Skip to main content

Chatting Effectively

Module: Core Features | Lesson: 1 of 7 | Time: ~10 minutes

What You Will Learn

  • How to write clear, specific prompts that get great results
  • How multi-turn conversations work (Claude remembers what you said)
  • How to refer back to earlier messages
  • The difference between asking for explanations and asking for actions
  • When to start a fresh conversation
  • Common mistakes to avoid

Prerequisites

Why This Matters

Claude Code is powerful, but it can only be as helpful as your instructions allow. Think of it like giving directions to a friend who has never been to your neighborhood. The more specific and clear you are, the faster they arrive. Vague directions lead to wrong turns.

This lesson teaches you how to "talk" to Claude Code so it understands exactly what you need.

Writing Clear Prompts

The single most important skill with Claude Code is writing good prompts. A prompt is simply what you type at the > input line.

Be Specific, Not Vague

Compare these two prompts:

Vague PromptSpecific Prompt
"Fix this file""In index.html, change the page title from 'Untitled' to 'My Portfolio'"
"Make it better""Rewrite the README.md to include an installation section and a usage section"
"Help me with Python""Create a Python script called hello.py that prints 'Hello, World!'"

The specific versions tell Claude exactly what file, what action, and what result you expect.

The Golden Rule of Prompts

Include three things: what you want done, where it should happen, and what the result should look like. The more of these you provide, the better Claude performs.

Give Context

Claude cannot see your screen or read your mind. If you are working on a specific project, tell Claude about it:

> I'm building a personal website. The main page is index.html in the current
folder. Can you read it and suggest improvements to the layout?

This is much better than:

> Suggest improvements

The first version tells Claude what you are building, where the file is, and what kind of help you want.

One Thing at a Time

When you are starting out, ask for one thing per message. This makes it easier to review what Claude does and catch any issues.

Avoid Cramming Multiple Requests

Instead of: "Read my config file, fix the database URL, add a new section for logging, and also create a backup script"

Break it up:

  1. "Read config.json and show me what's in it"
  2. (review the output)
  3. "Change the database URL to localhost:5432"
  4. (review the change)
  5. "Add a new section for logging settings"

Once you are comfortable reviewing Claude's work, you can combine related requests into a single message. But when learning, one step at a time is safest.

Multi-Turn Conversations

A multi-turn conversation means you and Claude go back and forth over multiple messages. This is one of Claude Code's most useful features: Claude remembers everything you have said in the current session.

Claude Remembers Context

Here is an example of a multi-turn conversation:

You:   > Create a file called grocery-list.txt with milk, eggs, and bread
Claude: (creates the file)

You: > Add butter and cheese to that file
Claude: (knows you mean grocery-list.txt — you don't need to repeat the filename)

You: > Actually, remove eggs — I already have some
Claude: (removes eggs from the same file)

Notice how you did not need to say "grocery-list.txt" in every message. Claude tracked which file you were talking about from the conversation history.

Referring Back to Earlier Messages

You can reference things Claude said or did earlier in the conversation:

  • "Go back to that config file you read earlier and change the port number"
  • "Remember that error message from before? Try a different fix"
  • "Use the same format as the first file you created"

Claude understands these references because it has the full conversation in memory.

How Far Back Does Claude Remember?

Claude remembers everything from the moment you started the current session (when you launched claude in your terminal). If you close the terminal or run the /clear command, the conversation resets and Claude will not remember previous messages.

Asking for Explanations vs. Asking for Actions

Claude Code can both explain things and do things. It is important to be clear about which one you want.

Asking for Explanations (No Changes Made)

Use these when you want to learn or understand something without Claude modifying any files:

  • "What does the package.json file do?"
  • "Explain what this error message means"
  • "What is the difference between a .txt file and a .md file?"
  • "Read server.js and explain what it does, step by step"

Claude will read, analyze, and explain without touching any files.

Asking for Actions (Claude Makes Changes)

Use these when you want Claude to create, modify, or run something:

  • "Create a new file called notes.txt"
  • "Change the background color in styles.css to blue"
  • "Run the command node --version to check if Node is installed"
  • "Delete the temp folder"

Claude will use its tools (Read, Edit, Write, Bash) to carry out the action. You will see permission prompts before anything changes.

Ask "What Would You Do?" First

If you are unsure about a change, ask Claude to explain its plan before acting:

  • "How would you reorganize these files?" (explanation only)
  • "Go ahead and reorganize these files" (action)

This way, you can review the plan and then decide whether to proceed.

When to Start a New Conversation

Sometimes your conversation with Claude gets long, goes off-topic, or you want a fresh start. Here is when to reset:

Use /clear to Reset

Type /clear at the prompt and press Enter. This erases the conversation history while keeping Claude Code running. Claude will not remember anything from before the clear.

> /clear

When to Start Fresh

  • Switching tasks: If you were working on file organization and now want to debug a script, start fresh so old context does not confuse things
  • Conversation feels stuck: If Claude keeps misunderstanding you or going in circles, a reset often helps
  • After a long session: Very long conversations can sometimes make Claude less focused. A fresh start helps
  • Unrelated questions: If you want to ask something completely different from your current task

When NOT to Start Fresh

  • You are in the middle of a multi-step task: Claude remembers what it already did, so clearing mid-task means you will need to re-explain everything
  • Claude just made a change and you want a follow-up: Keep the context so Claude knows what it just did

Common Mistakes to Avoid

Here are the most frequent mistakes beginners make, and how to fix them:

Mistake 1: Being Too Vague

ProblemBetter Version
"Help""Help me understand why my index.html page shows a blank screen"
"Fix it""Fix the typo on line 12 of readme.md — 'teh' should be 'the'"
"Do something with this""Create a summary of the key points in notes.txt"

Mistake 2: Forgetting to Give Context

Claude does not know your project unless you tell it. Always mention:

  • What file or folder you are talking about
  • What the project is (if relevant)
  • What you have already tried (if troubleshooting)

Mistake 3: Asking Multiple Unrelated Things at Once

> Read my config file, also what is the weather today, and can you write
a poem about cats?

This confuses the focus. Claude will try to address everything, but the results will be scattered. Ask one topic at a time, or at least keep requests related to the same project.

Mistake 4: Not Reviewing Claude's Output

Always read what Claude shows you before moving on. If Claude edited a file, check the changes. If Claude ran a command, read the output. This is how you learn, and how you catch mistakes early.

Try It Yourself

Practice improving vague prompts. For each vague prompt below, write a better version before checking the suggested answer.

Prompt 1: Vague Version

> Make a file

Think about it: What file? What should it be called? What should be in it?

Click to see a specific version
> Create a file called shopping-list.txt containing: apples, bananas, oranges, and grapes — one item per line

This tells Claude the filename, the contents, and the format.

Prompt 2: Vague Version

> Read my project

Think about it: Which file? What do you want to know about it?

Click to see a specific version
> Read the file package.json in the current folder and explain what each section does

This tells Claude which file to read and what kind of explanation you want.

Prompt 3: Vague Version

> Search for stuff

Think about it: What are you searching for? Where? What kind of files?

Click to see a specific version
> Search all .txt files in the current folder for the word "important"

This tells Claude what to search for, where to search, and which files to include.

Practice Makes Perfect

Try typing your improved prompts into Claude Code right now. See how the specificity changes the quality of Claude's response. Over time, writing clear prompts becomes second nature.

What You Learned

  • Be specific: Include the what, where, and expected result in your prompts
  • Give context: Tell Claude about your project and goals
  • One thing at a time: Especially when learning, keep requests focused
  • Claude remembers: Within a session, Claude tracks the full conversation
  • Explain vs. act: Be clear whether you want information or action
  • Start fresh when needed: Use /clear to reset context between tasks
  • Review everything: Always read Claude's output before moving on

Help Us Improve

How was this lesson? Take 2 minutes to share your feedback — it helps us make the tutorials better for everyone.

Give Feedback →

Next Up

Next: Reading Files