Prompt

In short
A prompt is the input that tells an AI what to do: a question, an instruction or a text with a task. The language model generates its answer from it. How good that answer is depends less on magic words than on whether the prompt clearly states the task, the context, the desired result and the limits. With cloud tools, whatever goes into a prompt leaves your business.
The common mistake: there is a perfect prompt
Collections of "prompt templates" often promise more than they deliver. A language model responds to context, not to a formula. A short template without details of your case returns a generic answer.
A prompt works better when it names four things: task, context, format and limits. For example: "Summarise these minutes in five points for management, list open decisions first and do not invent any figures."
What never belongs in a prompt
With a cloud tool, everything you type into a prompt or attach as a file goes to the provider. With a free personal account, your business has no say over what happens to it.
That is why every team needs a one-sentence data rule: which data never goes into a prompt, such as customer names, health or payroll data and passwords. What such a rule looks like is covered in An AI policy for SMEs.
For a Swiss SME
Swiss data protection law applies to AI too. If a prompt contains personal data, the same rules apply as for any other processing: a specific purpose, informing the people concerned and, with a provider abroad, the rules on disclosure abroad. The details are in AI and data protection in Switzerland.
How we handle it
We do not write prompts as one-offs but as part of a workflow: with fixed context, checked data and a point where a person reads the result. That is part of our AI consulting, process automation and AI governance, where people keep responsibility for important decisions.
Last reviewed:
A term in your quote that nobody explained?
In a strategy call we translate the offer in front of you – even when it did not come from us.