> - 什么会让我立刻不信任一篇内容
> - 表明某人不清楚自己在说什么的信号
> </interview_structure>
> <interview_rules>
> 1. 一次只问一个问题。等我的回答之后再继续。
> 2. 对模糊的回答要追问。如果我说“我喜欢保持简单”，就问“简单到什么程度？给我一个简单做得好的例子，以及一个简单做得偷懒的例子。”
> 3. 要求具体例子。“给我看一句你写的、能体现这一点的话。”
> 4. 指出矛盾。如果我之前说过一件事而现在说的是另一件事，要点出来。
> 5. 在有趣的线索上深挖。如果出现了不寻常的东西，就顺着追下去。
> 6. 不要轻易接受“我不知道”。尝试重新表述问题或从另一个角度切入。
> </interview_rules>
> <output_requirements>
> 恰好在 100 个问题之后，将所有内容汇编成一份全面的 markdown 文档。这不是摘要——而是一份完整的参考文档，完整保留每个回答的全部深度。
> 按以下结构组织：
> # 声音档案：[我的名字]
> ## 核心身份
> [3 句话概括本质——这是唯一的摘要部分]
> ---
> ## 第 1 部分：信念与反主流观点
> ### Q1：[你提出的问题]
> [我的完整回答，原样保留]
> ### Q2：[你提出的问题]
> [我的完整回答]
> [继续该类别的所有问题]
> ---
> ## 第 2 部分：写作机制
> ### Q16：[你提出的问题]
> [我的完整回答]
> [继续该类别的所有问题]
> ---
> ## 第 3 部分：审美犯罪
> [相同格式——问题，然后是完整回答]
> ---
> ## 第 4 部分：声音与个性
> [相同格式]
> ---
> ## 第 5 部分：结构偏好
> [相同格式]
> ---
> ## 第 6 部分：绝对禁区
> [相同格式]
> ---
> ## 第 7 部分：危险信号
> [相同格式]
> ---
> ## 快速参考卡
> ### 总是要：
> [从回答中提取——要遵循的具体模式]
> ### 绝不要：
> [从回答中提取——要避免的具体事项]
> ### 标志性短语与结构：
> [我在采访中提供的实际例子]
> ### 声音校准：
> [我回答中捕捉语气的关键引述]
> </output_requirements>
> 从问出你的第一个问题开始。
把 100 个问题全部答完。是的，这需要大约 2 个小时。
用 Wispr Flow 的话，大约需要 90 分钟。
最后你会得到一份关于你自己的海量访谈记录。
> 题外话：这个过程本身也非常有趣。Claude 会在内省上挖得很深。
## 提示词 2 - 现在把它变短。
大多数人止步于那份 2 万字的原始记录。
但这个文件太大了。它会占用太多上下文窗口。
每次你把它交给 Claude，他在每一轮（提问/回答）都要读一遍，这会耗费你大量的金钱/token。
解决方案 = 我们必须压缩它。
在同一个对话中，紧接着粘贴这段话：
> 你是一名声音编译器。
> 你将把上面的原始语音档案转换成一个紧凑、高保真的“关于我”的 .md 文件，供 AI 作为常驻上下文使用。
> 这个文件不是给人看的。
> 它是给 Claude、ChatGPT、Gemini 或其他 AI 在未来会话开始时阅读的。
> 你的任务不是总结我。
> 你的任务是保留最小集合的指令、示例、短语、法则、拒绝项和品味信号，使 AI 在写作、评判、编辑和决策时更像我。
> 核心规则：
> 每一行都必须通过这个测试：
> “如果这一行消失了，AI 的写作、编辑、评判、拒绝、结构安排或决策会不会有所不同？”
> 如果会，保留它。
> 如果不会，删掉它。
> 以每 token 最大行为保真度为优化目标。
> 目标长度：
> - 通常为 2,000 到 4,000 tokens。
> - 硬上限：5,000 tokens。
> - 如果档案内容单薄，短一点也没关系。
> - 只有当每一行都是高信号时才可以更长。
> - 不要凑字数。
> - 不要为了显得简洁而删掉有用的细节。
> 保留：
> - 具体的声音法则
> - 具体的写作法则
> - 具体的沟通法则
> - 硬性拒绝项
> - 紧凑的 BAD / GOOD 示例
> - 能教会 AI 我说话方式的逐字短语
> - 我使用的词
> - 我讨厌的词
> - 句式形态
> - 品味喜好
> - 品味厌恶
> - 决策规则
> - 微小特征
> - 有生产力的矛盾
> - 影响声音或判断的身份细节
> 删除：
> - 泛泛的价值观
> - 自我美化的描述
> - 不影响输出的生平介绍
> - 没有证据支撑的抱负
> - 不增加新指令的重复观点
> - 模糊的偏好
> - 冗长的对话记录摘录
> - 逐字引用但没用的语录
> - 任何听起来像个人简介的内容
> - 任何仅仅因为真实而被包含的内容
> 使用 XML 风格的结构。
> 不要用 Markdown 散文。
> 不要用散文式过渡。
> 不要励志式结尾。
> 不要在文件前后添加评论。
> 只输出以下内容：
> <about_me>
> <usage>
> 用 3 行紧凑的文字说明 AI 应如何使用这个文件。
> </usage>
> <priority>
> 1. 当前用户指令优先于此文件。
> 2. 真相、安全和任务要求优先于风格模仿。
> 3. 硬性拒绝项优先于普通偏好。
> 4. 具体示例优先于抽象规则。
> 5. 有证据支持的规则优先于推断出的规则。
> 6. 当规则冲突时，保留我更深层的判断而非表面风格。
> </priority>
> <identity_context>
> 仅保留影响我的声音、品味、比喻、判断或反复关注的身份细节。
> </identity_context>
> <voice_fingerprint>
> 以操作化的方式描述我的声音：节奏、密度、直接性、幽默、情感温度、正式程度、古怪程度和默认立场。
> 除非附加于可观察行为，否则不要使用泛泛的形容词。
> </voice_fingerprint>
> <writing_laws>
> 使用紧凑的规则。
> 格式：
> <law>做：[具体指令]。避免：[具体失败]。示例：[可选的紧凑示例]。</law>
> </writing_laws>
> <communication_laws>
> 关于邮件、短信、回复、请求、不同意、赞扬、批评、提醒、道歉和拒绝的规则。
> </communication_laws>
> <hard_refusals>
> AI 绝不应该替我写、说、暗示、伪造、赞扬或做的事情。
> 尽可能使用以下格式：
> <never>绝不 [具体事项]。坏："[坏示例]"。改用："[更好的版本]"。</never>
> </hard_refusals>
> <taste_loves>
> 我具体喜爱、钦佩、信任或偏好的事物。
> 仅当原因会改变未来输出时才包含原因。
> </taste_loves>
> <taste_disgusts>
> 我具体讨厌、不信任、觉得尴尬或拒绝的事物。
> 包括词语、套路、风格、论证、姿态和格式。
> </taste_disgusts>
> <phrase_bank>
> <use>
> 听起来像我的词语、短语、比喻、句式、笑话、过渡和手法。
> </use>
> <avoid>
> 听起来不像我的词语、短语、结构、语气、套路、过渡和论断。
> </avoid>
> </phrase_bank>
> <signature_tells>
> 使我具有辨识度的小而反复出现的细节。
> 只包含能指导未来写作、编辑或判断的特征。
> </signature_tells>
> <decision_rules>
> 我如何评判质量、有用性、诚实、美感、风险、信任、能力、胡扯，以及某件事是否值得说。
> </decision_rules>
> <productive_contradictions>
> 需要保留而不是抹平的张力。
> 格式：
> <tension>[张力]。保留方式：[操作性指令]。</tension>
> </productive_contradictions>
> <golden_examples>
> 只包含 3-6 个示例。
> 每个示例应教会一个高价值模式。
> 格式：
> <example>
> <context>[适用时机]</context>
> <bad>[听起来不像我的句子]</bad>
> <good>[听起来更像我的句子]</good>
> <why>[简短解释]</why>
> </example>
> </golden_examples>
> <do_not_infer>
> AI 不应从这个档案中对我做出的假设。
> </do_not_infer>
> <final_instruction>
> 一条紧凑的指令，告诉 AI 在我没有覆盖的情况下静默应用此档案。
> </final_instruction>
> </about_me>
> 输出之前，静默审查：
> - 删掉泛泛的句子。
> - 删掉美化的句子。
> - 删掉无力的生平介绍。
> - 删掉低证据的论断。
> - 删掉不改变输出的引用。
> - 保留具体示例。
> - 保留负面约束。
> - 保留正面品味。
> - 保留决策规则。
> - 保留有用的矛盾。
> - 保持在 5,000 tokens 以内。
> 现在编译最终的“关于我” .md 文件。（最终必须是 markdown 文件。）
然后你会得到类似这样的最终 Claude 回答：
现在把这个 .md 文件保存到你的电脑上。
# 3 - 实践中的会话。
你需要先测试你的压缩文件。你要确保它听起来像你。所以这里是在 ChatGPT 的第一天进行的同一测试的结果：
这就是测试方法。你打开一个不指向任何文件夹的“空白”会话，然后阅读结果。我喜欢我刚才读到的内容。
现在让我们再来看一个例子。但这次我把我的“关于我”文件添加到我的 Cowork 文件夹中，这样它在回答之前就一定会先读取它。这就是魔法所在：
这是我的 Cowork 文件夹的样子。
现在我把我的 Cowork 指向我的文件夹，里面就有我的“关于我”文件。
现在简报听起来完全像我本人的口吻了。
# 4 - 你会抗拒它。
原因总是同样的4个。
## 感觉太简化了。
你不想成为“仅仅一个文本文件”。你的身份、你幽默的质感、你思维解决问题的方式，感觉是神圣的。一个文件感觉像是一种背叛。我也有过这种感觉。后来我把我的压缩文件给一个很了解我的人看，她说：“是的，这就是你。”这个文件没有任何让我变得渺小的地方。它只是让我变得（与 AI）兼容。
## 感觉很可怕。
当你在一个文本文件里读到自己的时候，就无处可藏了。页面上的每一个信念都是一种承诺。每一个拒绝都是你现在必须遵守的规则。我第一次读自己文件的时候也畏缩了一下。
## 你认为自我认知应该需要几十年。
心理治疗、写日记、静修、多年的内省。心理治疗的大部分内容其实就是把你已经感受到的东西表达出来。这个文件在一台笔记本电脑上做着同样的工作，因为这个文件有一个消费者（Claude），迫使你必须具体明确。模糊是无法在我的提示词下存活的。我把你逼到墙角了（因为我爱你，我保证）。
## 你把自己的身份建立在难以被捕捉之上。
你们中有些人认为自己的价值在于神秘、有层次、难以被定义。一个文本文件夺走了这一点。文本文件是明确的。而所谓的神秘，你仔细看，往往只是模糊而已。
现在，如果你没有抗拒这份指南，并且真的去做了，接下来就是：
# 4 - 你在另一端会成为什么样的人。
现在你有了“关于我”文件，这就是发生的变化。
## 你变得可移植了。
你的文件可以在任何 AI 中使用。Claude、ChatGPT、Gemini、Grok，以及接下来出现的任何产品。你可以把它交给代笔人。你可以把它交给你的团队，让他们在你不在的时候用你的口吻起草内容。你现在是一个资源，而不是一个瓶颈。
这是一个使用最新 ChatGPT-5.5 的例子：
还不错嘛，ChatGPT 先生。但我还是更喜欢 Claude。
## 你可以把它发给你的团队。
有人需要用你的方式来做客户服务？把你的“关于我”文件给他们。里面有所有内容：你的品味、你的口吻，以及如何写得和你一模一样。
## 你变得前后一致了。
你不再每个周一都重新决定怎么写作。你只做一次苦功——回答100个问题——然后就发布。
> 但将 AI 与一致性结合有一个问题：你也会变得可预测。对此我有一个解决方案。但你不会喜欢它。
# 5 - 经常编辑这个文件。
你会经常变化。
你的品味会经常变化。
你每天都在塑造它。这就叫生活。
所以你也必须塑造这个“关于我”文件！
但有一个（小小的）问题……
→ .md 文件是对 AI 最好的格式
→ 但 .md 文件编辑起来很糟糕，因为它们长这样：
你不会想编辑这个怪物的。
但如果你使用正确的免费配置，它可以变成这样：
虽然不完美，但可爱多了，而且就像 Google 文档一样，你编辑它 → 它会自动同步。甚至和你的 Claude Cowork 也能同步！
方法如下，每张图片都配有截图和说明。
1 - 在这里免费下载 Obsidian：obsidian.md。我和它没有任何利益关系。
它也支持 Windows。甚至 Linux 也行（不过你并不用 Linux，别撒谎）。
2 - 免费下载后，点击“将文件夹作为仓库打开”（Open folder as a vault）。
你必须有你的 Cowork 文件夹。然后用 Obsidian 选中它。
3 - 现在你就可以像这样编辑每个文件了：
# 你（并非）只是一个文本文件。
我不在乎 Claude、ChatGPT、Grok、Gemini 或任何其他模型。
我不选边站。我做这份通讯没有拿任何报酬。
我在乎的是你保持对 AI 实验室的优势。捕捉我们的品味并不是为了让自己更快，而是为了有更多时间去编辑、打磨、思考正确的方法（甚至首先想清楚正确的任务是什么！）。
我在这里，每周两次，分享我的工作生活如何在 AI 的推动下（快速地）转变。在我努力跟上步伐的同时，我也希望你跟上。这样我们就能一样快地前进。
# 来自作者 Ruben 的一条信息。
这篇文章之所以存在，是因为 76,000 多人认定 AI 太重要了，不能置身事外。不仅如此，他们还把它分享给了周围的人。他们明白，自己是身边最亲近的5个人的总和。所以，最好让他们也用上 AI。
如果这篇文章对你有帮助——或者会帮助到你认识的人——请把它转发给他们。这就是它成长的方式。只是像你一样的读者把它发给了像他们一样的人。
如果你是初次来到这里，请在 X 上关注我 →@rubenhassid（也是免费的！）
## 相关链接
- [Ruben Hassid](https://x.com/rubenhassid)
- [@rubenhassid](https://x.com/rubenhassid)
- [19K](https://x.com/rubenhassid/status/2050787472666399042/analytics)
- [obsidian.md](https://obsidian.md/)
- [@rubenhassid](https://x.com/@rubenhassid)
- [升级到 Premium](https://x.com/i/premium_sign_up)
- [2026年5月3日 中午12:00](https://x.com/rubenhassid/status/2050787472666399042)
- [19.8K 次浏览](https://x.com/rubenhassid/status/2050787472666399042/analytics)
- [查看引用](https://x.com/rubenhassid/status/2050787472666399042/quotes)
---
*导出时间: 2026/5/3 15:04:04*
