What the pre-check asks
The pre-check itself is a tool in your browser. This page is its content as text — so you can see beforehand what is coming, and so a search engine or an AI assistant can read the questions at all.
No account, no email address, nothing transmitted: your entries stay in your browser. The pre-check is free and stays free.
Step 1 · Does the Regulation reach you at all?
Two questions. On a clear no we say so and you are finished here — that is explicitly a possible outcome.
1 · Do you use AI in the business?ChatGPT, Copilot, Gemini, an AI feature inside software you already run, a chatbot on your site — occasional use counts.
2 · Are you a deployer, not a provider?You use other people's AI. You don't develop or sell AI systems under your own name.
Step 2 · Prohibited practices (Article 5)
Five practices the Regulation forbids. The pre-check asks these first, because no file helps if one applies — then the answer is to stop, not to document.
- Emotion recognition on staff or trainees —
- Scoring people on general social behaviour —
- Inferring sensitive traits from biometric data —
- Untargeted collection of facial images —
- Deliberate exploitation of vulnerability —
Step 3 · Which systems are in the house?
Two entries per system: what it is used for, and who comes into contact with it. The classification proposal follows from that combination.
Used for
- Writing, summarising, translating text
- Software development
- Internal analysis, research, knowledge search
- Chatbot or assistant with customer contact
- Generating images, audio or video
- Realistic depictions of people, voices or places (deepfake-capable)
- Published text on matters of public interest
- Recruitment, personnel decisions, performance assessment
- Creditworthiness, credit scoring, insurance scoring
- Biometric identification or categorisation
- Access to essential services (education, health, social benefits)
- Control of critical infrastructure
Who comes into contact with it
- Staff only, internal
- Customers
- The public
Step 4 · The classification — five exits, fixed order
The classification is a fixed decision tree, not a language model. The order is fixed, and every proposal names the branch that fired.
Prohibited — Article 5Then no file helps. The pre-check says so and stops.
High risk — Annex IIIHiring, creditworthiness, biometrics, access to essential services, critical infrastructure. Here we do not classify — we halt and refer you to legal counsel.
UndeterminedAutomated decisions about people in an area we cannot place with confidence. That too is recorded as such rather than guessed.
Limited risk — Article 50Transparency duties. Who is bound turns on the paragraph: 1 and 2 bind the provider, 3 and 4 bind you as the deployer.
Minimal riskNo mandatory duty under the AI Act for this system. A permissible result — recorded and dated, not omitted.
Step 5 · What you end up with
In the full file: the transparency notices for the systems above, the AI usage policy in German and English, and the system register carrying the classification you confirmed for each system — dated documents.
Start the pre-check — free, no account →