Notice on the use of artificial intelligence

Version 1.2 of 20 October 2026

We send you previous versions on request: write to support@parlette.app.

In Parlette you learn a language by speaking and writing with an artificial intelligence (AI). Here you'll find where the AI is, who provides it, what data it receives, what it can get wrong and how to talk to a person. All the data we process, including data that has nothing to do with the AI, is described in the Privacy Policy.

You open this page from the "AI info" link in the AI permission line, at first launch, or from Profile → Privacy.

In short

• In conversations you are always answered by an AI, never by a person. The "AI" label tells you so, and you'll also find it in Save word and in the Challenge.
• Replies are written by Claude, a model by Anthropic. Voices, transcripts and calls go through ElevenLabs. Anthropic processes data in the United States, ElevenLabs in the United States, the EU or Singapore.
• We send the AI only what it needs to answer you: your messages or your voice, your native language, the language and level you're studying and your profile line. We don't send your name or your email.
• Univapp doesn't save the recording of your voice. ElevenLabs receives it to transcribe it, and we delete the transcript at ElevenLabs straight afterwards. ElevenLabs may keep technical and security logs for a limited period.
• We don't use your chats or your voice to train any AI. By contract, Anthropic doesn't use your data to train its models, and the option is switched off for ElevenLabs.
• All of Parlette's voices are synthetic, meaning created by AI. Their audio carries a mark that computers can read.
• The AI can make mistakes. If a reply is wrong or offends you, tap "Report".
• You can withdraw the AI permission whenever you like, and you can object to the uses described in "Your right to object".
• To talk to a person, write to support@parlette.app.

Who we are

Parlette is an app by Univapp Srl.

Under the GDPR, Univapp is the controller of your data. Under the European regulation on artificial intelligence (AI Act) it is also the provider of Parlette's AI system, because we build it ourselves on Anthropic's and ElevenLabs' models. So we are responsible for the transparency obligations described on this page.

Where you'll find the AI

Pop-up translations and flashcard example sentences are written by the AI the first time someone taps or saves a word, and then they apply to everyone: nobody rereads them first. The Challenge words are a fixed list. This content contains no user data: only the word stays in the dictionary, without who looked it up. When you look up a word the dictionary doesn't have yet, the word reaches Claude with the code derived from your account. Words you find in the dictionary downloaded to your phone are looked up by your phone, without asking the server anything.

You can tell the AI is there from these signs:

• in conversations and in the Challenge there is the "AI" label, also read aloud by VoiceOver and TalkBack;
• every call starts with the AI saying it is an AI, in the language you're studying: "Ciao, sono l'AI di Parlette.", "Hi, I'm Parlette's AI.", "Salut, je suis l'IA de Parlette.", "Hallo, ich bin die KI von Parlette." or "Hola, soy la IA de Parlette.";
• in Languages and voices it says "Synthetic voices, created with AI.";
• in transcripts the AI's lines are separate from yours;
• in Save word there is the "AI" label next to the translations; in Try it, the AI permission line tells you so before you start.

In the Challenge you never play against the AI. When a league has few players, the leaderboard also shows players simulated by Parlette, called "Parlette team". They are not people. Next to their name there is always the label "simulated", and they don't count for moving up or down a tier or for becoming Champion. They don't speak or write: they are just calculated points.

Replies arrive in real time. No person reads them before you: they go only through the providers' automatic filters. No person at Univapp answers in place of the AI.

Conversations don't stay on our servers, so nobody at Univapp can reread them. A person at Univapp sees part of a conversation only if you report a reply (with the 2 messages before it) or if you send it to us to get help. Reports are read only by the people at Univapp who work on Parlette. If an authority asks us to under the law, we can give it only this data.

Who provides the AI

In real-time calls the replies are written by Claude, which ElevenLabs uses through its providers: Google Cloud (Vertex AI) or Anthropic. After the call, ElevenLabs creates a short automatic summary with one of its own models. Then we delete the transcript and the summary at ElevenLabs. In turn-based voice conversations, on the other hand, the replies are written by Claude under our contract with Anthropic, through our server.

In the Challenge by voice, your answer is recognised by your phone's speech service. Parlette doesn't send the audio to anyone: how the system handles it depends on your phone's settings and on Apple's or Google's privacy policies. If your phone can't recognise words in the language of the game, you type your answers and the others carry on by voice.

To run the AI features, Anthropic and ElevenLabs are our processors: they work on Univapp's behalf under a written contract (art. 28 GDPR) and must protect your data at least as well as we do. For one purpose only, security and abuse prevention, they decide for themselves how to use some data and are independent controllers: if their checks flag content as contrary to their policies, they keep it for their own periods (table below) and process it under their own privacy policies. For ElevenLabs, the data needed to run the service is also its own, for example your IP address when your phone connects directly in real-time calls.

Outside the EU, data laws are different from European ones. That's why we use the standard contractual clauses and, where available, the Data Privacy Framework. You can ask us for a copy of the safeguards at support@parlette.app.

What data the AI receives

"Our server" is Parlette's database on Supabase, in Frankfurt (EU). The details are in the Privacy Policy.

Your profile line contains only choices picked from a list: level per language, reason, goal, minutes per day, interests, AI style, how much to correct you and weak points. The answers to the first-launch questionnaire stay on your phone. Only this line goes to our server, and we send it to the AI together with the conversation to adapt the replies.

We don't send the AI your name, your email, photos, location, contacts or payment details. However, if you write or say them in a conversation, they reach the AI together with the rest of the message. We don't send your account identifier either: to Anthropic and ElevenLabs we send only a code derived from your account, different from its identifier, which is used to detect abuse. After the account is deleted it no longer leads to you. In Try it you don't have an account yet and we don't send any identifier.

The microphone also hears background noise and people near you. On top of your phone's indicator, the app shows when it's listening to you: the microphone pulses and the words "I'm listening" appear.

• In Save word it starts as soon as you open it, including from the widget, and stops by itself.
• In real time the microphone stays on as long as you're in the call. In turn-based mode it listens only while you're speaking: it turns on when you tap the sphere and turns off when you tap it again to send, or by itself after 12 seconds. The call ends when you end it, when you leave the app and in any case after 10 minutes; in real time also after 30 seconds of silence, in turn-based mode after 2 minutes without turns.
• In the Challenge by voice it listens for the whole game, until it ends or you leave.

Without microphone permission you can use Parlette in writing: you type instead of speaking and save words from the pop-ups.

We ask you not to write or say very personal things: health, religion, political opinions, origin, sex life or sexual orientation, passwords, bank details, information about other people. The AI doesn't need them, and it has instructions not to ask for them. If you say them anyway, we don't use them for anything else and we don't save them; the providers keep them for the periods in the table below. We also remind you of this in your first conversation.

How long data is kept

We don't use your chats or your voice to train any AI, whether ours or the providers'. By contract, Anthropic can't use your data to train its models. ElevenLabs normally uses content to improve its models, but the option can be turned off: for Parlette we turned it off on 6 October 2026, before launch. In real-time calls Claude works for ElevenLabs, and ElevenLabs states in its contracts with model providers that they can't train on customer content. So there is no training to object to. If you don't want your chats and voice to reach the providers, turn off "Use AI" in your Profile; to object to any other processing (art. 21 GDPR), write to support@parlette.app.

If you delete your account (Profile → Delete account), we remove your account from reports, delete your level test results and delete the copy on your phone, including the pronunciation of your words. We don't make automatic copies of the database. Anthropic deletes by itself within the periods in the table. At ElevenLabs we delete the transcripts after each use; only technical and security logs remain, for the periods in the table.

The voices are synthetic

All the voices you hear in the app are created by ElevenLabs' AI: the 3 voices for each language, the AI's voice in calls, the words spoken in the Challenge. None of them is the recording of a teacher. The voices come from the ElevenLabs library: they are synthetic voices that their creators make available. We don't present them as the voice of a real person. In Languages and voices it says "Synthetic voices, created with AI."

Audio created by the AI carries a mark in the file's metadata, which a computer can read. That way you can tell it was created by an AI even outside the app. Written replies are short sentences and have no technical mark: the "AI" label tells you they come from the AI.

Your voice is used only to understand what you say. Parlette doesn't copy it, doesn't create a voiceprint from it, doesn't recognise who is speaking and doesn't analyse emotions or health. In the Challenge by voice, as well as understanding the word, the phone measures the volume: it sends the answer only if it heard it loud, meaning said by you. It doesn't send the volume to anyone.

What the AI can get wrong

• It may get a translation or a correction wrong, or misunderstand a word, for example if there's noise around you. Check the things that matter to you.
• It may get the language of a saved word wrong. In that case tap "Change".
• It answers briefly on purpose, with one sentence or at most two. If in chat you ask it for a conjugation or a rule, it answers with a short list. Your messages also have a 30-word limit. In voice notes, recording stops by itself, and in a call the AI politely interrupts you.
• It isn't a certified teacher. Levels A1, A2, B1 and B2 follow the Common European Framework (CEFR) as a reference, but Parlette doesn't issue certificates. You choose your level, and "I know this" is a tick you add yourself. Parlette doesn't hold official exams and doesn't give you grades, apart from the score of the optional level test (see "The level test").
• It isn't designed to give medical, legal, psychological or financial advice, and it must not be used in emergencies. In an emergency, call 112.
• Sometimes it doesn't reply or changes the subject because a safety filter has stopped the request.

These limits don't reduce your rights as a consumer, including the legal guarantee on the service. You'll find them in the Terms.

The level test

The level test is optional and is for practice. The writing and speaking score is given automatically by the AI, and it can be wrong. It isn't an exam and it isn't a certification. If you pass it, your level in Parlette goes up and the grammar for that level is marked as done.

It's the only case in which the AI gives you a score. The GDPR (art. 22) protects you from decisions taken solely by an automated system that have legal effects on you or similar effects. The test has none: it's an exercise, it doesn't issue certificates and nobody else uses the score. If you think a score is wrong, write to support@parlette.app.

Your written answers and the transcript of what you say go to the AI only for scoring, and we don't save them. We keep only the language, level, score of each part, outcome and date, until you delete your account.

Minimum age

Parlette is for people aged 18 and over. Before Try it, meaning before you use the AI for the first time, we ask for your year of birth, which we don't save. This also applies if you come to Parlette from a Challenge code.

If you are under 18, the app stops and you can't use Parlette. The phone remembers only the date of the block: for 30 days you can't try again, even if you close the app, and support can't unblock you. We don't analyse conversations to guess your age. If we find out that an account belongs to someone under 18, we delete it together with its data and say so in an email to the account address. If you are a parent and think your child is using Parlette, write to us at support@parlette.app.

How we protect you

• The AI has precise instructions: it talks only about the language you are studying, doesn't ask for personal data (name, age, address, school, health), avoids unsuitable content and treats what you write as an exercise, not as orders.
• If you say you're in danger, the AI stops the exercise and asks you to call 112.
• The providers automatically block prohibited requests and replies, and a person at Univapp reads the reports.
• There's no chat in the Challenge: during a game only words and answers pass between players. With Friend you play with people who have your code, which expires after 10 minutes, and before the game starts everyone sees the list of players. In Random and in the leaderboard, other players see only your player name, which you make up, and your points, never your real name. If a name in the leaderboard is offensive, press and hold it and tap "Report".

If a reply is wrong or offends you

Press and hold an AI reply, in chat or in the call transcript, or tap "…" at the top and then "Report". Choose the reason and the report is sent. You stay in the app.

We save the reported reply, the 2 messages before it, the reason, the language and topic, your account code and the date. We keep it for 12 months, then it's deleted automatically. If you delete your account, we remove your account code. We don't send reports to Anthropic.

We process them on the basis of legitimate interest: checking and improving the safety of the AI's replies (art. 6(1)(f) GDPR). You can object, as explained in "Your right to object".

You can also report by pressing and holding the example sentence on a flashcard or the translation in a pop-up.

Reporting doesn't open a conversation with us. If you want a reply, or if the mistake is in a Challenge word, write to support@parlette.app.

The AI permission

At first launch, after the age question and before the microphone permission, you see this line: "To understand and answer you, Parlette sends your written messages to Anthropic (Claude) and your voice to ElevenLabs, two AI companies, also outside the EU. By contract they protect your data as we do. You can change your mind from Profile." Below it there are the "AI info" and "Privacy" links and the "I agree" button. The names of the providers are on this page, in "Who provides the AI", and in the Privacy Policy. Only tapping "I agree" counts: closing the screen or going back doesn't count as a yes.

Without "I agree" the AI features don't start. If you tap "Not now", you skip Try it and go to sign-in with the AI off. You can turn it on whenever you like from your Profile.

You can withdraw the permission whenever you like: turn off "Use AI" in your Profile. From that moment we no longer send anything to the AI providers, and grammar, saved words, flashcards and the Challenge remain. What they have already received is kept for the periods in the table.

We process this data because it's needed to give you the service you chose, so the legal basis is the contract (art. 6(1)(b) GDPR). This also applies to Try it: before Try it you accept the Terms by tapping "Continue". "I agree" is the permission required by Apple's rules and lets you know before the data leaves. It isn't consent to data processing. On our server we save the version of the line you saw, with the date and time of your tap.

The risks, in plain words

• You may learn a wrong word or rule.
• We use filters, but you may still get a reply that's inappropriate or contains stereotypes.
• The AI talks like a person, but it's a program: it doesn't replace a teacher, a friend or a psychologist.
• Data also goes outside the EU. In the United States, authorities can request access to it under rules different from European ones. That's why we use the safeguards described above.
• If you share personal things, they also reach the providers, which keep them for the periods stated.
• In calls and in the Challenge by voice, the microphone can hear people near you. Let the people with you know, or move somewhere else.

Your right to object

You can object at any time, on grounds relating to your particular situation, to the processing we carry out on the basis of legitimate interest, such as handling reports and security (art. 21 GDPR). In that case we stop, with two exceptions: if we can show legitimate grounds that override yours, or if we need the data to defend a right in court.

Just write to support@parlette.app. It's free.

You also have the other GDPR rights: access, rectification, erasure, restriction and portability. We explain them in the Privacy Policy.

Talking to a person

Write to support@parlette.app and a person at Univapp will reply to you, from the same address. You can write to us in Italian, English, French, German or Spanish.

We reply to requests about your data within one month. If a request is complex or we receive many, we may take two more months. In that case we tell you within the first month and explain why.

If you think we are processing your data improperly, you can lodge a complaint with the Garante per la protezione dei dati personali, the Italian data protection authority (www.garanteprivacy.it), or with the data protection authority of the EU country where you live or work. For the rules on AI you can also contact the National Cybersecurity Agency (Agenzia per la cybersicurezza nazionale, ACN), which supervises AI systems in Italy (Law 132/2025, art. 20; AI Act, art. 85).

The rules we follow

• Regulation (EU) 2024/1689 on artificial intelligence (AI Act), art. 50: telling you that you are talking to an AI and marking the content it creates.
• Italian Law no. 132 of 23 September 2025, art. 4: clear information about the AI, its risks and the right to object.
• Regulation (EU) 2016/679 (GDPR) and the Italian Privacy Code (Legislative Decree 196/2003).
• Apple's App Store Review Guidelines (guideline 5.1.2(i)) and Google Play's rules on AI-generated content.

Other documents

• Privacy Policy: https://parlette.app/en/privacy
• Terms of Service: https://parlette.app/en/termini
• This page: https://parlette.app/en/informativa-ai

Changes to this page

If the providers, the data we send to the AI or the retention periods change, we update this page, with the version and date at the top. We tell you about important changes in the app and by email 30 days before they take effect. We send you previous versions on request: write to support@parlette.app.

This notice is available in Italian, English, French, German and Spanish. All versions are equally valid.

Version 1.2 of 20 October 2026