AI Assistants at Work – How to Keep Track of Your Data

AI assistants like Microsoft Copilot, ChatGPT or Gemini have arrived in many offices. They summarise emails, draft texts and help with spreadsheets. That's genuinely useful. At the same time, it's worth taking a quick look at what happens to your input – and which open alternatives exist.

What happens to your input?

When you ask an AI assistant a question, your text – often together with content from open documents – is sent to a data centre and processed there by a language model. A few questions are interesting here:

• Where is the data centre?
• Who is involved in processing (for example other service providers)?
• How long are prompts stored?
• What may they be used for?

The big providers answer these questions in their privacy and contract documentation. Details can change over time, so it makes sense to check back now and then.

A few simple tips

1. Look at the settings: The admin interface often has options for storage locations, history and data use.
2. Read change notices: Providers announce new features and subprocessors – a quick look helps.
3. Clear team rules: For example: no health data, no client data, no passwords in AI tools.
4. Anonymise: If you need help phrasing something, simply replace names with placeholders.

A few terms, briefly explained

• Language model (LLM): The software that understands and generates text.
• Inference: The moment the model processes your request and responds.
• Subprocessor: Other companies a provider uses for processing.
• Local AI: A model that runs on your own computer or server.

The open alternative: AI on your own server

Open language models have become really good in recent years. With open-source tools you can run an AI assistant where the data never leaves your premises:

Local models aren't as strong as the largest cloud models for every task. But for many office jobs – summarising, rephrasing, writing drafts – they work very well. And you know exactly where your data is.

With me, a local AI is in the works for this. If you're interested, just let me know.

What AI is really good for at work

So it's not all about caution – here are a few tasks where AI assistants genuinely save time:

• Drafts: A first draft of an email, newsletter or job ad.
• Summaries: Cutting long texts or minutes down to the essentials.
• Rephrasing: Making a text friendlier, shorter or simpler.
• Translating: Quick translations as a starting point, which you then check.
• Brainstorming: Ideas for events, social media posts or product names.
• Spreadsheets and formulas: Help with setting up a table or a tricky formula.

For all of these: the AI provides a suggestion, you make the decision.

A small team rule to copy

Many teams get on well with a short agreement, something like this:

1. We use these AI tools: …
2. General texts, drafts and research are fine.
3. Health data, client and HR data and passwords are not allowed.
4. Results are always double-checked before sending.
5. When in doubt, we ask first.

A page like this is quickly written and gives everyone peace of mind.

Conclusion

AI assistants are a real help day to day. With a quick look at the settings, a few clear rules and – where it fits – a local open-source solution, you can use them with peace of mind. In my workshops I'm also happy to show you what that looks like in practice.

If you'd like to give it a try, I'm happy to help – just drop me a line.