# The AI house rules

One page for everyone who works with AI – on the job and in private life: what goes into the prompt window, what never does, and the three rules in between. Pin it up, pass it on, work by it. Replace the [placeholders], cut what doesn't fit.

---

## Goes into the AI

- Your own drafts, your own texts, your own ideas.
- Anything already public: websites, press releases, published figures.
- Thinking and structuring work: outlines, summaries of public sources, help with wording.
- Anonymised cases: "a client in construction" instead of name and place.

## Never goes into the AI

- Names and data of other people – clients, colleagues, friends, family – unless they've explicitly agreed.
- Health, money, conflicts: anything that can harm a specific person. That covers the personnel file and your friend's story alike.
- Credentials, passwords, keys. No exceptions.
- Contracts, internal figures, third-party material under NDA.

**The rule of thumb:** If you wouldn't email it to a company you don't know, it doesn't belong in a prompt window either.

## The three rules in between

1. **AI answers are drafts.** Check numbers, names and claims before anything goes out. The AI sounds most confident when it's guessing.
2. **Whoever sends it, owns it.** "The AI wrote that" is not an excuse – what you send out carries your name.
3. **When in doubt: anonymise or ask.** Take the names out, blur the details – at work you ask [responsible person], in private you ask the person concerned. The doubt is the signal, not the problem.

## When something goes wrong

Say it openly, contain it quickly, learn from it. In a team: whoever reports a mistake gets help – whoever hides it makes it bigger. Point of contact: [name]. And if it concerns another person's data, tell them – the earlier, the smaller the damage.

---

Free to use. From hohl.rocks – more on safe AI use: ki-sicherheit.jetzt