AI transparency

The characters you meet in role-plays (a shop assistant, a colleague, a doctor) are played by the AI. They are not real people, and the facts they mention may be invented for the exercise.

An organisation may be configured to use an alternative model through OpenRouter. Pronunciation of single words and example sentences uses your device’s built-in text-to-speech engine, not an AI provider. Details of each provider, including where data is processed, are in the Privacy Policy.

• The app presents the conversation partner as an AI throughout — from onboarding (“Real conversations with an AI tutor”) to the conversation screen.
• When the synthetic voice is speaking, the app shows that the AI is speaking. The voice is generated, not a recording of a real person.
• Corrections, feedback and level estimates come from the AI, not from a teacher, and are not an assessment.
• Course texts and translations may have been drafted with AI help and reviewed by people; machine translations into your native language are generated automatically.

• AI makes mistakes. A correction can be wrong, a real error can be missed, an explanation can be inaccurate, and speech recognition can mishear you — especially with accents, noise or short utterances.
• Bias. Language models learn from large amounts of text and can reflect its biases — for example favouring one variety of a language, or stereotypes in role-play characters. We instruct the models to stay respectful and inclusive, but cannot rule this out.
• Not a certified assessment. CEFR estimates are indicative, not aligned to or validated against official examinations, and not a certificate. Treat scores and error counts as a guide for practice.
• Not professional advice. The AI is a language partner; do not rely on it for medical, legal, financial or other important decisions.

• No emotion recognition. Inqui does not infer your emotions or intentions from your voice, face or text. Speaking rate and pauses are measured only as indicators of fluency. Where a study asks how you feel, you answer that yourself in a questionnaire.
• No biometric identification or categorisation. Your voice is transcribed, not used to identify you or to sort you by age, gender, origin or any other characteristic. We do not use the camera.
• No automated decisions with legal or similarly significant effects. AI outputs steer your practice inside the app; they do not grade you, admit you or certify you.
• No training by the AI providers; our own training only on voice recordings. The AI providers do not use your content to train their models, and we do not use your conversation texts to train AI models. We do store recordings of your spoken turns and may use them to train our own models that detect speech errors (for example in pronunciation or grammar). In regular use you can switch this off at any time in Settings → Privacy; in a research study it needs your separate, optional consent. Details are in the Privacy Policy.

Inqui is intended for language practice and formative feedback: helping learners see their mistakes and progress, and choosing suitable practice. Its outputs are not intended for, and must not be used for:

• grading learners or deciding whether they pass a course, module or examination;
• admission to, access to, or assignment to an educational institution or programme;
• deciding the level of education a person will receive or be able to access;
• certifying language proficiency, including for employment, immigration or visa purposes;
• monitoring or detecting prohibited behaviour of students during tests.

For institutions (deployers): teachers may use the dashboard as one input for their own judgement about what to practise next, but any assessment of a learner must remain a human decision based on appropriate evidence. An institution that uses Inqui’s outputs for the purposes above does so outside the intended purpose and on its own responsibility, and may take on the obligations the AI Act attaches to high-risk uses in education. Institutions must also tell their learners that they will be practising with an AI system. See Section 7 of the Terms of Service.

• The AI works within instructions and guardrails that we write and review: it stays in its teaching role, stays on the learning scenario and declines harmful or unrelated requests.
• Flagged misuse is reviewed by a person — the operator and, for institutional accounts, the institution’s administrators. No account is restricted automatically.
• Teachers can see their learners’ progress and, depending on permissions, conversations, and remain responsible for teaching decisions.
• We monitor AI quality and errors (for example failed or empty replies) and change models or instructions when needed.

Report a problematic output. If the AI said something wrong, offensive, harmful or inappropriate, e-mail [email protected] with the date and time, the activity and, if possible, a screenshot or the text. Where your app version offers an in-app report option, you can use that too. A person reads every report.

Sending your conversation content to the AI providers is part of how the Service works. What is sent, to whom, where it is processed, the safeguards for transfers outside the EU and how long data is kept are explained in the Privacy Policy. Questions: [email protected].