Claude, End to EndA Hands-On Guide From First Chat to Working System

Claude answers questions, runs multi-step tasks in the cloud, builds files and designs, works in your browser and apps, and repeats work on a schedule. This guide shows you how to use each part, in the order you will need them.

How Claude fits together

Every result depends on five layers. When an answer misses, one of these is usually the reason.

Where you can use it:

• Web (claude.ai) and mobile (iOS, Android): conversations, tasks and files.
• Claude Desktop (macOS, Windows, Linux beta): everything above, plus local folders, the built-in browser, computer use and desktop extensions.
• Claude in Chrome: a side panel that reads and acts on the page you're on. Chrome only, paid plans.
• Claude for Microsoft 365: add-ins for Excel, PowerPoint, Word and Outlook. Paid plans.
• Claude Code: for software work in a code repository. Paid plans.

Check which version you have. Anthropic is merging chat and Cowork into one conversation, starting with Pro and Max. If your message box shows Chat and Cowork options, you're on the previous version. If it doesn't, you have the new one. This guide notes where the steps differ.

Set up in 10 minutes

1. Start on the web. Install Claude Desktop when you want Claude to work with local files, your browser or your apps.
2. Write instructions for Claude. Settings > General > Instructions for Claude. They apply to every conversation. Use the template below.
3. Check code execution is on. Settings > Capabilities > Code execution and file creation. Files, artifacts and skills depend on it.
4. Review memory. Settings > Memory. It's on by default on Free, Pro and Max. Everything Claude remembers is listed under Topics, where you can edit or delete it.
5. Set your training preference. Settings > Privacy. Turn off the model improvement setting if you don't want new chats used for training.
6. On a Mac, turn on quick entry. Double-tap Option to open Claude from any app, with a screenshot or window attached.
7. Check your usage. Settings > Usage shows your five-hour session limit and weekly limit.

About me: I work as [role] on [main types of work]. I already know [skills or tools], so skip the basics there.

How to respond:
- Lead with the answer, then the reasoning.
- Keep it short unless I ask for depth.
- Use tables for comparisons and numbered steps for processes.
- Tell me when you're unsure or when information may be outdated.
- If my request is ambiguous, ask one question. Otherwise state your assumption and continue.

Avoid: [filler, hype, phrases or formats you dislike].

Test it: open a new conversation and ask three things you ask often. Rewrite any instruction Claude ignored.

Choose a plan

US prices as listed on claude.com/pricing on September 16, 2026.

Know before you choose:

• Limits reset in two ways: a session limit every five hours and a weekly limit. Chat, tasks, Claude Code and Claude Design all draw from the same allowance.
• Usage credits (paid plans) let you keep working past your limit at API rates. Turn them on and set a monthly cap in Settings > Usage.
• Fable models run on usage credits on Pro and standard Team seats. On Max and premium Team seats, you can spend up to 50% of your weekly limit on them.
• Team and Enterprise content isn't used for model training by default. Free, Pro and Max users choose in Settings > Privacy.

Choose a model and effort

Click the model name next to the send button. Select More models to see the full list.

Effort controls how thoroughly Claude works. The recommended level is marked Default.

• Low or Medium for routine tasks. Your usage lasts longer.
• High for the best balance of quality and speed.
• Extra high for long coding and agentic tasks.
• Max for correctness-critical work, with longer waits.

Thinking is a separate toggle under Effort. It's always on for Fable 5.1 and Opus 5. Fable 5.1, Opus 5 and Sonnet 5 support a 1M-token context window on paid plans.

Heads up: Opus 5 switches a narrow set of flagged cybersecurity requests to Opus 4.8 and labels the reply. To pause instead of switching, turn off Switch models when a message is flagged in Settings > Capabilities.

Quick answers, tasks and code

In the new version, you don't pick a mode. Claude decides whether your request needs a quick answer or a task, and several tasks can run at once. Longer tasks keep running in the cloud after you close your laptop, and you can check in from your phone.

Tasks that touch your computer (local folders, the built-in browser, computer use) need Claude Desktop open.

Choose how much Claude checks with you using the permission setting in the message box.

Claude always asks before permanently deleting files.

Claude Code is the right tool when the work lives in a repository. Start with the Claude Code docs.

Prompts that work

A quick question needs no structure. For anything bigger, cover five parts.

Goal: [the finished thing you need]
Context: [files, links, connected apps, who it's for]
Output: [format, length, structure]
Boundaries: [what to keep, avoid, or ask me about first]
Before you finish: [the check to run on your own work]

Redirect instead of restarting. Stay in the same conversation and say:

• "Shorter. Keep the numbers."
• "Give me three different approaches before you pick one."
• "What assumptions did you make?"
• "Ask me what you need to know, then continue."
• "Which parts are you least confident about?"

Save usage: put related questions in one message, and start a new conversation when you change topics. Long conversations use more of your limit.

Let Claude ask: Claude sometimes replies with clickable options, such as a budget range or priorities to rank. Click them or type your own answer.

Give it the right context

Include what changes the answer and leave out the rest. Connectors and long files use more of your limit and your context window.

File limits: up to 20 files and 500MB per file in a conversation. Claude reads charts and images in PDFs up to 100 pages. For longer PDFs (up to 1,000 pages), it reads only the text.

Project habits:

• Keep project instructions short and general. Put task-specific details in the conversation.
• Conversations inside a project don't share context with each other. Add anything they all need to the project's knowledge.
• Reusing project files costs less usage than uploading them again. Remove files you no longer need.
• Each project has its own memory, separate from your other conversations.
• Free plans can have up to five projects.

Many connectors? Open + > Connectors > Tool access and choose On demand once you have ten or more, so they load only when needed.

Help me write instructions for a Claude project about [topic].
Ask me up to five questions about the goal, audience, sources, format and quality bar.
Then draft instructions under 200 words that every conversation in this project should follow.

Research with sources

Quick, current facts: in the new version, Claude searches the web when it helps. In the previous version, turn on + > Web search. Replies include citations.

Research (paid plans) runs many searches that build on each other and returns a cited report in minutes. It can also pull from connected Gmail, Google Calendar and Google Docs.

• New version: type /deep-research, or + > Research.
• Previous version: + > Research. Web search must be on.

Research [question] for [who will use it and what they need to decide].
Scope: [time range, regions, source types to include or exclude].
Deliver a short report: the answer first, then a comparison table, then linked sources.
Separate what sources state from your own inference. List anything you couldn't verify.

Pressure-test the report:

• "Cite the source for each major claim."
• "Which claims rest on a single source?"
• "What would change your conclusion?"
• "Which links did you actually open?"

Then open the sources behind every number, quote or date you plan to use. Claude can produce broken links or misread a source, and the original page may carry context the summary dropped.

Create files and analyze data

Claude builds Excel spreadsheets with working formulas, PowerPoint decks, Word documents and PDFs. Download them or save them to Google Drive.

Using [attached files or connected sources], create a [Word document, 8-slide deck or Excel workbook] for [audience].
Structure: [the sections, slides, sheets or columns you expect].
Cite each number to its source file and flag anything that needs my review.
Before you finish, check that [the formulas calculate correctly / the summary matches the data].

Revise without starting over:

• Name the exact spot: "slide 4, chart title" or "the Q3 tab, column D".
• Say what stays: "Keep the layout and colors."
• In a Markdown document Claude drafted, highlight the text, click Edit with Claude and type the change.

Analyze the attached [CSV or spreadsheet]. Don't change the original.
First, describe the columns, row count and any quality issues such as missing values, duplicates or inconsistent formats.
Then answer: [your question]. Show the key numbers in a table and one chart, with each finding explained in one plain sentence.

Before you use a file: open every page, tab and slide, and spot-check numbers and formulas against the source.

Artifacts, visuals and designs

Claude doesn't generate photos or illustrations. It builds things with code instead: diagrams, charts, documents, interactive tools and designs.

Visuals in the conversation

Ask "draw this as a diagram", "chart this data" or "show me how this changes over time". You can click and adjust the result, then keep asking questions. Available on web and desktop.

Visuals aren't saved automatically. To keep one, copy it as an image, download it, or save it as an artifact.

Artifacts

An artifact is something you'd put in front of someone: a document, dashboard, small app or interactive tool. It opens beside the conversation, and you can find all of yours in the Artifacts tab.

• Iterate: ask for changes, or switch between earlier versions with the version selector.
• Share: artifacts start private. Click Share and choose who can open it. People need a Claude account to open a shared artifact. In the previous version, Publish on a chat artifact creates a public link anyone can open.
• Add AI: an artifact can call Claude itself. When others use it, the usage counts against their plan, not yours.
• Save data (paid plans): published artifacts can store data. Before entering anything sensitive, check whether other users can see what's stored.

Build an interactive [tool, tracker or calculator] for [audience] to [main job].
It needs [inputs, views and controls]. Keep the design simple and readable on a phone.
List any assumptions in the tool itself.

Designs, slides and docs

On paid plans, select Output in the message box, or just ask:

• Claude Design: mockups, prototypes, one-pagers and landing pages that follow your design system. Export to PDF, PowerPoint or HTML, or send it to tools like Canva.
• Claude Slides: presentations you can edit, present and export to PowerPoint or PDF.
• Claude Docs: living documents you write with Claude and your team.

All three are in beta. For targeted design changes, click the canvas and leave an inline comment. Use the conversation for bigger structural changes.

Voice, browser and your computer

• Voice mode offers hands-free or push-to-talk, which works better in noisy places. Fable models aren't available in voice mode.
• Claude picks the fastest route: a connector first, then a browser, then your screen. Connect the apps you use most so it doesn't have to click through them.
• Computer use asks permission before each app. On macOS 15 or later, it works in background windows while you keep using your computer.

Browser and computer safety:

• Don't give Claude access to banking, health or government apps and sites.
• Logins you save in the built-in browser stay available to Claude in future sessions. Choose them deliberately.
• A page, email or document can hide instructions meant for Claude. If it starts doing something you didn't ask for, stop the task.
• Claude takes screenshots to see your screen. Close anything sensitive first.

Use the browser to [task] on [site]. Stop before submitting anything and show me exactly what you're about to enter.

Connectors, skills and plugins

Everything lives under Customize in the left sidebar.

Connectors

1. Go to Customize > Connectors, click + and pick a service.
2. Sign in and review what it can access.
3. In a conversation, turn it on under + > Connectors, or just ask and let Claude use it.

Set tool permissions per connector. Select a connector and set each tool to Always allow, Needs approval or Blocked. For example, let Claude search your email but block it from sending.

Use a desktop extension (Claude Desktop > Settings > Extensions) for tools that run on your own computer.

Skills

Once you've pasted the same prompt three times, turn it into a skill.

• Built-in skills for Excel, Word, PowerPoint and PDF run automatically.
• Write one: ask Claude to draft a skill with you. A skill is a folder with a skill.md file. Its description (up to 200 characters) tells Claude when to use it, so make it specific.
• Record one (Pro, Max and Team, Claude Desktop for Mac): click + > Record a skill, do the task once and talk through your choices. Recordings run up to about 10 minutes.
• Add it: Customize > Skills > + > Create skill > Upload a skill (a ZIP of the skill folder).
• Use it: Claude loads it when relevant, or type / to pick one.

Skills follow the open Agent Skills standard, so a well-written skill can work in other tools that support it.

Help me write a skill for [task].
It starts from: [inputs, such as meeting notes or a link].
A good result looks like: [format, length, sections]. Example: [paste one].
Always: [rules]. Never: [rules].
Before finishing, check: [quality check].
Write a specific description so you know exactly when to use it.

Plugins

Customize > Plugins > Browse plugins > Install. Plugins cover roles like sales, finance, legal, marketing and data analysis. Type / to use their skills. Plugin Create helps you build your own.

Install only what you trust. Plugins and extensions can run local programs with the same permissions as any other app on your computer.

Scheduled tasks and long-running work

Schedule a task

Scheduled tasks are on paid plans and run in the cloud, so they keep going while your computer is off.

• New version: say what to do and how often, for example "Every Monday at 9am, summarize last week's messages in #product." Review the name, schedule and instructions, then click Schedule.
• Previous version: go to Scheduled in the sidebar, then New task > Create with Claude or Set up manually. Cadences: hourly, daily, weekdays, weekly or on demand.

Manage every task from Scheduled: edit, pause, resume, run now or delete. Scheduled tasks work from your connectors and files saved to your Claude account. A task that needs local files or apps runs on your computer instead, so Claude Desktop has to be open.

Before you schedule:

1. Run the prompt by hand until the output is reliable.
2. Make sure it says what to check, what's worth reporting, and when to stop.
3. Keep it to reading and drafting. Don't schedule tasks that send, buy or delete.
4. Review the first few runs, and pause tasks you no longer need.

Every [cadence], check [connected sources, such as Slack #channel and a Google Drive folder] for [what to look for].
If nothing meets [threshold], reply "Nothing new" and stop.
Otherwise, write a brief: [format]. Draft any messages for my review. Don't send or change anything.

Not sure a schedule is the right fit? Read Routine vs Schedule vs Loop.

Hand off a long task

• Describe the outcome, format and inputs in the first message, then say "I'll check back later."
• Step away. In the new version, cloud tasks keep running when you close your laptop. You'll get a notification on your phone when Claude finishes or needs you.
• Steer mid-task from web, desktop or mobile. Stop it any time if the scope drifts.
• If Claude says it sent, saved or scheduled something, confirm it in the connected app.

[Outcome]. Deliver it as [format]. Use [files, links or connected apps].
Don't [actions to avoid]. Done when [checks it can verify, such as every figure linking to its source].

Privacy and safety

Good to know:

• Incognito chats are kept for 30 days for safety and can't be turned into regular chats. They aren't available inside projects. In the new version, they can't create files or run code.
• A public chat link shares a snapshot with anyone who has it. Treat it as public.
• Thumbs up or down sends the whole conversation to Anthropic as feedback, stored for up to five years and separated from your account ID.
• Flagged conversations may still be reviewed for safety, even with training off.

Rules to work by:

• Don't paste passwords, card numbers or data you aren't allowed to share. On a work account, follow your company's policy.
• Keep Manual approval for anything that sends, publishes, buys or deletes. Write "draft only" into the prompt.
• Give Claude a dedicated working folder instead of broad file access, and keep backups.
• Treat websites, emails and shared artifacts as untrusted. They can carry instructions meant for Claude.
• For legal, financial, medical and security decisions, use Claude to prepare, then get expert review.

Five exercises to get hands-on

Each exercise builds on the one before. Together they take you from a first prompt to a workflow that runs on its own.

1. Calibrate. Set your instructions for Claude, ask three questions you'd normally ask, and fix the instruction that didn't hold.
2. Research. Run Research on a decision you actually face. Pressure-test it with the four questions in Research with sources and open three of the sources.
3. Produce. In a project, turn real files into a document or spreadsheet you'll use this week. Revise it twice without starting over.
4. Package. Turn the prompt from exercise 3 into a skill, add it under Customize > Skills, and run it on new inputs.
5. Automate. Schedule a task that uses the skill. Review the first three runs and tighten the prompt.

Quick reference

Go deeper: Claude help center · ChatGPT, End to End · How to Give AI Taste · Routine vs Schedule vs Loop