Skip to main content

Claude Onboarding – First Steps for Every Employee

Welcome. This page gets you using Claude safely and productively in your first sitting — no prior AI experience assumed. Read it once, keep the checklist at the end, and you are ready.

Claude is a work tool, not a decision-maker. It drafts, summarizes, structures, and researches. You stay responsible for the result. Everything below follows from that one idea.

Who this is for

Every employee, regardless of team. If you lead a team or set it up for others, continue with the Adoption & Governance Guide. If you administer the Enterprise environment, see Schulung 4 in the Training Plan.

1. What Claude is for here​

Claude supports everyday knowledge and office work:

  • Drafting texts, emails, and documents
  • Summarizing and structuring information
  • Preparing meetings, reports, and templates
  • Sorting ideas, notes, and concepts
  • Researching and organizing what you already know

It does not replace people or decisions. Think of it as a fast, tireless assistant whose work you always check before it leaves your hands.

2. Your first 30 minutes​

  1. Sign in through the company access only. Use the company Single Sign-On (SSO). Never use a private Claude account for work content.
  2. Pick the right surface. For most people that is Claude Chat or a shared Project. Developers also use Claude Code. If you are unsure where to work, the Surfaces & Features Guide maps it out.
  3. Open a Project for recurring work. A Project keeps instructions, files, and context in one place so you do not re-explain yourself every time. Your team may already have shared Projects — use those.
  4. Try one real task. Summarize a document you may share internally, or draft a routine email. Read the result critically. That is the whole loop.
Chat vs. Project in one line

Use Chat for one-off questions. Use a Project when you keep coming back to the same topic (a customer, a product, a process) and want Claude to remember the context. See Artifacts for how Claude hands you reusable deliverables like documents and tables.

3. The data traffic light​

Before you type or upload anything, ask: which category is this?

LightMeaningExamplesRule
🟢 GreenPublic or uncritical internal infoPublic product info, general drafts, public web contentFine to use
🟡 YellowInternal info with limited confidentialityInternal drafts, non-sensitive internal documentsOnly inside approved company environments
🔴 RedSpecially protected dataPatient and health data, dental-practice customer data, access credentials, secrets/tokens, confidential contracts, critical production dataDo not enter — or only after a separate, documented approval
Red data is the line you do not cross

Our customers are bound by professional secrecy (§ 203 StGB). Never enter patient data, health data, diagnoses, or findings into Claude — not even paraphrased. The same goes for passwords, API keys, and tokens. If you are unsure whether something is red, treat it as red and ask.

4. Allowed vs. not allowed​

✅ Allowed❌ Not allowed
Drafting texts, summaries, templatesUsing a private account for company data
Structuring information for meetings/reportsEntering passwords, tokens, or secret keys
Sorting ideas, notes, conceptsProcessing sensitive personal data unchecked
Working with approved data and sourcesTaking Claude's output as final, unchecked
Using non-approved plugins, connectors, or sources

The deciding factor is always: only approved information, only approved data, always a human check.

5. The five everyday rules​

These five cover almost every situation:

  1. Use Claude only through the company access.
  2. No red data and no secrets in the prompt or context.
  3. Always check results before you use or forward them.
  4. Use only approved extensions and sources (connectors, plugins).
  5. When in doubt, ask IT, Security, or Data Protection.

Short rule: Only approved data. Only the company access. Always check results.

6. Getting genuinely good results​

Most disappointing results come from vague requests, not from the model. A few habits make a large difference:

Ask for an outcome, not a topic. "Draft a 150-word follow-up email to a dental practice that asked about service intervals, friendly and concrete" beats "write something about service intervals."

Give context up front. Who is it for, what tone, what length, what must it include or avoid. The more you frame, the less you correct.

Iterate in small steps. Change one thing at a time — "make it shorter," "add a bullet list," "warmer tone" — instead of rewriting the whole prompt. With Artifacts, each change keeps a clean version history.

Make it sound like you, not like generic AI. Paste two or three of your own past texts and ask Claude to match your tone, sentence length, and vocabulary. Tell it what to avoid (filler, buzzwords, exclamation marks). Generic output is a prompt problem, not a Claude limit.

Reuse context with Projects. For anything recurring, put the standing instructions and reference files into a Project once. Then every chat in that Project already knows the rules.

Want to go deeper on the craft?

For practical, bite-sized lessons on prompting, tone, avoiding usage limits, and getting more out of Cowork and Skills, the external learning portal claude101.com is a useful complement to this internal material. Treat it as background learning — our rules in sections 3–5 always take precedence.

7. Quick wins by role​

TeamA good first use
SalesPrepare a draft offer or structure discovery questions
MarketingStructure campaign ideas and draft variants in our tone
SupportDraft answer templates and summarize a thread (no customer-identifying data)
ManagementCondense reports and notes into a clear summary

In every case the review stays with you.

8. Common mistakes to avoid​

  • Pasting large, unstructured dumps of information with no clear task
  • Not stating what the result should actually be
  • Forwarding output without reading it critically
  • Using private tools or accounts for work
  • Misjudging how sensitive an input is

Good results need a clear task and clean boundaries.

9. Your onboarding checklist​

[ ] I use Claude only through the company access (SSO)
[ ] I never enter passwords, tokens, or secrets
[ ] I check the data traffic light before entering anything
[ ] I never enter patient, health, or customer-practice data
[ ] I use only approved extensions and sources
[ ] I check every result before using or forwarding it
[ ] I ask when I am unsure

10. Where to get help & keep learning​

All AI-generated content is a draft and requires human review before external use.