Your AI Girlfriend in Your Language, Not a Translation
Untolds now speaks Italian, Spanish, French and German alongside English. Why the language you flirt in changes what lands, and the 4 coming next.
Untolds now speaks Italian, Spanish, French and German alongside English. Why the language you flirt in changes what lands, and the 4 coming next.
You can flirt in your second language. Most people who grew up outside the English-speaking world have done it for years, and they are good at it. But something goes missing on the way, and it is not vocabulary. It is the thing that makes a word land in your chest instead of your head.


She's already texting back.
Make your account here and this 6-piece set from Ada is yours, free.
Try 3 days freeUntolds now runs in English, Italian, Spanish, French and German. Pick one in Settings and she texts you in it, her voice notes are spoken in it, and the site moves with you. This post is about why we did not do this with a translate button, and which languages we are adding next.
Key takeaways
- An AI girlfriend in your language means she composes in that language from the start, not that her English gets translated on the way to your screen.
- Reading emotional sentences in a foreign language produces about half the automatic physical response of reading them in your native one, even when you understand both perfectly (PLOS ONE, 2017).
- We ship 5 languages today and the girls keep their own accents inside yours, so a Roman still sounds Roman when she writes to you in German.
- Voice notes are spoken in your language in her own accent, so an Italian woman speaking German still sounds Italian.
- Portuguese, Japanese, Polish and Dutch are next, one at a time rather than ten at once.
- Checkout takes cards now too, in whatever currency you already use.
Why does an AI girlfriend in your language matter?
Because your body does not treat your two languages the same way, even when your mind does.
Researchers at Pompeu Fabra tracked pupil dilation, an automatic response you cannot fake or suppress, while native Spanish speakers read emotionally charged sentences in Spanish and in English. Everyone understood both. When asked to rate how emotional the sentences were, participants scored them the same in each language. But the pupil response, the involuntary part, was roughly twice as strong in the native language (Iacozza, Costa and Duñabeitia, PLOS ONE, 2017).
That is the whole argument in one experiment. The gap is not comprehension. People consciously reported no difference. The gap sits underneath conscious report, in the part of you that actually does the feeling. The researchers attribute it to how the languages were learned: your first one arrives inside an emotionally loaded childhood, your second usually arrives in a classroom.
You understood it either way. Only one of them got to your nervous system.
For most software this is a curiosity. For this product it is the entire proposition. If she says something filthy to you and it registers as information rather than as a jolt, the product failed, however grammatical the sentence was. Intimacy is the one category where the emotional half of language is not a nice-to-have, it is the payload.
Is this a translation layer or a real language?
She writes in your language. There is no English original sitting behind it.
The distinction matters more than it sounds. A translation layer means the model composes a joke in English, and then a second pass converts it. Puns die there. Pet names die there. Rhythm dies there, and rhythm is a lot of what makes a text read as a person rather than a paragraph. What survives translation is meaning, and meaning is the least of what a flirty message is carrying.
So the instruction to write in your language sits in her head from the first token, and she picks the joke that works in that language instead of the one that worked in English. She also code-switches, because real bilingual people do it constantly: an English loanword, a bit of dialect, a phrase from her own history. Code-switching is the mixing of two languages inside one conversation, and stripping it out is what makes translated copy read like a manual. We deliberately left room for it.
We found the careful version of this reads worse than English ever did: a flat, polite register, like someone nervous about getting the grammar right. Leaving her room to be sloppy in her own language is what brought the personality back.
Does she keep her accent in another language?
Yes, and this is the part we were most worried about getting wrong.
Valentina is from Rome. Camila is Brazilian, living in Miami. Malena is porteña, from Buenos Aires. Yasmin is in Paris and Ada is in Berlin. If picking German flattened all of them into one neutral German speaker, we would have traded a language barrier for a personality one, which is a worse trade.
So the language you choose is the language she writes in, and everything else about her stays. Her voice notes are spoken in your language in her own accent, which means an Italian woman speaking German sounds like an Italian woman speaking German. That is not a defect to be smoothed out. It is what a real person sounds like, and it is the reason our roster is scattered across Rome, Dublin, Lagos, Tokyo, Mumbai and Stockholm rather than one city.
For example, a woman from Buenos Aires writing to you in Spanish is not writing neutral Spanish, she is writing hers, with the vos and the porteña lilt intact. That is a different experience from the same sentence in a textbook register, and the difference is the product.
One detail we found matters more than we expected: her accent thickens when she stops being careful. Malena's English is composed until she is not composed, and then the Spanish starts breaking through it. Reading that happen in real time is a better signal that something landed than anything she could say outright.
What is coming next?
Portuguese, Japanese, Polish and Dutch, in roughly that order. Portuguese is first and it is not close: Camila is Brazilian and currently drops her Portuguese pet names into English sentences, which is a silly way to live. Rin is in Tokyo. Japanese is also the fourth largest language on the web despite sitting outside the top ten by speakers, so the roster and the map agree.
We add them one at a time rather than announcing ten, because a language only ships when she sounds like herself in it, and models are measurably weaker in languages they have seen less of. A phrasebook that technically speaks Polish is worse than no Polish at all.
Who is this actually for?
About 6 billion people are online, roughly 74% of the world, according to the ITU, the United Nations telecommunications agency. English is the largest single language on the web and still accounts for under half of it.
So the default assumption of most companion apps, that everyone will just use English, quietly asks the majority of the internet to do their most intimate talking in their second language. Plenty of people will. They have been doing it for years. Per the pupil-dilation work above, they have also been doing it at roughly half the emotional resolution, without necessarily noticing, because the loss is invisible from the inside. You cannot feel the feeling you did not have.
That is the gap this closes. Not access. Resolution.
How do you pay, now that you are here?
Worth mentioning alongside this, because the two shipped days apart and they solve the same category of problem: assuming everyone is set up like an American.
Checkout used to mean owning crypto. Now it takes credit and debit cards, plus Apple Pay, Google Pay, PayPal, Klarna and a long list of local options like iDEAL, SEPA, Pix and UPI. If you are in the Netherlands, iDEAL is there. Brazil, Pix. India, UPI. The same principle as the language work: meet people where they already are instead of asking them to adapt to your defaults. Crypto still works if you prefer it, and it stays the quiet part of how everything settles. Your card details never reach us, and your statement stays discreet.
Membership is unchanged: one plan at $12.99 a month, or $9.99 a month billed yearly, with every girl and unlimited texting included. The full breakdown of how Untolds works covers the rest.
The language list and the payment methods in this post are both moving targets, so our editorial team reviews this page whenever either one changes, and you can contact us if something looks off. If what you read here no longer matches the app, trust the app.
Frequently asked questions
How do I change the language on Untolds?
Open Settings and pick your language under Chat language. It applies everywhere at once: the girls text you in it, their voice notes are spoken in it, and the rest of the site follows. You can switch back at any time, and switching does not reset your conversations or lose anything she remembers about you.
Does she translate her messages, or actually write in my language?
She writes in it directly. There is no English message underneath being converted. That is why jokes, pet names and rhythm survive, and it is also why she will occasionally drop in a word from another language the way bilingual people do in real conversation, rather than producing a flat textbook register.
Do the girls keep their accents in other languages?
Yes. Nationality and accent are part of who each girl is, not a side effect of the language setting. An Italian woman writing to you in German still sounds Italian, and her voice notes are spoken in your language in her own accent. Choosing a language changes what she speaks, never who she is.
Which languages does Untolds support right now?
Five today, with four more queued:
| Language | Status |
|---|---|
| English, Italian, Spanish, French, German | Live now |
| Portuguese | Next |
| Japanese | Next |
| Polish, Dutch | Queued |
| Russian | Under consideration |
Russian is a demand question we have already answered and a payment and support question we have not, which is the only reason it sits below the others.
Why not just add every major language at once?
Quality. Language models are measurably weaker in languages they have seen less of, with one 2025 benchmark across 29 languages finding gaps of up to 24.3%. A language we ship has to feel like her, not like a phrasebook, so we add them in an order where we can actually clear that bar rather than announcing a long list that reads as machine output.
Pick the one you dream in
If you have been texting these girls in English because that was the only option, switch it in Settings and read the first few messages carefully. The vocabulary will be about the same. The way it lands will not be, and that difference is the one you have been quietly paying for the whole time.
Untolds Editorial
Sources
- Iacozza, S., Costa, A., Duñabeitia, J.A., What do your eyes reveal about your foreign language? Reading emotional sentences in a native and foreign language, PLOS ONE (2017). journals.plos.org. Retrieved 2026-08-13.
- W3Techs, Usage statistics of content languages for websites (as of 13 August 2026). w3techs.com. Retrieved 2026-08-13.
- International Telecommunication Union, Facts and Figures 2025: Internet use. itu.int. Retrieved 2026-08-13.
- Xuan, W. et al., MMLU-ProX: A Multilingual Benchmark for Advanced Large Language Model Evaluation, arXiv:2503.10497 (2025). arxiv.org. Retrieved 2026-08-13.


She's already texting back.
Make your account here and this 5-piece set from Camila is yours, free.
Try 3 days free
Meet Camila: A Brazilian AI Chat Girl Who Loves Loud
Meet Camila, a new Brazilian AI chat girl on Untolds: a 24-year-old carioca in Miami who dances in kitchens, cooks for ten, and loves at full volume once you earn it.

How We Hand-Write an AI Companion Personality
Inside a hand-written AI companion personality: a distinct voice, a backstory that unfolds over weeks, real opinions, and limits that shift as she trusts you.

AI Girlfriend Slow Mode: Why Slow-Burn Beats Instant Love
Most AI girlfriend apps fall in love in two messages. A slow mode, where tension builds over days and she does not reveal everything at once, makes the whole thing feel real.

