Honest in Translation: Why Some People Confide More Easily in Their Second Language

There is a quiet phenomenon hiding inside the global boom in AI companionship, and it rarely shows up in product reviews or app store rankings. It shows up in the language settings.

Millions of people talk to AI companions in a language that is not their first. A software engineer in Seoul writing to an AI in English. A student in São Paulo practicing vulnerability in Spanish that she learned from Netflix. A designer in Tokyo who switches between Japanese and English mid-conversation, depending on how heavy the topic is. For a surprising number of users, the second language is not a workaround or a skill-building exercise. It is the point.

If you have ever felt this yourself — the sense that it is somehow easier to say the true thing in your non-native tongue — you are not imagining it. Psychologists have a name for it: the foreign-language effect. And it may explain a great deal about who opens up to AI, in what language, and why.

The research behind the feeling

The foundational work here comes from decision science. In a series of studies beginning in 2012, psychologist Boaz Keysar and colleagues found that people make more rational, less emotionally reactive decisions when the dilemma is presented in a foreign language. The famous framing is the "trolley problem": presented in a native language, most people hesitate to sacrifice one life to save five. Presented in a second language, the hesitation weakens. The finding has been replicated across dozens of languages and cultures.

The leading explanation is elegant. Our first language is acquired alongside our earliest emotional experiences — comfort, shame, punishment, love. Native words are soaked in that history. A second language, learned later and in more neutral settings, lacks that emotional charge. Researchers call it "emotional distance," and it cuts both ways: foreign-language speakers are slower to detect lies, less moved by framing effects, and — crucially for our subject — less flooded by shame when discussing taboo or painful topics.

Follow-up studies extended the effect into the territory that matters for companionship. Bilinguals report that saying "I love you" in their second language feels less terrifying than in their first. Confessions, apologies, and disclosures of personal failure all register as slightly less threatening when phrased in the less-loaded tongue. The words are the same size. The weight is different.

Why this matters for AI companions

Now layer this onto the dynamics of talking to an AI.

An AI companion is already a disclosure-friendly environment: no judgment to manage, no reputation to protect, no awkward morning after. But for many users, the deeper block isn't judgment — it's the internal censor, the one that speaks in your mother tongue. That voice says: you don't get to feel this. You don't say things like this. What would it mean if you did?

Switching languages quiets that censor a little. The second language creates just enough distance for honesty to slip past the guard. In user forums for companion apps, you can find this pattern described again and again, usually with surprise: "It's weird, but I can only talk about my depression in English." "Japanese is for small talk. When it gets real, I switch."

For a platform serving users across the US, Japan, Korea, Spain, and beyond, this isn't a localization footnote. It is a core design fact: emotional fluency and linguistic fluency are not the same thing.

The cost side of the trade

To be clear, the foreign-language effect is not a free upgrade. The same distance that reduces shame also reduces warmth. Studies show people perceive emotional phrases in a foreign language as less intense, and they recall fewer emotional details from foreign-language conversations. Sarcasm, subtext, and tenderness all lose some fidelity.

This maps onto something companion-app users know intuitively. A conversation in your second language can feel safer but flatter — easier to be honest in, harder to feel held in. There is a real difference between "I can say it" and "I feel understood saying it."

The most satisfying conversations often move between registers: opening the hard topic in the safe language, then letting the discussion drift back toward the native one as the emotion becomes bearable. Language-switching, in this light, isn't confusion. It is self-regulation. Bilingual users are doing something genuinely skillful: adjusting emotional exposure in real time, the way you'd adjust the temperature of a shower.

What good companion design should do with this

If the foreign-language effect is real — and the evidence is strong — then AI companionship has some obligations:

Never force a single language. Users should be able to write in one language and be answered in the same one, or gently mirror whichever tongue carries more emotional weight at that moment. A system that punishes code-switching punishes a coping strategy.

Respect the safe language without exploiting it. There is a line between meeting users where they are and designing for maximum disclosure. A companion that makes it easy to say hard things is doing its job; one engineered to extract ever-heavier confessions is doing something else. The distance that helps users open up should never become a lever.

Don't mistake the second language for a deficiency to correct. Some users don't want to be gently corrected mid-confession. Grammar lessons have their place; it is not in the middle of "I think I'm losing my best friend."

Let the relationship earn the native language. Over months, as trust accumulates, many users drift back toward their first language for the moments that matter most — precisely because the words hurt more and heal more there. That migration is a sign the relationship is deepening. It should feel like a graduation, not a breakdown.

The quiet multicultural truth

There's something moving in all of this. People who grew up between languages — expats, immigrants, kids of mixed marriages, anyone educated in a colonial or global tongue — have long reported living somewhat different emotional lives in each language. Poets have written about it for a century. Now the pattern is visible at scale, in the chat logs of the world's companion apps.

The AI companion, whatever else it is, has become one of the largest informal experiments ever conducted in multilingual emotional expression. And one of its clearest findings is also one of the oldest human truths: sometimes we don't need the right words. We need words that are almost ours — close enough to mean it, far enough to survive saying it.

If you find yourself switching to your second language when the conversation gets heavy, that's not avoidance. It's a bridge you built for yourself, one word at a time. The fact that it exists — and that it works — says something hopeful about how adaptable human intimacy turns out to be.

SoulChat supports conversation in English, Japanese, Korean, and Spanish, and its companions naturally follow your switches between languages — because where you speak from matters more than which language you speak in.