Acceptance of AI: how staff come to trust the new technology
Acceptance grows when staff experience that AI takes tedious work off their hands and does not monitor or replace them. Take their concerns seriously and answer them in a binding way, for example in an agreement with the employee representatives. Start with tasks nobody likes doing, with volunteers and with real examples from your organisation. Answers with sources help because they can be checked. Usage quotas and pressure have the opposite effect.
Why are staff sceptical?
Because they have legitimate questions. Anyone who dismisses them as resistance loses trust. The most common concerns and what helps:
| Concern | What lies behind it | What helps |
|---|---|---|
| “AI will replace me.” | Unclear goals and consequences | State clear goals, make binding commitments with the employee representatives, make relief rather than job cuts the purpose |
| “I am being monitored.” | Logs, analyses, usage figures | Rule out monitoring in an agreement, disclose what is logged |
| “The AI makes mistakes and I am liable.” | Experience with made-up answers | Results are drafts and a person decides; show sources so they can be checked |
| “I can’t do this.” | Little experience with digital tools | Ready-made apps instead of an empty input field, training on people’s own tasks, contacts within the team |
| “My data, or that of patients and citizens, ends up who knows where.” | Reports about AI services abroad | Explain where the AI runs and which data is allowed |
| “This is even more work.” | Experience with earlier IT rollouts | Start with small, noticeable relief, do not ask for any additional documentation |
Where should you start?
With tasks nobody likes doing and where a draft saves a lot of time: summarising minutes, drafting standard replies, searching long documents, translating texts or putting them into plain language. The benefit is visible at once and the risk is low, because the person reads the result before using it.
Less suitable for the start are tasks where mistakes are hard to spot or where decisions about people are made.
Who should take part first?
Volunteers from several departments, not only the technology enthusiasts. A nurse who shows how she drafts a report in half the time convinces her team more than any presentation by the management. Ask these first people to report openly, including on what does not work. Honest limits make the good results credible.
What can managers and employee representatives do about concerns?
- Managers: They use the AI themselves, talk about their experience in the team and protect time for training and trying things out. They do not set usage quotas as a goal.
- Works council, staff council or employee representation body (in church-run organisations, the MAV): Involving them early is not only an obligation (§ 90 and § 87(1) no. 6 BetrVG, the German Works Constitution Act). An agreement that rules out monitoring and provides for training during working hours answers many concerns in a binding way. How this works is explained in the guide Codetermination on AI.
- Data protection: The data protection officer explains which data may go into which app. This gives people confidence in their daily work.
All roles in the project and the phases of the rollout are described in the guide Change management for the AI rollout.
What should you avoid?
- Usage quotas and rankings: They create pressure and amount to performance monitoring, see AI and performance or behaviour monitoring.
- Big promises: Announcing “30 % less effort” and not delivering costs credibility. Better: concrete examples from your organisation.
- Empty input fields: A chat window without guidance overwhelms many people. Ready-made apps for specific tasks lower the barrier.
- A rollout without rules: Unclear rules on data lead to uncertainty or to the use of private AI services.
- One-off training: Questions only arise during work, see Organising AI training.
How does this work with basebox?
- Ready-made apps instead of an empty chat window: Staff choose the task and get the result, without any knowledge of AI.
- Answers with sources: Every answer shows where it comes from. A click on the source opens the passage in the document (traceability).
- Recognisably AI: It is clear that people are chatting with an AI platform; a notice appears below the input and output window.
- No monitoring through the logs: Logging serves security, not performance monitoring. The content of questions and answers is logged only if your organisation switches on detailed logging; everyone then sees a notice. The dashboard shows usage, including per person, but no questions or content; what these figures are used for is set out in the agreement with the employee representatives.
- Data stays within bounds: The AI runs in our cloud in Germany or on a server dedicated to you. Your data does not train any model.
- Template and training: A template agreement for the works council or staff council (non-binding draft) and training on the use cases of your organisation.
Frequently asked questions
Should we make using the AI mandatory?
Not at the start. Voluntary use with good first examples is more convincing than an obligation. Whether and for what use will later be expected is something you agree with the employee representatives; the basebox template provides two variants for this.
What do we say when asked whether jobs will be cut?
Only what you can commit to in a binding way. Many organisations agree with their employee representatives that the rollout will not lead to redundancies for operational reasons and that changes to tasks will be discussed jointly. Especially with a shortage of skilled staff, the aim is usually to relieve people, not to cut jobs.
How do we deal with staff who reject AI?
With respect and patience. Many reservations have good reasons, such as experience with wrong answers or concerns about data. People who see colleagues saving time and being able to check results often change their minds on their own.
Sources
This article was created with the help of AI, based on the sources listed. Responsible for the content: basebox GmbH. Not legal advice, no liability. Research as of 11 October 2026.