AI transparency

1. You always know when you are dealing with AI

Before every action in which your answer is evaluated by artificial intelligence, a clear notice is shown above the button. It appears before any interaction takes place, not after it (Art. 50(1) of Regulation (EU) 2024/1689). The notice is the same for everyone — there are no accounts on this site.

2. Exactly where AI is used

This list is complete. If something is not named here, AI is not used for it:

• Evaluation of writing, picture-description and speaking tasks — the text you type or a short voice recording is assessed by an AI system, which returns points with comments.
• Speech transcription — in speaking tasks your recording is first turned into text.
• The listening recordings — these are synthetic voices, not recordings of real people. The pages that play them say so separately.
• Exercise content and illustrations — texts, answer options and images are produced with the help of AI and then checked by us.

3. What the AI evaluation means, and what it does not

The evaluation is study feedback for practice. It is not an official state language proficiency examination result, it is not given by an assessor, and it has no legal effect on you whatsoever. AI can be wrong in both directions — awarding too many points as well as too few. The official examination is organised and marked by the competent state authority, not by us. Any result can be reviewed by a human on request — write to us.

4. How AI content can be recognised automatically

Material made with AI is marked machine-readably in three layers, and all three name the same IPTC concept, so an automated tool reading any one of them receives an identical claim:

• page level — `<meta name="ai-generated-content">` on every page carrying AI-made material (the legal pages do not have it, because a person wrote them);
• content level — a `data-ai-generated` attribute on the page’s main content container;
• file level — an `X-AI-Generated` response header on every image, audio and exercise-data file.

http://cv.iptc.org/newscodes/digitalsourcetype/trainedAlgorithmicMedia

5. Who receives what you enter

For evaluation, your text or voice recording is sent to a single AI provider — OpenAI Ireland Ltd. The request carries no identifier of yours. The listening recordings are generated by Microsoft Ireland Operations Ltd. (Azure Speech) from texts we wrote ourselves; that provider never receives your recordings or your answers. The full account of recipients, legal bases and retention periods is in the Privacy Policy.

6. What we deliberately do not do

We do not use AI to make any decision about you, we build no profile of you, we do not adapt content to your behaviour, and we pass your answers to nobody for advertising. Your answers are not used to train AI models. We also deliberately do not name a specific AI model: models are replaced from time to time, and a model name in a document nobody re-reads would become a false statement.