25+ Claude Hacks, Settings and Token-Saving Tips
Use Claude more effectively with practical hacks, essential settings, prompt techniques and token-saving workflows collected into one complete guide.
The Claude hacks that make the biggest difference are switching on the settings that are off by default (web search, instructions, Skills, connectors, memory), handing multi-step work to Cowork, prompting with the goal instead of the steps, and saving tokens by starting fresh chats and matching the model to the task.
Most people open Claude, type a question, and never touch a setting. That works, but it leaves most of what Claude can do switched off. This guide collects 30+ Claude hacks in one place: 8 one-time settings, 6 productivity tools, 14 prompt techniques, 5 token-saving habits, and one safety rule.
What "Claude Basic Setup" Actually Means
Claude basic setup is turning on the features that are off by default, such as live web search, custom instructions, saved processes, connections to apps like Gmail or Drive, and memory across conversations, so every new chat starts with more context instead of a blank slate. None of these cost anything extra on top of your plan. They're built into Claude and need to be switched on once.
Part 1: The 8 Core Settings Worth Turning On
Set these up roughly in this order. Each one builds on the last.
Turn on Web Search, and use Research for deeper answers
By default, Claude answers from what it already knows, which has a training cutoff. Click the "+" next to the message box and enable Web Search. For anything time-sensitive (prices, current events, recent releases), this is the difference between a guess and a checked answer. For bigger questions, switch on Research mode so Claude searches more widely before answering.
Move from the browser to the desktop app
The Claude desktop app carries more capability than the browser tab: local file access, connector integrations, and more stable long-running sessions. If you use Claude daily, it's worth the one-time install.
Turn a repeated process into a Skill
If you type the same instructions every time ("format this as a table," "review this the way I like it"), save it as a Skill once. After that, trigger it by name or command instead of re-teaching Claude the same process in every conversation.
Write your preferences into global instructions, and build an "About me"
Custom instructions (set once in your settings) tell Claude how you like answers formatted, what tone to use, and what to always check before responding. Add a strong "About me": your role, your audience, your work, and your writing style. Every future chat inherits it, so personalisation improves without repeating yourself.
Match the reasoning effort to the task
The effort setting in the model picker controls how much Claude thinks before answering. Low effort is fine for quick, simple questions. For anything with several moving parts, such as planning, analysis or debugging, raise the effort so Claude can reason through the problem before committing to an answer.
Pick the right model for the job, not just the default
Claude is a family of models. Claude Haiku 4.5 is the fastest, for simple tasks. Claude Sonnet 5 is the balanced everyday choice. Claude Opus 5.5 is for long, demanding coding and knowledge work. Claude Fable 5.1 is the most capable, for the hardest long tasks. Match the model to what you're asking for.
Connect the apps you already work in
Connectors let Claude read directly from Gmail, Slack, Google Drive, and similar tools, with your permission, instead of you copy-pasting context into the chat. Set them up once and Claude can pull in the relevant document, thread, or email on its own.
Turn on chat memory and build a Project
"Search and reference past chats" lets Claude recall context from earlier conversations. Projects go a step further: bundle your files and instructions once, and every chat inside that Project carries that context across sessions.
Part 2: 6 Productivity Tools Most People Never Open
Once the basics are on, these move Claude from a chat window to something that plans, automates, and delivers finished work.
Use Cowork instead of treating Claude like a chatbot
Delegate complex, multi-step work through Claude Cowork. It breaks the task down, works through each part across your files and tools, and delivers a finished result, instead of you managing it prompt by prompt. Cowork is available on paid plans.
Point Claude at a folder and let it work in your files
Give Claude a project folder in the desktop app and it can read, understand, and edit your documents right there. No more copying content in and out of the chat to get something reviewed or updated.
Use Dispatch to run Claude from your phone
Give Claude instructions, get updates, and keep work moving when you're away from your desk. Being away from your laptop no longer means the work stops.
Automate recurring work with Scheduled Tasks
Set a task up once (a weekly report, a daily briefing, a recurring review) and Claude runs it hourly, daily, weekly or on weekdays. Scheduled tasks run remotely, so they keep going when your computer is asleep, unless the task needs a folder on your computer.
Turn data into reports and dashboards with Artifacts
Instead of a wall of plain text, Artifacts turn Claude's output into interactive reports and dashboards you can use and share.
Talk instead of typing with Voice Mode
Speak to Claude and it responds hands-free. Useful for thinking out loud, or any time typing isn't convenient.
Part 3: 14 Prompt Techniques for Better Answers
Give Claude the goal, not step-by-step instructions
Describe the outcome you want and let Claude work out the process. Micromanaging each step usually gives a worse result than a clear goal.
Ask Claude to ask you questions when information is missing
Tell Claude to ask you clarifying questions before it starts. It fills gaps with your answers instead of guesses.
Ask for five completely different versions
Before settling on one draft, ask for five versions that take different angles. Choosing from options beats refining the first idea.
Push beyond the basics for creative work
Claude's first creative answer is often the safe one. Ask it explicitly to go beyond the obvious ideas.
Never ship the first draft
Treat the first output as a starting point. Give specific feedback and refine it.
Ask Claude to audit its own response
After an answer, ask Claude to review it for errors, gaps and weak arguments. It often catches problems it missed the first time.
Upload multiple documents and ask for patterns
Give Claude several files at once, such as interview notes, reports or feedback, and ask what patterns run across them.
Give Claude your hardest problems
Don't save Claude for easy tasks. It is most useful on the problems you've been stuck on.
Use Claude mainly for coding and knowledge work
Writing, analysis, research and code are where Claude is strongest. Lean on it there.
Stop relying on old prompt libraries
Long copied prompt templates were written for older models. Current Claude models do better with a clear goal and your own context.
Double-check important answers
Claude can sound confident even when it's wrong. Verify facts, numbers and sources before you use them.
Don't mistake agreement for accuracy
If Claude agrees with you, that doesn't make you right. Ask it to argue the other side.
Outsource the thinking, not your understanding
Let Claude do the heavy lifting, but make sure you understand the result well enough to explain and defend it.
Focus on the outcome instead of the process
Judge the work by whether it does the job, not by whether Claude followed your preferred route to get there.
Part 4: Token-Saving Tips
Every token costs money, or a share of your plan's usage limits. These habits keep more of it for the work that matters.
Start a fresh chat when a conversation gets long
Long chats become less effective, and Claude re-reads the whole thread each turn. When a topic is done or the chat drifts, start a new one.
Clear old chats to cut unnecessary context
Remove chats you no longer need so old, irrelevant context doesn't get pulled back into new work.
Clean up messy documents before you share them
Edit rough notes into shape in an editor like Obsidian first, so Claude reads only what matters.
Use the smallest model and effort that does the job
Usage grows from Haiku to Sonnet to Opus to Fable. Use Haiku or Sonnet at low effort for quick lookups and simple edits, and save Opus, Fable and high effort for hard problems.
Put repeated context in a Project, Skill or instructions
Pasting the same brief into every chat wastes tokens and time. Store it once where Claude picks it up automatically.
One Safety Rule
Never paste passwords, API keys, or sensitive credentials
Keep secrets out of chats, files and prompts you share with Claude.
Default Claude vs. a Properly Set Up Claude
The model is the same either way. What changes is how much context and capability it's allowed to use.
| Setting | Left on default | After basic setup |
|---|---|---|
| Current information | Answers from training data only | Checks live sources via Web Search |
| Repeated instructions | Re-typed in every chat | Saved once as a Skill or global instruction |
| Reasoning depth | One fixed effort level | Matched to task difficulty |
| Your workspace context | Pasted in manually each time | Read directly via connectors |
| Conversation history | Forgotten after each chat | Referenced automatically, or carried in a Project |
| Multi-step work | Managed through back-and-forth prompts | Delegated whole to Cowork and delivered finished |
| Your files | Copy-pasted into the chat one at a time | Read and edited directly from the folder |
| Being at your desk | Required to give Claude any instruction | Optional — Dispatch runs it from your phone |
| Recurring tasks | Manually re-run every time | Automated on a schedule |
| Sharing results | Plain text you reformat yourself | Interactive reports and dashboards via Artifacts |
One trade-off worth knowing: higher reasoning effort and larger models take longer to respond and use more of your usage limits. Reserve the highest effort and the biggest models for hard tasks. For quick lookups and simple edits, the fast, lightweight settings are the better choice.
Key Takeaways
Claude's most useful features (web search, memory, connectors, Skills) are off by default and take a few minutes to switch on.
Set global instructions, an "About me" and Skills up once, so you stop re-explaining yourself in every chat.
Cowork, Dispatch and Scheduled Tasks let Claude run multi-step or recurring work without you prompting every step.
Prompt with the goal, ask for options, and never ship the first draft. Verify anything important.
Save tokens with fresh chats, the smallest model that works, and context stored once in a Project.
FAQ
How do I save tokens in Claude?
Start a new chat when a conversation gets long, use Haiku or Sonnet at low effort for simple tasks, and store repeated context in a Project, Skill or custom instructions instead of pasting it each time. Cleaning up documents before you upload them also helps.
Do I need a paid plan to use these settings?
The Free plan includes chat, web search, memory, connectors and Research. Projects, Cowork, scheduled tasks, Claude Code and Claude Design need a paid plan such as Pro, Max, Team or Enterprise. Check your account settings to see what's on yours.
Which Claude model should I use by default?
Claude Sonnet 5 is the balanced everyday choice. Use Claude Haiku 4.5 for quick, simple tasks, and move to Claude Opus 5.5 or Claude Fable 5.1 for long, high-stakes work where the extra usage is worth it.
What's the difference between Claude Skills and custom instructions?
Custom instructions are always-on preferences that apply to every chat. A Skill is a specific, repeatable process, like a formatting template or a review checklist, that you trigger only when you need it.
What is Claude Cowork used for?
Cowork is for handing Claude a real, multi-step task instead of walking it through each step. You describe the outcome, and Claude works across your files and tools and delivers finished work such as a report, spreadsheet or organised folder.
More Claude Tips Coming Soon
Follow along for more practical Claude settings, prompts and workflows.
Follow on Instagram →