AI Transparency Notice

    Last updated: 16 November 2026

    This notice explains, in plain language, where WhatsLove uses AI, what it does with your voice and messages, and the choices you have. It is published to meet the transparency obligations of Regulation (EU) 2024/1689 (the EU AI Act), alongside our obligations under the GDPR.

    This page is maintained by WhatsLove S.R.L.S. to answer common questions about how AI is used in WhatsLove. It describes our own practices and is not an independent audit, certification, or third-party attestation.

    1. Your voice is transcribed to text

    Voice intros and voice messages are converted to text by a speech-to-text system. The other person only hears your voice; they never see your transcripts.

    • Voice intro transcripts are used to write a short profile summary and to help suggest people to you.
    • Voice message transcripts are checked for safety, for example links, contact details, scams, harassment and sexual content. Flagged messages are hidden and sent for human review.
    • Questions you ask in your messages are counted (from transcripts) for the Spark Meter.

    2. How match suggestions work

    Suggestions use your voice intro transcript, your profile details and your preferences, such as who you want to meet and your interests. Gender is used only to respect who you want to meet; it is never used to boost anyone. Suggestions are recommendations, and you are free to ignore every one of them. Elite and VIP profiles are shown more often in suggestions.

    3. The Spark Meter

    The Spark Meter is a shared chemistry meter that grows as you talk. It is calculated by a fixed formula from six conversation signals, not predicted by AI:

    • Both speaking: whether you are both sending voice messages.
    • Balance: how evenly voice time is shared.
    • Reply rhythm: how quickly you each reply.
    • Back and forth: how often you take turns.
    • Curiosity: questions asked, counted from transcripts and text messages.
    • Returning: coming back to the conversation on different days.
    • The Spark Meter is the same for both of you. No score can be bought, boosted or adjusted by any plan or payment. Paid plans only unlock a detailed breakdown and conversation tips.

    4. Bond Preview

    Bond Preview is an optional, short and playful read on how your voices sit together, written by AI from your voice intro transcripts and profiles. It is for fun and is "not a scientific measure". It never makes relationship predictions.

    5. Voice Style, and what we do not do with your voice

    WhatsLove measures speaking pace, pauses, pitch variation and loudness variation from your voice recordings to describe your speaking style (Voice Style) and suggest people with a similar style. We do this only with your consent, which you can withdraw at any time in Settings. We do not identify people by voice, we do not store voiceprints, and we do not detect emotions or personality. Speaking-style descriptions are available on paid plans. With your consent, Voice Style is one signal used for matching, not the only signal.

    • We do not use your voice to identify you, and we do not perform face recognition or remote biometric identification.
    • Uploaded audio, photos and videos may be automatically screened for prohibited content (for example nudity or content involving minors). This is safety moderation only, not a judgement of you.

    6. AI providers we use

    • OpenAI: Whisper speech-to-text for voice intros and voice messages, and text embeddings of voice intro transcripts.
    • Google Gemini, via the Lovable AI gateway: profile summaries, match suggestion support, Bond Preview, conversation starters, photo checks and media safety screening.
    • OpenAI models via the Lovable AI gateway: safety classification of message transcripts.
    • AI-generated audio: our podcast episodes and advertising voiceovers use a synthetic narration voice and are labelled as AI-generated.

    7. Human oversight and limitations

    No irreversible decision about a person is left to a machine on its own. Account restrictions and removals triggered by automated moderation are reviewed by a human before they become permanent, and you can contest any outcome by writing to us. Automated systems can make mistakes, so please report anything that looks wrong.

    8. What we deliberately do not do

    • No social scoring or rating of users as people.
    • No emotion inference from your voice.
    • No subliminal, manipulative, or deceptive techniques designed to distort your behaviour.
    • No exploitation of age, disability, or economic vulnerability.
    • No fake profiles, AI companions or bot conversations presented as people.
    • No predictive profiling of criminality, health status, political opinions, religion, or sexual orientation.

    9. Risk classification

    Based on our own assessment, the AI systems in WhatsLove fall into the limited-risk category of the EU AI Act, which triggers transparency obligations rather than the regime applied to high-risk systems. This is our documented self-assessment, not a certification issued by a regulator or notified body.

    10. Your rights

    • Be told when you are interacting with an AI system, and when content is AI-generated.
    • Request an explanation of how an automated decision affected you.
    • Ask for human review of any automated moderation outcome.
    • Access, correct, export, or delete your voice recordings, transcripts and profile data.
    • Lodge a complaint with your national data protection authority or AI market surveillance authority.

    11. Contact

    Questions about AI in WhatsLove, or a request for human review, can be sent to info@whatslove.app. We aim to respond within 30 days, and sooner for deletion requests.

    12. Daily voice prompt

    Your daily voice prompt answer is shown on your profile to other users for 24 hours, then removed.

    13. AI conversation starters

    To suggest a conversation idea, profile details and recent conversation transcripts of both people are processed by Google Gemini via Lovable AI. Ideas are suggestions only.

    14. Product testing

    We may show different versions of features to different groups of users to improve matching, on the basis of legitimate interest (GDPR Art. 6(1)(f)). Results are analysed in aggregate. This product testing is separate from the optional, consent-based Research setting.

    15. Notifications

    Spark, voice note, daily prompt and streak notifications are limited to a combined maximum of 3 per day. Each notification type can be turned off in Settings.