Introduction

Are you using Claude like this—opening the chat, typing a question, waiting for a response, and then leaving? Next time you open it, it feels like a brand new “stranger”?

To be honest, this is not much different from using a search engine.

Claude has been available for two years, with daily active users exceeding ten million worldwide. Yet, most users only tap into about 10% of its capabilities.

This isn’t because it’s too complex, but because no one has explained what the remaining 90% looks like.

In this article, we will clarify that 90%. After reading this, your Claude will “know” you and “understand” you, working alongside you like a true partner.

1. Basic Setup: Let Claude “Remember” You

1. Create a Project Instead of Starting a Chat

Every time you start a new conversation, Claude is in a “forgetful” state. It doesn’t know who you are, what you do, or what style of responses you prefer.

You either spend the first few messages reintroducing yourself or endure its “one-size-fits-all” replies.

Solution: Use a Project.

A Project is a persistent workspace where all conversations share memory. You set it up once, and from then on, Claude will automatically start working with an understanding of you.

It’s simple: Open Claude → Click on Projects in the sidebar → Create a new project → Name it “Work” or “Personal”.

All future conversations will take place within this Project.

2. Tell Claude Who You Are

Most people skip this step and then wonder why Claude’s responses always feel “off.”

Fill out the following template and paste it into the Project’s knowledge base:

My name is [Your Name].
My profession is [Your Role], and I mainly handle [2-3 Tasks].
My current most important goals are [1-3 Specific Goals].
I primarily use Claude for [Writing/Research/Analysis/Learning/Programming, etc.].
My knowledge background: [What you are good at, what you are learning, what you know nothing about].
My preferred response style: [Concise/Direct/Detailed with examples/Step-by-step/Avoid bullet points, etc.].
What I do not want to see: [No disclaimers/No corporate tone/No restating my question/No starting with "Good question"].
My areas of interest: [Your interests, industry, niche].

The more specific you are, the more tailored Claude’s responses will be.

3. Have Claude Write “Custom Instructions”

Having background information is not enough. You also need a set of Custom Instructions that tell Claude the default behavior mode when interacting with you.

After filling out the above information, send this prompt to Claude:

Based on all the information I’ve provided, help me write a set of custom instructions for my Claude project. Requirements: Describe who I am, set the default communication style and format, specify what should never be done, define the tone, and include default behaviors to follow in every conversation. Write in the second person, keeping it under 400 words.

Paste the generated content into the Project Instructions, and you’re done.

Set it once, and permanently change your experience.

2. Cognitive Upgrade: Claude Is Not What You Think

4. Claude Is Not a Search Engine

Most people treat Claude like Google—inputting questions and waiting for answers.

This is the lowest value usage.

Claude is not an information retrieval tool; it is a thinking partner. It can reason, synthesize, debate, and delve deeper based on context. If you treat it like a search engine, you cut off 80% of its value.

Don’t ask Claude, “What is XXX?” Instead, ask, “Help me clarify XXX.”

❌ “What is prompt caching?”

✅ “I’m working on a workflow that requires calling Claude 20 times per session. Help me analyze the mechanism of prompt caching and whether it can actually reduce costs in my scenario.”

The first question asks for a definition, while the second invites Claude to help solve a problem.

5. Let Claude Ask You Questions First

This is a powerful technique that almost no one uses.

Before Claude starts any complex task, have it gather information from you first.

Why? Because without this step, Claude will make assumptions. If those assumptions are wrong, you’ll spend time correcting things that could have been done right the first time.

Universal phrase:

Before you begin, ask me the 5 most important questions to help you do this well. Once I answer, you can start.

Or for specific tasks:

I need you to help me write a client outreach email. Before you start writing, ask me what you need to know to create truly good (not generic) content.

3. Advanced Usage: Operations Even Old Users Don’t Know

6. Style Cloning—Have Claude Write in “Your Voice”

Claude’s default writing is grammatically correct but often misses the right tone. It reads like AI because it is using its own writing style.

Solution: Provide Claude with 3-5 samples of your own writing for it to analyze your style.

Here are 3 samples of my writing:
[Paste Sample 1]
[Paste Sample 2]
[Paste Sample 3]

Analyze my writing style in detail: sentence length, rhythm, word choice, how I start and end paragraphs, what I avoid, formality, and any features that make my writing recognizable.

After the analysis, whenever I ask you to write anything, strictly match this style. Do not revert to your default mode.

After analysis, Claude’s output will resemble your own writing, no longer sounding like a “writing robot.”

7. Use Claude as a “Sparring Partner”

Most people ask Claude for ideas—Claude just goes along with you, expanding your thoughts and providing affirmation.

This can be useful sometimes, but it’s not pressure testing.

Before making any important decisions, have Claude challenge your ideas:

Here’s my plan: [Describe your plan]

Your task is to destroy it. Identify every potentially erroneous assumption I’ve made, find every way it could fail, and argue against it as strongly as possible. Don’t hold back, don’t use qualifiers, just attack.

After attacking, help me build a strong argument supporting my viewpoint.

Finally, tell me your true opinion.

Good ideas can withstand criticism; what’s dangerous is never having let anyone challenge them.

8. Enable “Deep Thinking” Mode

Most Claude users have never activated this feature.

Extended Thinking will prompt Claude to reason step-by-step before providing an answer, rather than just outputting directly.

Simple tasks don’t require this. But for complex decisions, deep analyses, or when you need Claude to “truly think” rather than just “pattern match”—turn it on.

Operation: Click the brain icon before sending a message. Or add this line to your prompt:

Think carefully before answering. Reason step-by-step, show your thought process, mark areas of uncertainty, and then provide a conclusion.

The quality of answers to difficult questions will visibly improve.

9. Have Claude Help You Write Prompts

This is the most underrated usage.

If you’re unsure how to instruct Claude to get the best results—ask Claude to write the prompt itself.

It knows better than you what kind of instructions yield better responses.

I need Claude to help me [describe your task]. Help me write the best prompt, including role setting, context, format requirements, and constraints. After writing, execute this prompt directly.

4. Token Saving Tips: Spend Less, Get More

10. Specify Output Length

Claude typically writes to a length it deems appropriate—often much longer than you need. More tokens, more reading time, more noise.

Before asking, directly tell it how long you want the response:

• • “Answer in 3 sentences.”
• • “Give me 5 bullet points, no explanations, just the points.”
• • “Keep it under 150 words.”

This single instruction can reduce token consumption for most tasks by 40-60%, while still providing the information you need.

11. Eliminate “Fluff Openings”

Claude defaults to adding unnecessary introductory phrases—“Good question, let me analyze that,” restating your question, adding disclaimers, and summarizing at the end.

All of these are unrequested, consume tokens, and waste your time.

Add to your custom instructions:

Never start with an opening statement, affirmation, or restatement of my question. Get straight to the answer. Unless I specifically request, do not add a summary at the end. Avoid disclaimers unless the topic genuinely requires it.

12. Don’t Repeat Background Information Every Time

If you paste the same background information in every new conversation—you’re wasting tokens, and this waste will grow with increased usage.

This is precisely why Projects and Custom Instructions exist. Set it once, and let Claude automatically read it each time, eliminating the need to repeat.

13. Start a New Conversation When Changing Topics

Claude carries all context from the current conversation. When you suddenly change topics in a long discussion, all previous content remains, leading to more token consumption, slower processing, and interference with the new topic.

When changing topics, start a new conversation within the Project. You retain the Project memory while shedding irrelevant baggage.

One conversation should only tackle one task; this is the simplest way to save money.

5. Ready-to-Use: 4 High-Value Prompt Templates

14. Feynman Learning Technique—Use Analogies to Understand Anything

Claude’s default explanations are often “technically correct but practically useless”—it uses terminology similar to the concept itself, leaving you with just the definition without real understanding.

Explain [Topic] to me using analogies and everyday examples. Avoid any technical jargon. Assume I have no background in this field.

After each analogy, ask me a question to test my understanding. Based on my answer, decide whether to delve deeper or adjust the explanation.

Continue until I can explain it in my own words without using any technical terms.

15. Personalized Travel Planning

I’m going to [Destination] for [Number of Days] days. My daily budget is about [Amount] (including accommodation).

My travel style is: [Describe—do I prefer a slow pace or a packed schedule, tourist spots or local favorites, museums or food, sports or leisure, etc.].

What I want to avoid: [Crowds, tourist traps, expensive restaurants, long-distance travel, etc.].

Create a daily itinerary for me, including where to stay, what to do, where to eat, and any transportation information I need to know. Mark any items that need to be booked in advance.

16. Monthly Spending Analysis

I’m going to paste last month’s spending details. Help me analyze:
1. Where did I spend the most?
2. Which expenses seem unusual relative to my goals?
3. What can I cut without significantly affecting my quality of life?
4. What important things might I be spending too little on?
5. One specific change that could have the biggest financial impact.

My spending: [Paste bills or spending list]
My current financial goals: [Describe what you want to do—save money, pay off debt, invest, etc.]

17. Claude as Your “Thinking Partner”

Most people lack someone in their lives who will listen non-judgmentally, ask the right questions, and help clarify their thoughts on what’s troubling them.

Claude can play this role.

I want to discuss something I’m struggling with. Your task is not to give immediate advice.

First, ask me questions to fully understand the situation: What happened, how I see it, what I’ve tried, and what outcome I hope for.

After understanding the full picture, summarize what you heard—not just the facts, but also the "underlying" feelings you sensed.

Only after that, give your honest opinion, including parts I might not want to hear.

What I want to discuss is: [Describe what you want to clarify]

18. Business Idea Pressure Testing

I have a business idea and want to pressure test it before investing a lot of time.
The idea is: [Describe in detail—what it is, who it’s for, how it makes money, and why you think it’s viable].

Your task is to identify all the issues:
1. What potentially erroneous assumptions have I made?
2. Who is already doing this, and why might I lose to them?
3. Why might target users not pay for this?
4. What prerequisites must be met for this to succeed, and how likely are they?
5. What is the single biggest problem with this idea?

Be specific. Vague risks like "the market might not be ready" are not useful. Give me the real version of each question.

After analysis, tell me what this idea needs to look like to truly work.

Conclusion

Claude is not smarter than you. It doesn’t have better ideas than you.