Most people open Claude, type into it like it’s Google, get a bland answer, and decide it’s overrated.
They’re not wrong about the answer. They’re just using it before it knows anything about them. Out of the box, Claude has no idea who you are, what you do, or what you’re working on.
This guide fixes that. You do not need any technical experience. There’s no code, nothing to install beyond one app, and nothing you can break. If a step doesn’t apply to you, skip it.
Before you start
What you need: a Claude account. If you don’t have one, go to claude.ai and sign up. The free plan is enough for everything here except one optional part, which is flagged.
How to use this guide: do the steps in order. Each one says how long it takes and how to tell it worked.
One note on wording. Software changes. If a button is called something slightly different on your screen, look for the closest match. Everything here is in Settings or on the main chat screen.
Step 1. Get the desktop app
Time: 2 minutes. Most people only ever use Claude in a web browser. The app you install on your computer can do more.
- Go to claude.com/download.
- Click the button for your computer: Download for macOS or Download for Windows.
- Open the file and follow the prompts, the same as installing any other app.
- Open Claude and sign in with the same account.
It’s available for Mac, Windows and ChromeOS, with a Linux beta for Ubuntu and Debian.
Why bother
- Cowork, which does multi-step jobs across your own files rather than just answering questions
- Claude Code, for anyone who builds software
- Desktop extensions, which let Claude reach files and apps on your actual computer
- Quick entry, a keyboard shortcut that opens Claude from anywhere
Cowork and Claude Code need a paid plan. Ordinary chat works on every plan including free.
You’ll know it worked when Claude is sitting in your dock or taskbar and opens without a browser.
Step 2. Tell Claude who you are
Time: 3 minutes. This is the one almost nobody does, and it’s the one that changes the most. There’s a box in Settings where you describe yourself once, and Claude reads it at the start of every conversation.
- Click the small circle with your initials in the bottom left corner.
- Click Settings.
- Find the box labelled Instructions for Claude.
- Type into it. Then save.
What to write
Copy this, paste it into the box, and replace anything in square brackets. Delete any line that doesn’t apply.
Who I am:
[Your name]. I'm a [your job] at [company or "working for myself"], based in [city].
Day to day I [describe what you actually spend your time doing].
What I'm working on right now:
- [project or client one]
- [project or client two]
How I want you to talk to me:
Be direct and give me the answer first, not a long build-up.
If you think my idea has a problem, tell me that before you help me do it.
Don't flatter me or open with compliments.
Use [Australian / British / American] English.
What to skip:
Don't explain [things you already understand well] to me.
Don't add a warning or disclaimer to every answer.
Don't give me five options when one good one will do.A filled-in example
Who I am:
Sarah. I'm a bookkeeper working for myself, based in Brisbane.
Day to day I do BAS returns, payroll and monthly reconciliations for about fifteen small business clients.
What I'm working on right now:
- Moving four clients from spreadsheets onto Xero
- Writing up a standard onboarding process for new clients
How I want you to talk to me:
Be direct and give me the answer first.
If my approach has a problem, say so before helping me do it.
Don't flatter me. Australian English.
What to skip:
Don't explain basic accounting terms to me.
Don't tell me to consult an accountant, I am one.The mistake to avoid
Most people write only the first part, who they are, and stop. The sections on how to talk to you and what to skip are where the quality actually jumps.
A good habit: any time you catch yourself correcting Claude on the same thing twice in different conversations, that correction belongs in this box.
You’ll know it worked when you start a brand new conversation, ask something about your work, and it answers at your level without you explaining your situation first.
Step 3. Connect your tools
Time: 2 minutes. A connection lets Claude read from something you already use, like your email or files, instead of you copying and pasting things in.
- Open a new conversation.
- Look for the connectors or tools menu near the message box.
- Pick a service and click to connect it.
- Sign in to that service and approve access.
You can see everything available at claude.com/connectors.
The three worth starting with: Gmail, so it can read the actual email thread. Google Drive, so your documents are searchable inside a chat. Slack, so it can find what your team already decided.
Before: you open the email, copy it, paste it into Claude, then ask your question. After: you ask “find the thread with the Henderson job and draft a reply agreeing to the new timeline” and it goes and gets it.
Only connect what you’ll actually use. You can disconnect any of them later from the same menu.
You’ll know it worked when you ask Claude to find something in a connected service and it comes back with the real file or message, not a guess.
Step 4. Learn the model picker
Time: 1 minute. There’s a dropdown at the top of the chat where you choose between versions of Claude. Most people never click it. Ignore the names, they change. Think about three sizes:
| Size | Best for |
|---|---|
| The light one | Quick answers, lookups, simple summaries. Fastest. |
| The middle one | Writing, analysis, coding, anything with a few steps. The everyday choice. |
| The heavy one | Deep research, complicated reasoning, long documents. Slowest, and it uses more of your allowance. |
Stay on the middle one. Drop to the light one for quick jobs. Move up to the heavy one only when the middle one has already had a go and didn’t do well enough. Which options appear depends on your plan.
Step 5. Make a project
Time: 3 minutes. A project is a folder for one ongoing thing. You put your files and rules into it once, and every conversation inside it already knows them.
- Find Projects in the menu down the left side.
- Create a new one and name it after the client or thing you’re working on.
- Upload your files into it.
- Add project instructions, which work like Step 2 but only inside this project.
Projects work on every plan including free. On the free plan you can have up to five.
What to put in it: your brand or style guide, three or four examples of work you were happy with, templates you reuse, and any list of products or jargon you’d otherwise have to explain.
You’ll know it worked when you start a conversation inside the project, ask it to do something using your files, and it does it without you uploading anything.
The final check
- The Claude app opens from my dock or taskbar
- The Instructions for Claude box has all four sections filled in, not just my job title
- At least one service is connected and I’ve tested it
- I’ve opened the model dropdown and know which one I’m on
- One project exists with my real files already in it
If something looks different
I can’t find the box in Settings. Click your initials in the bottom left, then Settings. It’s a large text box. If the wording is slightly different, look for the closest match.
A connector asked for a lot of permissions. That’s the standard sign-in screen for that service. You can revoke access any time.
I only see one or two models. That’s a plan thing, not a mistake.
It still feels generic. Nine times out of ten it’s Step 2. Add the “how I want you to talk to me” and “what to skip” sections properly. Vague instructions produce vague results.