Ask anyone who's tried to learn a language what was missing, and you'll usually get the same answer: someone to talk to. Not an app with multiple-choice questions — an actual patient conversation partner, available whenever you have ten minutes, who won't sigh when you stumble over the same tense for the fifth time. For most learners, most of the time, that person has never existed.
Voice mode changes that. A capable AI assistant will now hold a spoken conversation with you, in dozens of languages, at any level, for as long as you like. It's the single biggest thing to happen to solo language learning in a generation — if you use it well. Here's a field guide.
Set it up as a partner, not a lecturer
Open your assistant's voice mode and brief it before you start, in text or speech:
Let's have a conversation in [language]. I'm at roughly [level]. Speak a little slowly, keep your sentences simple, and gently correct my mistakes — but note the corrections at the end so we don't break the flow of the conversation.
That last part matters. If it stops to correct every error mid-sentence, you'll never build any fluency; if it never corrects you, you'll cement your mistakes. Errors noted at the end of an exchange is the sweet spot — you keep talking, and you still learn.
Actually talk — that's the whole point
The temptation is to keep drilling grammar. Resist it. The thing you can now do that you never could before is speak, so speak. A few formats that work:
- Roleplay real scenarios. Order a coffee. Do a mock job interview. Make small talk with a stranger. Argue about football. The more it resembles something you'll actually do, the better it sticks.
- Narrate your day. Tell it what you did today and what you'll do tomorrow. Simple, endlessly repeatable, and it surfaces exactly the vocabulary you personally need.
- Let it interview you. Ask it to keep putting questions to you on a topic. That forces you to produce language rather than merely recognise it — the hard, valuable direction.
Use its corrections properly
After a chunk of conversation, ask: "What did I get wrong, and how should I have said it?" Then — crucially — ask why. "Why that tense and not this one?" turns a correction into a rule you'll remember. The AI is a genuinely excellent explainer of grammar in context, which is where grammar actually lives.
Mix voice with the rest
Voice is the breakthrough, but it isn't the whole meal. Use text chat for the things speech is bad at: reading a short article together and unpacking it, drilling verb conjugations, quizzing vocabulary. A good rhythm is spoken conversation to build fluency and confidence, plus text work to firm up the structure underneath.
Know the limits
Two honest caveats. First, an AI's pronunciation and accent are good but not flawless, and can vary by language — treat native audio (films, music, podcasts, real speakers) as your source of truth for how the language actually sounds. Second, a model is a superb practice partner, not a substitute for contact with the living language and its culture. Use it to get fluent enough to be brave, then take that out into real conversations and real media.
Consistency beats intensity
The oldest rule in language learning is still the truest, and voice mode makes it easy to honour: ten minutes of talking every day will take you further than a two-hour session once a week. The whole point of a tutor in your pocket is that it's there — so use it little and often, and it compounds.
This is exactly what the Modern Languages node in the library is built around — voice-mode-first, with prompts that set your AI up as a conversation partner from the first message. Pick a language, and start talking today.