CHAT & WRITING
The all-rounders.
Drafting texts, clarifying questions, sorting ideas. Strong at phrasing, weak at fact-checking.
- ChatGPTOpenAI · USA
- ClaudeAnthropic · USA
- GeminiGoogle · USA
- Le ChatMistral AI · France
- CopilotMicrosoft · USA
NOT EVERY
AI CAN DO
EVERYTHING.
Which tools exist, who is behind them and what to look for when choosing. No ranking, no advertising, no version numbers that will be out of date tomorrow.
WHAT DO YOU NEED?
OUR ADVICE
Use a chat assistant. Give it a role, context and format, and edit the result yourself.
Learn to prompt ↗OUR ADVICE
Use a research tool that cites sources, and open the sources. Chatbots without web search love inventing evidence.
Why AI hallucinates ↗OUR ADVICE
Use a local model or do not enter anything personal. Health, sexuality and addresses do not belong in other people's clouds.
More on privacy & ethics ↗OUR ADVICE
Try several generators and watch out for stereotypes. Label AI images when you publish them.
Report bias in images ↗OUR ADVICE
Music AI is a good starting point, not a finished product. Edit it, break it and rebuild it, the way we do in our archive.
To our music ↗THE CATEGORIES
RESEARCH
With sources, not gut feeling.
Search engines with AI summaries. Helpful if you actually open the sources they cite.
- PerplexityPerplexity AI · USA
- Web search in chatbotsvarious providers
- DeepLDeepL · Germany (translation)
CODE
For everyone who builds.
Assistants that write code, explain it and hunt for bugs. Always test the results; never copy them blindly.
- GitHub CopilotGitHub/Microsoft · USA
- Claude CodeAnthropic · USA
- CodexOpenAI · USA
- CursorAnysphere · USA
IMAGE
Words become images.
Image generators inherit the biases of their training data. Pay attention to whom they show and whom they do not.
- MidjourneyMidjourney · USA
- FLUXBlack Forest Labs · Germany
- Stable DiffusionStability AI · UK, open weights
- Adobe FireflyAdobe · USA
- GPT ImageOpenAI · USA
VIDEO
Moving images at the push of a button.
The youngest and most expensive category. Here deepfakes are not a side issue but the core risk.
- SoraOpenAI · USA
- VeoGoogle · USA
- RunwayRunway · USA
- KlingKuaishou · China
AUDIO & MUSIC
Voices and songs.
Music and voice generators. Legally contested, because many were trained on copyrighted music.
- SunoSuno · USA
- UdioUdio · USA
- ElevenLabsElevenLabs · voices & speech
LOCAL & OPEN
Your data stays with you.
Open-weight models run on your own computer without data leaving the device. A bit of tech curiosity helps.
- LlamaMeta · USA
- MistralMistral AI · France
- GemmaGoogle · USA
- QwenAlibaba · China
- Ollama / LM StudioApps for running models locally
Examples, not recommendations and not a complete list. Providers and models change fast. As of September 2026.
CHECKLIST BEFORE YOU CHOOSE
EUROPETHERE ARE
THERE ARE
ALTERNATIVES.
Mistral from Paris, FLUX by Black Forest Labs from Freiburg, DeepL from Cologne: European providers are bound by the GDPR and the AI Act. That does not automatically make them better, but it makes them more predictable.
Where does my data end up?
Server location, retention period and whether your input is used for training. Many services offer an opt-out; you just have to switch it on.
Who is behind it?
Providers are subject to the laws of their country. Take particular care with services from states without independent data protection oversight.
What does it really cost?
Free often means you pay with data or with usage limits. Subscriptions are only worth it if you use the tool regularly.
How good is it in your language?
Most models are trained mainly on English text, so quality in German and other languages varies. Test with your real tasks, not with demos.
Can I check the result?
Use tools that show sources when it comes to facts. And tools you can switch off when it comes to you.
THE BEST
MODEL IS
YOUR OWN HEAD.
Tools come and go. Judgement stays. Learn the basics and every new model is just another tool.
AI basics ↗