AI in your business: what can go into ChatGPT and co., and who stays in control?
Which data may go into AI tools and which may not, who checks results and what the EU AI Act requires on AI literacy. With a rule that fits on one page.

An employee copies a customer's complaint into an AI tool in the evening and adds: "Make this friendlier." Including name, address and invoice number. Another has an AI summarise a client's bill of quantities (German: Leistungsverzeichnis, the list of work items for a job). Both mean well. Both use their private account on their own phone. And nobody in the business knows about it.
This article is for small trade, construction and service businesses, for property management companies and for self-employed people who use AI or want to. It's not about which tool can do the most. It's about what is allowed to go in, who checks the results and what you should write down for that.
In short
- First clarify purpose and access: What is AI used for, with which account, and are the inputs used for training? According to the DSK, applications that don't do this are preferable.
- Personal data stays out as long as nothing else has been clarified. According to the DSK, deleting names often isn't enough, because the reference to a person remains recognisable from the context.
- A human checks and decides: AI delivers drafts. Someone in the business reads and takes responsibility for what goes out or gets decided.
- AI literacy is mandatory, even in a small business: Under Art. 4 of the EU AI Act you have to take measures. According to the Bundesnetzagentur, a certificate is not required.
- One page is often enough: approved tools, forbidden content, who checks, who to ask. In writing, with a date.
First: what for, with which access, with which settings?
The Datenschutzkonferenz (DSK), the body of the German data protection supervisory authorities, published guidance on AI and data protection in 2024. It is explicitly aimed at businesses and authorities that use AI. Four questions follow from it that you clarify before anything else.
What for? According to the DSK, you should decide in advance which tasks AI is used for. For example: drafting texts for the website, polishing general emails, turning bullet points into service descriptions. Not: "for everything that comes up".
With which account? According to the DSK, employers should provide devices and accounts for professional use. Employees shouldn't have to work with private accounts, because otherwise profiles of individual people can arise. If an email address is required, the DSK recommends a functional address of the business.
Are your inputs used for training? According to the DSK, you should check whether inputs and outputs are used for training and whether you can opt out. Applications that don't do this are preferable. With many providers this depends on the plan and the settings. So check instead of assuming. The DSK also recommends switching off training right when setting up business accounts and, where possible, not storing the history beyond the session.
Who is the provider? If you use an AI tool as a cloud service and process personal data with it, according to the DSK there is often a relationship of processing on behalf (German: Auftragsverarbeitung). Then you need a contract with the provider under Art. 28 GDPR. The DSK also points out that with such services, data often goes to countries outside the EU. You need to check that before customer data goes in.
Sounds like a lot. For most small businesses, the simplest solution is still clear: one business account, training off, and no personal data in it.
What may go in and what may not
A traffic light helps more in everyday work than a long text. Here is a suggestion you can adapt to your business.
Green, may go in:
- bullet points about a service without reference to a customer, such as "remove old flooring, check the subfloor, lay tiles"
- general texts for the website, a notice or social media posts
- questions about wording, structure, spelling
- translation of general notices, for example house rules
Yellow, only after checking:
- texts in which you replaced names with placeholders. According to the DSK, it is regularly not enough to remove names and addresses, because the reference to a person often follows from the context. A complaint about "the water damage in the attic flat on Monday" is quickly traced in a small town.
- documents from clients, such as bills of quantities or plans. Check beforehand whether you're allowed to pass them on.
Red, stays out:
- names, addresses, phone numbers, emails of customers, tenants or employees
- photos from flats or with people in them
- health data, for example a sick note. According to the DSK, it belongs to the special categories under Art. 9 GDPR, whose processing is prohibited in principle and only allowed by exception.
- personnel matters such as warnings, references or job applications
- login details and passwords
- your costing, purchase prices and terms
On the last point: under § 2 GeschGehG (the German Trade Secrets Act), information is only protected as a trade secret if you take secrecy measures that are appropriate to the circumstances. Anyone who copies their costing into an open AI service without checking should at least know what the provider does with it.
Who stays in control?
A human checks every result. According to the DSK, providers of language models themselves often make clear that the texts they generate make no claim to being correct. According to the DSK, results relating to people must be checked before they are processed further. For you this means: no AI text goes to a customer unread. You always check numbers, deadlines, prices and commitments yourself.
A human decides. According to the DSK, under Art. 22 GDPR decisions with legal effect may in principle only be made by humans. The DSK gives the example of an AI that evaluates job applications and invites people to interviews on its own. That, it says, is a violation. And according to the DSK it isn't enough if a human is only formally involved and simply adopts the suggestion because of time pressure. Whoever decides must really be able to decide.
One person is responsible. Decide who in the business answers questions about AI and approves new tools. In a small business that is often the owner. If you have a data protection officer or a works council, the DSK recommends involving both early.
AI literacy: what Art. 4 of the EU AI Act requires
Under Art. 4 of the EU AI Act (German: KI-Verordnung), deployers of AI systems must take measures that promote the AI literacy of their staff. A deployer is anyone who uses AI in their business, including the painting business with an AI tool for texts. The provision was amended in 2026. The current version explicitly says that no specific level of AI literacy has to be guaranteed.
According to the Bundesnetzagentur (Federal Network Agency), the obligation applies regardless of the sector and size of the business. The regulation doesn't prescribe a specific format, and according to the Bundesnetzagentur there is no certification requirement. The measures can take place internally or externally. The Bundesnetzagentur recommends documenting them: which measure, what content, how long and who took part.
For a small business it can look like this: you go through the AI rule in a team meeting, show two examples of where AI makes mistakes, and note the date and participants. What else the EU AI Act requires, for example for chatbots on the website, is covered in the article EU AI Act: What small businesses really need to do now.
Note: This article gives an overview and does not replace legal advice. As of: November 2026.
Your AI rule on one page
According to the DSK, businesses should issue and document clear internal rules on whether, for what and with which tool AI may be used. It explicitly recommends concrete examples of what is allowed and what is forbidden. These points belong on your page:
- Approved tools: name of the tool and which access, business accounts only
- Settings: training with inputs off, history off where possible
- What for: two to five approved tasks with an example
- What never goes in: the red list from above
- Checking: every AI text is read before sending, numbers and commitments are always checked
- Deciding: AI decides nothing about customers, tenants or employees
- Contact person: who answers questions and approves new tools
- Date and signature: and when you'll look at the rule again
Step by step in one week
- Take stock: Who uses which tool for what, with which account? Ask without blame, otherwise you won't get honest answers.
- Check what you have: Do you already have Microsoft 365 or industry software with AI features? Then look there first for the data protection rules before introducing something new.
- Choose one tool: set up a business account, check training and history in the settings, and for personal data clarify the contract for processing on behalf.
- Write the rule: the page from above, in your own words.
- Go through it with the team: briefly, with examples from your everyday work. Note the date and participants.
- Ask again after a few weeks: Where are there problems, what's missing from the rule? Adjust it.
And if you notice that your business doesn't really need AI right now: that's fine too. Then write that in the rule. At least it's clear that private accounts are off limits for business data.
How we can help
We start by looking with you at where AI really helps in your business and what you already have, for example in Microsoft 365. If a simple rule and the right settings are enough, we'll tell you. In the area of Artificial intelligence we advise you on choosing suitable tools, setting up business accounts and wording your AI rule. We support you with data protection and the responsible use of AI, but we don't replace legal advice. If AI is then to be built into processes, for example for drafts based on enquiries, you'll find more in the area of Automation.
In the free initial consultation we look together at which step makes sense for you first. We'll get back to you promptly, the way you prefer: by email or by phone.
Sources & further reading
- Datenschutzkonferenz (DSK): Orientierungshilfe Künstliche Intelligenz und Datenschutz, Version 1.0 vom 6. Mai 2024 datenschutzkonferenz-online.de
- KI-Verordnung, Art. 4: KI-Kompetenz (AI Act Service Desk der EU-Kommission) ai-act-service-desk.ec.europa.eu
- Bundesnetzagentur: KI-Kompetenz nach Art. 4 KI-Verordnung bundesnetzagentur.de
- § 2 GeschGehG: Begriffsbestimmungen (Gesetze im Internet) gesetze-im-internet.de


