Most people who start a language app stop somewhere around A2. The streak survives, the language does not. AI has not changed that, and the parts of it that do help are not the chatbot on the marketing page.
Three things move a learner: spaced repetition that adapts to what you forget, pronunciation feedback on single sounds, and reading and listening material pitched just above your level. This post covers which apps do those three well, a routine that fits in twenty minutes a day, and the list of things no app will teach you.
The three AI features worth paying for
Spaced repetition that adapts to you. The app tracks each word or phrase and brings it back just before you would forget it. Words you keep missing return tomorrow, words you know return in a month. This is the one feature with decades of memory research behind it, and Anki has done it since 2006 without calling it AI. If an app does not have it, the vocabulary will not stick.
Pronunciation scoring on single sounds. The app compares your recording with native speech and tells you the R is too soft or the stress sits on the wrong syllable. It is reliable on individual sounds and common words. It is unreliable on intonation and rhythm, so a green tick on a whole sentence means less than it looks.
Reading and listening at your level. Second-language research calls it i+1: material slightly above what you can already handle. Apps that pick articles and podcasts for you on that basis save the hour you would spend hunting for them. When I find a text myself, I paste it into the readability checker first. The score tells me whether it is a stretch or a wall.
Two features get more screen time in the ads and deserve less. Grammar correction on your writing helps in languages with cases and particles (German, Japanese, Finnish), where memorising patterns is not enough, but it explains the rule and does not make you apply it. Conversation roleplay (ordering food, asking directions) is a warm-up, not a conversation. The AI is patient, predictable and never interrupts. People are none of those.
The apps, and who each one is for
Duolingo is where most people start and where most people stop. The gamification keeps the streak alive, the lessons are short, and the list of languages is the longest. The paid Max tier adds AI explanations of your mistakes, roleplay, and a video-call practice partner. It gets you through basic vocabulary and grammar. It rarely gets anyone past intermediate, because the lessons never get long enough.
Anki is not a language app and has no AI to speak of. It is a flashcard scheduler, free on desktop and Android and paid on iOS, and it does spaced repetition better than anything else on this list. Since version 23.10 it ships FSRS, a scheduler that fits itself to your own forgetting curve, next to the older SM-2 algorithm. You make the cards or download a shared deck, and you show up every day. That last part is where it loses people.
Pimsleur is audio only, thirty minutes a day, with the spacing built into the scripts. Good for a usable accent and the first few hundred phrases. The vocabulary stays small.
Busuu pairs an AI study plan with corrections from native speakers on your written exercises. That human feedback loop is the reason to pick it. Fewer languages than Duolingo.
Lingvist does one thing: vocabulary in sentence context, spaced, at a level the model adjusts as you go. If your gap is words, it is the most direct route. Short language list.
ChatGPT or Claude as a tutor works once you are past beginner and know what you need. Ask it to correct a paragraph and explain each change, to drill the subjunctive, to roleplay a job interview in the target language. It has no curriculum, so it will not tell you what to learn next.
I have not put months into every app here. The comparison is about what each one does mechanically, which a free week is enough to check for yourself.
**Duolingo** is where most people start and where most people stop.
Twenty minutes a day, and what goes in them
No single app covers all four skills, so stack them. This is the routine I would set up, and it is short because the short version gets done.
Every day, 20 minutes
- 10 minutes of flashcard review, in Anki or the app's own queue. Do this first. Reviews pile up when you skip a day.
- 5 minutes of listening at your level: a podcast episode, a short video.
- 5 minutes out loud: read a paragraph, or shadow a recording. The text to speech tool reads any text you paste with the voices installed on your machine, so you can shadow your own paragraph instead of one the app chose.
Three times a week, 30 to 60 minutes
- One grammar topic with exercises, from a textbook or the app.
- One piece of writing (a journal entry, a message), corrected by the AI tutor or a person.
- One conversation with a person: a language exchange partner, or a tutor on a marketplace like iTalki or Preply.
Once a week
- Something long entirely in the language: an episode, a chapter, a whole podcast.
- Five minutes on whether the words you are learning are the words you need. Tourist vocabulary is no use in a work meeting.
Twenty minutes a day beats three hours on Sunday. Every spaced repetition scheduler assumes daily exposure, and a week's gap turns tomorrow's review into a wall of two hundred cards.
For scale: about 3,000 words covers everyday conversation, and 8,000 to 10,000 covers professional work. Keep your list in a text file and run it through the word counter once a month. That number going up is the only progress bar that means anything.
What no app will teach you
- When to be formal. An app teaches the words for please and thank you. It does not teach when the Dutch u (the formal you) sounds respectful and when it sounds like you are keeping someone at arm's length. That comes from being in the room.
- Sarcasm, humour and register. Knowing how to say something is not knowing when. The grammar of a request is in every app. The social cost of getting it wrong is not.
- The local accent. Pronunciation models are trained on the standard variety. Learn Spanish for Buenos Aires from a Castilian-trained model and you will be corrected on sounds nobody there uses.
- Showing up. The hard part of a language is month four, not day one. AI makes the session better. It does not make you open the app.
- The stretch from B2 to C1. Getting from intermediate to advanced takes novels, films without subtitles, arguments and jokes. The models are weakest exactly there.
- A real conversation. Interruptions, background noise, someone with a cold and a regional accent talking fast. Only people prepare you for people.
- **When to be formal.** An app teaches the words for please and thank you.
FAQ
Can an AI app make me fluent on its own?
No. It makes the study hours count for more: retention goes up, the grammar drills hit your actual gaps, pronunciation gets corrected early. The conversations that turn study into fluency still happen with people.
How many hours does it take?
The US Foreign Service Institute puts languages close to English (Spanish, French, Dutch) at roughly 600 to 750 class hours, and Mandarin, Arabic, Japanese and Korean at about 2,200. Better tooling might shave a tenth off that. The rest is hours.
Is Duolingo enough?
For the first three to six months, for vocabulary and basic grammar, yes. After that it becomes a habit rather than a course. Add one human conversation a week and native content, and keep Duolingo for the streak if the streak is what keeps you going.
What is the best stack that costs nothing?
Duolingo for structure, Anki with a shared deck for vocabulary, and the free tier of ChatGPT or Claude for corrections and grammar explanations. Add a language exchange partner, which costs an hour of your own language in return.
AI Trip Planners: What They Get Right, the Prompt That Works, and What to Verify Before You Pay
A chatbot writes a decent day-by-day itinerary in a minute. It also invents opening hours and prices. How I prompt for a plan shaped like the trip, which planning apps add something a chat cannot, and the seven things to check before booking.
AI Bedtime Story Generators: What to Ask For, and What to Check Before You Read It Aloud
A generator writes a story with the child's name in it in thirty seconds. Whether it is a bedtime story, short, calm and ending asleep, depends on the prompt. The tools I could verify, the prompt that gets a usable story, word counts per age, and the one-minute check before you read it out.
AI Podcast Tools: Where the Script, the Voice and the Editing Come From
Which AI tools write the script, read it aloud, clean the audio and produce the show notes, what each one costs, where the seams show, and the one disclosure to make before you upload.
AI Speech Recognition Accuracy in 2026: What the Percentages Hide
What word error rate measures and what it misses, why vendor accuracy numbers come from clean audio, where transcription still fails in 2026 (names, numbers, accents, two people talking), which services to test, and how to raise accuracy in your own application.