The recognition trap
Most language apps are built around multiple choice, matching, and tapping words into order. These exercises test recognition: can you identify the right answer when you see it? Speaking is production: can you retrieve a word from nothing, assemble a sentence, and say it before the other person loses patience? Recognition and production are stored differently in memory, and practising one improves the other only slightly.
That is why learners with long streaks and thousands of flashcards freeze in conversation. They have not practised the skill they want. Any app that will build speaking has to make you speak, in full sentences, repeatedly, with some way of finding out whether what you said was right.
The four types of Japanese apps
| Type | Examples | Good for | Weak for speaking because |
|---|---|---|---|
| Gamified courses | Duolingo, LingoDeer, Busuu | Habit building, basic vocabulary, gentle grammar | Mostly tap and match; little free speech; feedback is right or wrong, not why |
| Spaced-repetition flashcards | Anki, WaniKani, Renshuu | Vocabulary and kanji retention at scale | Zero production under time pressure; no conversation |
| Human tutor platforms | italki, Preply | Real conversation, cultural nuance, accountability | Cost per hour, scheduling, and nervousness limit frequency for many learners |
| AI conversation tutors | Unihongo | Daily spoken practice, roleplay, pronunciation scoring, instant corrections | Less cultural judgment than a good human teacher; best combined with input |
None of these types is bad. The mistake is expecting a flashcard app or a gamified course to produce conversation. They will not, because they never ask you to hold one.
Seven features to look for if speaking is the goal
- Voice input required. If you can complete the lesson silently, it is not a speaking tool.
- Open-ended prompts. Reading a scripted line aloud is useful for pronunciation but not for fluency. You need questions with no fixed answer.
- Corrections with explanations. Knowing a sentence was wrong is not enough; you need the fixed version and the reason.
- Level adaptation. A tool that speaks the same way to beginners and advanced learners will bore one and drown the other.
- Realistic scenarios. Ordering, asking directions, self-introductions, doctor visits, and meetings are what you will face. Practise those.
- A review loop. Session history, saved corrections, and something to revisit tomorrow turn practice into progress.
- Low friction. If starting a session takes more than a minute, you will skip it on busy days, and speaking only works daily.
How Unihongo approaches speaking
Unihongo was built around the daily conversation problem. You open the app, choose a topic and level, and talk with an AI Sensei by voice. The tutor listens, replies naturally, and adapts to how you speak. Roleplay missions give conversations an objective, such as checking into a hotel or explaining a symptom to a pharmacist, so you practise situations rather than abstract prompts. A pronunciation lab scores individual lines, and after each session a report shows what you said, the corrected version, and what to work on. Flashcards and a kana reference support the speaking work rather than replacing it. It is free to start, with no card required.
A recommended stack by goal
- Travel in three months: one AI conversation app for daily ten-minute sessions on travel topics, plus a small phrase deck in a flashcard app. Skip gamified courses; the time is better spent speaking.
- Daily conversation in a year: daily AI conversation, a flashcard app for the words that come up in your own sessions, and a human tutor every week or two for cultural judgment and motivation.
- Work-ready Japanese: daily AI practice on business topics and keigo, a kanji app for reading email and documents, extensive listening to natural-speed Japanese, and a human tutor or class for feedback on register.
- JLPT plus speaking: a structured grammar and kanji resource for the test, and AI conversation to make sure the grammar you study actually comes out of your mouth.
How to evaluate an app in one week
Give any app seven days and keep a simple log: minutes used, full sentences spoken aloud, corrections received, and whether you could say one new thing at the end of the week that you could not say at the start. An app that scores well on all four is worth your time. An app that scores well only on minutes used is entertainment, which is fine, but it is not going to get you talking.