KIKIneAhnung
Back to Tutorials

Level 2 — Advanced

You use AI regularly and want to take your results to the next level. Here you'll learn advanced techniques: system prompts, document analysis, data work, local AI, and the limits of the technology.

Lesson 1

System Prompts and Custom Instructions

Explaining the same role every time? You don't have to. System prompts(also called “Custom Instructions”) are permanent instructions you set up once and that apply to every conversation.

Where to Find This?

  • ChatGPT: Settings → “Personalisation” → “Custom instructions”
  • Claude: Project settings → “Project instructions” or individual “Styles”
  • Gemini: Create “Gems” with predefined instructions

Example: System Prompt for Your Business

You are my personal business assistant. I run a small café in Hamburg with 5 employees. Always respond in English, in a professional but casual tone. Keep it short — max 200 words per response unless I ask for more. If you're unsure about facts, say so honestly.

With a good system prompt, you start every conversation on the right foot — without repeating yourself each time. This saves an enormous amount of time.

Lesson 2

AI Workflows for Everyday

Real efficiency comes when you connect multiple steps into a workflow. Instead of handling each task individually, you build a chain:

1

Research

Use Perplexity or Gemini to gather current facts and sources about your topic.

2

Summarise

Feed the research results into Claude or ChatGPT and have them reduced to key points.

3

Create a Draft

Turn the summary into a text, presentation, or concept — with clear instructions about the format.

4

Review and Polish

Have the AI critically review its own draft: “What are the weaknesses in this text? What's missing?” Then: your own proofreading.

Real-world example:You want to write a newsletter. Step 1: Research industry trends. Step 2: Summarise the top 3 topics. Step 3: Newsletter text in your company's voice. Step 4: AI proofreading + your final check. Result: 30 minutes instead of 3 hours.

Lesson 3

Analysing Documents and PDFs

One of the most underrated AI capabilities: you can upload entire documents and use the AI as an intelligent reader. This works with:

  • ChatGPT (Plus): Upload PDFs, Word documents, images directly in the chat
  • Claude: Particularly strong with long documents — can process entire books at once
  • Google NotebookLM: Free tool specifically for document analysis

Practical Applications

  • Upload a rental agreement: “What notice periods apply? Are there any unusual clauses?”
  • Research paper: “Summarise the key findings and explain the methodology for laypeople.”
  • Manual: “How do I set up [Feature X]? Quote the relevant sections.”
  • Contract: “Compare this contract with the previous one — what changed?”

Privacy Note

Never upload highly sensitive documents (personnel files, health data, passwords) to cloud-based AI tools. For such cases, there are local alternatives — more on that in Lesson 6.

Lesson 4

AI and Spreadsheets/Data

AI isn't just for text — it can also work with structured data. This is especially useful when you're dealing with spreadsheets, CSVs, or statistics.

Analyse Data

Upload a CSV file and ask: “What are the 3 most important trends in this data?” or “Create a summary of revenue figures by quarter.”

Generate Charts

ChatGPT can generate charts directly: “Create a bar chart from this revenue data.” Claude creates code for this that you can use in tools like Google Sheets.

Have Excel Formulas Written

Describe what you want to calculate: “I need a formula that calculates the average of column B, but only for rows where column A contains ‘January.’”

Here is a CSV file with our sales figures for the last 12 months. Analyse the data and answer: 1) Which month was strongest? 2) Is there a seasonal trend? 3) Which products have the highest revenue?
Lesson 5

Browser Plugins and Extensions

You don't always have to switch to ChatGPT. There are browser extensions that bring AI directly into your workflow:

ExtensionWhat It DoesFree?
SiderAI sidebar for any website — summarise, translate, explainBasic yes
Perplexity ExtensionQuick research with source citations directly in the browserYes
GrammarlyAI-powered grammar and style checking in real timeBasic yes
DeepLTranslations directly on websites and in text fieldsBasic yes

Tip: Choose one or two extensions and truly integrate them into your daily routine, rather than installing ten and using none.

Lesson 6

Local AI: Ollama and LM Studio

Want to use AI but keep your data from leaving your computer? You can. With local AI models, everything runs on your own machine — no cloud, no subscription.

Ollama

Free, open source, runs on Mac, Windows, and Linux. Installation takes 5 minutes. Then you can download models like Llama, Mistral, or Gemma and use them locally. Perfect for: confidential documents, offline work, experimenting.

LM Studio

Graphical interface for local AI models. Particularly user-friendly — you pick a model from a library, download it, and start chatting. No command line needed.

Important to Know

Local models are smaller than the cloud versions of ChatGPT or Claude. They're good for simple tasks but can be weaker for complex analyses or long texts. The big advantage: Your data stays with you.

Lesson 7

Recognising Limits — When AI Fails

The better you use AI, the more important it becomes to know its limits. Here are the most common pitfalls:

Hallucinations

AIs sometimes invent facts that sound plausible but are wrong. Especially dangerous with: quotes, statistics, legal advice, and medical statements.

Outdated Information

AI models have a knowledge cutoff that can be months behind the current date. For up-to-date information, use Perplexity or a search engine.

Complex Logical Reasoning

With multi-step logical problems, mathematics, or legal argumentation, AIs can make mistakes. Use Chain-of-Thought prompting (from Level 1) to reduce the error rate.

Bias and Prejudice

AI models reflect the biases of their training data. Be especially critical with statements about people, cultures, or social topics.

The golden rule:The more important the decision, the less you should rely exclusively on AI. AI is an excellent first draft — but the final responsibility lies with you.

Summary

You've now mastered system prompts, multi-step workflows, document analysis, data work, and local AI models. And you know when not to blindly trust AI. In the next level, we cover automation, AI agents, and professional integration. You can find more hands-on knowledge in our AI course.