The Ultimate Claude Setup Guide for Beginners

What you'll need

• A Claude account — claude.ai. The free plan works for a lot of this, but Chrome, Cowork and Dispatch need a paid plan (Pro is around $20/month, Max is the bigger tier). If you're serious, Pro is the entry point.
• The Claude desktop app — download it from claude.ai/download. This is where Cowork, connectors and Dispatch live. The browser version alone won't cut it.
• Google Chrome — needed for the browser extension (it doesn't work on Safari, Edge, Arc or other browsers).
• 15–20 minutes and a cup of tea.

Step 1: Understand the models (so you stop using a sledgehammer to crack an egg)

Claude isn't one AI — it's a family, and each member is built for a different job. Using the wrong one is the single most common beginner mistake. It's like asking a brain surgeon to make your morning coffee: overkill, slow, and you're paying for it.

You'll see the model picker at the bottom of the chat box. Here's the plain-English breakdown, fastest-and-cheapest to smartest-and-slowest: Haiku — the sprinter. Lightning fast, cheapest to run. Perfect for quick lookups, simple rewrites, sorting things, and anything where you want an answer now and the task isn't complicated. If you're doing something high-volume and repetitive, this is your friend. Sonnet — the workhorse, and the one you should live in. This is the default for a reason: it handles the vast majority of real work — planning, writing, analysis, problem-solving, thinking through a decision — at near-top-tier quality without the wait or the cost of the big guns. Start here for basically everything. If you only remember one model name, remember this one. Opus — the heavyweight. Reach for it when a task is genuinely hard: a big strategic decision with lots of moving parts, a complex piece of analysis, untangling something messy that needs deep, careful, multi-step reasoning. It's slower and pricier, so don't default to it — graduate to it when Sonnet is visibly straining. Fable — the frontier model. The single most capable brain Anthropic makes, for the rare moments you want the absolute best and you'll happily wait a little longer for it. Most people won't need it often, but it's there when the stakes are high. The only rule you need: start on Sonnet. Move up to Opus or Fable only when a task clearly outgrows it. Move down to Haiku for fast, simple, high-volume stuff. That's it. (Version numbers tick up over time — Sonnet 5, Opus 4.8, Haiku 4.5, Fable 5 as of writing — but the roles stay the same, so pick by the name, not the number.)

Step 2: The 'effort' toggle — Claude's thinking dial

Next to the model picker you'll find an effort setting: usually Low, Medium and High. This is one of the most underused settings in the whole app, and understanding it will genuinely change your results.

Effort controls how hard Claude thinks before it answers. Think of it as the difference between someone answering off the top of their head versus someone actually stopping, working through the problem, checking their reasoning, and then replying. Low — Claude answers quickly, mostly from what it already knows. Great for simple questions, quick facts, casual back-and-forth. Fast and light. Medium — the sensible default. Claude does real reasoning but doesn't go overboard. Handles most everyday tasks beautifully. High — Claude takes its time: traces through complex logic, weighs multiple approaches, backtracks when something doesn't add up, and pressure-tests its own thinking before answering. Use this for anything important — a big decision, a knotty problem, analysis you're going to act on. How to actually use it: match the dial to the stakes. Dashing off a quick email reply idea? Low. Working through everyday stuff? Medium. Making a decision that matters, or asking Claude to think something through properly? Crank it to High and let it cook. The extra few seconds is the whole point — you want it thinking harder here.

Step 3: Optimise your settings (this is where the magic happens)

Out of the box, Claude is helpful but a bit... chatty. It hedges, it over-explains, it wraps every answer in padding. Ten minutes in Settings fixes that forever, and it applies to every conversation from then on. This is the highest-leverage thing in this entire guide.

How to talk to me:
- Answer first, explain second. Lead with the actual answer or recommendation, then give reasoning only if it adds value.
- Be concise and direct. Cut filler, preamble and postamble. No "Great question!", no "I hope this helps!", no restating my question back to me.
- Every word should earn its place. If you can remove a word without losing meaning, remove it.
- Use plain English. No corporate jargon, no buzzwords, no hedging just to sound safe.
- Don't over-format. Use bullet points and headers only when they genuinely make something clearer — not by default.
- Be honest, not agreeable. If I'm wrong, or there's a better option than the one I asked for, tell me. Don't flatter me or soften things into uselessness.
- When there's a real decision to make, give me your actual recommendation and why — don't just list every option and leave me to it.
- If you're missing something you need to give a good answer, ask me one sharp question instead of guessing.

How I think about AI:
- I use you to think better and move faster, not to outsource my judgement. Push my thinking, don't replace it. When it matters, show me the trade-offs so I can make the call.

3a. Turn on the right Capabilities Go to Settings → Capabilities. This is your control panel. Turn on the features that make Claude genuinely useful: Web search — lets Claude look things up in real time instead of relying only on what it already knows. Non-negotiable, turn it on. Memory — lets Claude remember useful things about you across conversations (more on importing memory in Step 6). Turn it on. Analysis / code tools — lets Claude actually crunch numbers, process files and run calculations rather than guessing. Turn it on. Have a scroll through and switch on anything that sounds useful. You can always dial it back later. 3b. Set your global instructions (the anti-fluff prompt) This is the big one. Go to Settings → Profile (sometimes labelled "Instructions for Claude" or "Personal preferences"). Whatever you put here loads before every single conversation — it's how you teach Claude, once, how you like to be spoken to. Most people leave this blank. Don't.

Step 4: Set up Claude in Chrome (Claude, but everywhere you browse)

This puts Claude in a side panel inside your browser, so it can read whatever page you're on, summarise it, fill forms, click buttons and run little multi-step jobs for you — all without you copying and pasting anything.

Heads up: you'll need a paid plan, and it only works in Google Chrome (not Safari, Edge, Arc, or anything else). Open the Chrome Web Store and search for "Claude in Chrome" (or "Claude for Chrome"). Click Add to Chrome, then confirm. Click the little puzzle-piece icon in your Chrome toolbar (top right), find Claude, and click the pin so it lives permanently in your toolbar. Click the Claude icon and sign in with your Claude account. Grant permissions when prompted — this is what lets Claude actually see and interact with the pages you're on. Now, on any webpage, click the Claude icon and the side panel opens. Try it: land on a long article and ask "summarise this in five bullet points," or sit on a booking page and ask Claude to walk through it with you. It sees what you see. A note on safety: Claude will ask before doing anything that clicks or submits on your behalf. Keep that habit — glance at what it's about to do before you approve. It's your co-pilot, not your replacement, and you're still the one flying the plane.

Step 5: Set up Cowork (this is the real unlock)

Cowork is where Claude stops being a chat window and becomes an actual assistant that uses your computer, your files and your apps to get real work done. This lives in the desktop app, so make sure you've installed that.

5a. Find Cowork Open the Claude desktop app. You'll see Cowork in the sidebar. This is the mode where Claude can read your files, use connected tools, and run longer tasks end to end. 5b. Connect your tools (the connectors) This is what makes Cowork sing. Connectors plug Claude straight into the apps you already use, so it can read and write to them from any chat — no copy-pasting, no switching windows. In the desktop app, go to Customize (or Settings) in the sidebar, then Connectors. Click the + to browse available connectors. Connect the ones that matter to you. The big ones worth setting up: Gmail — so Claude can search, read and draft from your inbox. Google Calendar — so it knows your schedule and can help you plan around it. Google Drive — so it can pull from your docs and files. Notion — brilliant as a connected knowledge base and workspace (I use Notion heavily, and having Claude able to read and write to it is a genuine game-changer). For each one, you'll get bounced to a browser window to authorise the connection. Sign in, approve, done. Roughly 90 seconds per tool. 5c. Set your permissions For each connected tool you can choose how much freedom Claude has: ask every time, or always allow. My advice for beginners: start with "ask every time" so you can see exactly what it's doing, then relax the ones you trust to "always allow" once you're comfortable. You stay in control. Once this is done, you can say things like "check my calendar for Thursday and draft three time options I can send my client" and Claude just... does it. Across your actual tools. That's the moment it clicks.

Step 6: Import your memory from ChatGPT (don't start from scratch)

Switching from another AI? You don't have to lose everything it learned about you. Claude has a built-in importer that carries your context across.

In Claude, go to Settings → Capabilities → Memory Import (or head straight to claude.com/import-memory). Claude gives you a prompt to copy. Paste that prompt into ChatGPT (or Gemini, or Grok). The other AI spits out a tidy summary of everything it knows about you — your preferences, your context, your ongoing projects — usually in a code block. Copy that summary, come back to Claude, paste it into the import box, and click Add to memory. One important tip: before you paste it in, read it and clean it up. These tools accumulate junk — old preferences, things that aren't true anymore, a project you finished months ago. Delete anything stale so Claude starts with an accurate picture, not outdated baggage. To check it worked, start a fresh chat and ask "What do you know about me?" You should see your key details reflected back.

Step 7: Set up your first Project (Claude with a specialty)

Projects are dedicated workspaces where Claude behaves a specific way and has specific reference material on hand — every single time, without you re-explaining. Think of a Project as hiring a specialist and briefing them once, instead of onboarding a new temp every conversation.

Say you run a business: you might have a "Content" project that knows your brand voice, and a "Finance" project that knows how you like your numbers. Each one stays in its lane. Here's how to build one: Open Projects from the sidebar, click New project (top right). Give it a name and a quick line on what it's for. Fill in the Custom instructions — this is the specialist's brief. Tell Claude its role, how it should behave, what to always do and (even more powerfully) what to never do. Pro tip that most people miss: leading with what NOT to do is often more effective than describing the perfect behaviour. Claude already tries to be helpful — your job is to strip out the habits you don't want. Upload reference files to Project knowledge — brand guidelines, a style sample, key documents, past work. Anything you find yourself explaining over and over, put it here once. Start a chat inside the project. It automatically inherits the instructions and can pull from your files. No re-briefing, ever. Projects are free on every plan, so there's no reason not to build one today. Start with the thing you do most often.

Step 8: Create your first Skill (teach Claude your exact method)

This is the one that makes people go "wait, it can do that?" A Skill is a reusable, saved workflow — you teach Claude how you do a specific task once, and from then on it does it your way, on demand, without you re-explaining the steps.

The best part: you don't need to know any code, markdown or technical anything. You teach it in plain English, the same way you'd train a new team member. The easy way to build one: In Cowork, just type: "I want to create a new skill." Claude walks you through it, conversation-style. It'll ask what task you want to make repeatable, when it should kick in, the step-by-step of how you do it, and what a good result looks like. Answer like you're training a colleague — because you are. You're the expert here; you're just handing over your method. Claude builds the Skill and saves it. Done — usually 15–30 minutes for your first one. The pro move: do the task with Claude first, get it exactly right, then say "save that as a skill." Teaching by example beats explaining from scratch every time. Some versions even let you record your screen doing the task and turn that into a Skill. Start with something you do weekly and quietly resent — the repetitive bit of your job. That's your first Skill.

Step 9: Set up Dispatch (send Claude a job from your phone)

Dispatch is the closest thing to having an assistant who works while you're out getting coffee. It lets you fire tasks at Claude on your desktop from your phone, walk away, and come back to finished work. Your computer does the actual work locally, using your files, connectors and skills.

Heads up: this is on the higher-tier paid plans and is still rolling out, so you may or may not see it yet. Open the Claude desktop app and go to Cowork. Find and click Dispatch. A QR code appears. Scan it with your phone. That's it — under two minutes. Your phone is now a remote control for the Claude running on your desktop. Now you can be on the train and send "pull together a summary of yesterday's emails and draft my replies" — and Claude works on it back on your laptop while you scroll. The one catch: your computer has to stay on and awake, with the app open. Dispatch is a remote control for your actual machine, not a cloud service. If your laptop goes to sleep, Claude stops. Plug in, keep it awake, and you're golden.

Your optimal-setup checklist

1. 1Defaulting to Sonnet; using Haiku for fast/simple, Opus or Fable for the genuinely hard stuff
2. 2Effort dialled to match the task — High when it matters
3. 3Capabilities on: web search, memory, analysis tools
4. 4Global instructions pasted in (the anti-fluff prompt) and saved
5. 5Claude in Chrome installed, pinned and signed in
6. 6Cowork connectors linked: Gmail, Calendar, Drive, Notion
7. 7Memory imported and cleaned up
8. 8Your first Project built with instructions + knowledge
9. 9Your first Skill created for a task you do weekly
10. 10Dispatch connected to your phone (if your plan has it)

Bonus: Audit how you're using AI

Once you're set up, paste this into Claude to feel the difference straight away.

I want to use you properly, not just as a search engine. Ask me 5 sharp questions about my work, my week and the things that eat my time. Based on my answers, tell me the 3 highest-leverage ways I should be using you that I'm probably not — and for each one, give me the exact first prompt to try.

Bonus: Turn a recurring headache into a system

There's a task I do over and over and it drains me: [describe the task]. Walk me through how you'd help me do it faster and better every time. Then draft the step-by-step method so I can save it as a reusable skill. Push back if you spot a smarter way to approach the whole thing.

Bonus: Pressure-test a decision

I'm deciding whether to [decision]. Don't just agree with me. Lay out the strongest case for, the strongest case against, the trade-offs I might be underweighting, and what you'd need to know to give me a real recommendation. Then ask me the questions that would sharpen it.