Why This Matters

A botched setup is the silent killer of Claude adoption. The most common reason new users give up in week one is hitting the Free tier's daily limit or never discovering Projects. Spending ten minutes on a deliberate setup saves hours of confusion later.

Concept Explained

Setting up Claude is a three-step pipeline: account → plan → configuration. Each step has a small failure surface, and the failure is usually silent — Claude looks like it's working but isn't, or works but with the wrong privacy posture.

How It Works

Once you have an account, the claude.ai interface has three things: (1) a model picker dropdown (top-right); (2) a Projects sidebar (left); (3) a chat input (bottom). Sign-in uses your email and password. There's no OAuth flow for claude.ai itself; API keys are separate.

Step-by-Step Tutorial

1. Step 1 — Create an account

Go to claude.ai and click "Sign up". Use an email you control. Verify your email. You're now on the Free tier.

2. Step 2 — Subscribe to a plan (if needed)

For most beginners, Free is fine for the first week. When you hit daily limits, upgrade to Pro. Visit the plans page to subscribe.

3. Step 3 — Configure privacy

Visit Settings → Account. Two settings deserve attention: training data opt-out (turn off if you don't want your conversations used for model training — Pro and higher default to off) and activity log (review periodically).

4. Step 4 — Pick your model

In the chat interface, use the model dropdown (top-right). Sonnet 4 is the default — good balance of speed and quality. Opus 4 is most capable but slower and uses more of your quota. Haiku is fast and cheap.

5. Step 5 — Trigger your first three conversations

Try: (1) "Explain what a JWT is in 3 sentences"; (2) "Draft a 100-word email apologizing for a missed meeting"; (3) "What are 5 questions I should ask before signing a SaaS contract?"

Real-World Example

A new engineer joined a fintech and was given a Claude Team seat. They followed the official quickstart, which skipped the privacy settings step. Three weeks later, during a security review, their team lead noticed the training-data opt-out was still on its default (which for Team plans is off — meaning Anthropic may use the data for service improvement, though Team data is not used for model training). The engineer's prompts — including some that referenced internal API patterns — were technically being used for service improvement logging. The fix took two minutes; the trust repair took longer.

Example Prompts / Commands / Code

# After sign-in, the claude.ai interface shows:
- Top-right: model picker (Sonnet 4 / Opus 4 / Haiku)
- Left sidebar: Projects + new chat
- Bottom: chat input with attachment button

# Hover the model picker for tooltips on each model.

# In claude.ai chat:

# Basic prompt
Explain X in 3 sentences

# With file attachment
[click paperclip, attach PDF]
Summarize this PDF in 5 bullet points

# With Project context
[select Project from left sidebar]
Based on this Project's knowledge, draft X

# Slash commands (if available)
/summarize
/translate

Common Mistakes

• Skipping the privacy settings — defaults are not always appropriate for work.
• Assuming your API key works for claude.ai — it doesn't.
• Forgetting to pick a model — defaults are fine but not always optimal.
• Not testing after setup — silent failures surprise later.

Best Practices

• Set up Claude on your primary browser first.
• Take a screenshot of your privacy settings page after setup.
• If your company has a Claude policy, follow it instead of defaults.
• Use a personal access token with minimal scopes for API experimentation.
• After any UI update, briefly check Claude still works.

Troubleshooting

Practical Exercise

Professional Challenge

Frequently Asked Questions

Further Reading

• Claude.ai
• Claude plans

Official References

• Claude.ai
• Claude settings

Related lessons: BE-03, BE-05, BE-06