#77 | We use this for AI language learning

TL;DR: AI is a strong language tutor and a terrible judge of its own mistakes. Two friends found the exact point where it breaks, and figured out what to do next. If you love languages, this edition is for you.

👋 Hello,

My friend Andrew tried Spanish twice before and quit both times. And my friend Jessy took four years of French in school and came out able to read a menu, badly.

Neither expected AI to change that. Both are three months in now, and both have gone further than expected. But for reasons that have nothing to do with talent.

Andrew picked Spanish, with German on the side. Jessy picked French, with a little Italian on weekends.

Both leaned on the same tool, ChatGPT, and built their own systems around it without knowing the other was doing the same.

They used the same tool and the same rough timeline. Weirdly, they had completely different failure points and solutions. Let’s see what they can teach us.

What actually worked

Andrew’s first real move: stop using one chat thread, use four. One for daily conversation, one for reading, one for writing corrections, and one just for a weekly vocabulary dump.

A single thread gets messy fast. So four it is.

AI (still, for gods’ sake) has no memory of what it told you yesterday, and mixing five kinds of practice into one conversation makes every session feel scattered. Splitting by function keeps each one focused.

Jessy went a different direction. Ten minutes of AI small talk every morning, no exceptions, aimed at one specific problem: the freeze-up.

She could conjugate French in her sleep and still lock up the second a real person was listening. At least, AI doesn’t get impatient. It doesn’t smirk.

That alone removed enough pressure for her to actually speak rather than just study grammar tables.

She’d heard about a story-based language course called StoryLearning* around this time, too, from a friend who swore by it, but she filed it away.

Not yet. The AI drilling still felt like enough.

Both of them also flipped the usual habit: instead of asking AI to explain a rule, they asked it to quiz them on one.

Andrew built a running list of false-friend words, the ones that look like English but mean something else entirely.

He then had the AI test him on those specifically, since those are the mistakes that survive the longest. Producing an answer sticks with you longer than reading one.

Quick takeaways so far

Split AI chats by function, not by topic
Ask AI to quiz you, not just correct you
Build a running false-friends list and drill it separately
Ten minutes daily beats one long session weekly

Where both hit a wall

Andrew’s wall showed up in week five. He asked ChatGPT to confirm a verb conjugation, and it gave him a confident, detailed, completely wrong answer.

A Spanish-speaking colleague caught it, laughing out loud, reading it back.

Not a rare failure. Learners studying other languages have reported the same pattern: an AI inventing a verb form that doesn’t exist and stating it with total certainty.

The tool that never gets tired also never doubts itself. That’s a problem when it’s wrong.

There’s another well-known version of the same wall, too.

His sentences were technically correct, and somehow nobody talks that way. (We’ll never escape textbook language, right?)

A 2026 review of ChatGPT in language learning found that it supports personalized, real-time feedback well but still lacks any coherent structure that holds the pieces together over time.

It’s a strong AI tutor for a single sentence and a weak one for a whole skill. The fix people land on is almost funny in its simplicity: go listen to a podcast, hear how humans actually talk, then come back.

Jessy’s wall looked different from the outside.

Her AI conversations felt fine. Her feedback didn’t. When she asked for pronunciation notes, she got vague encouragement instead of anything specific: close, without saying close to what.

The thing under both walls

Both learning problems trace back to the same mechanism.

Real language fluency doesn’t come from memorizing rules. It comes from understanding things slightly above your level, over and over, until your brain assembles the grammar on its own without being told.

AI is genuinely strong at generating that kind of practice on demand. What it’s structurally bad at is knowing when it’s wrong. Or noticing when a sentence doesn’t sound human or native.

That’s where both of them, separately, started looking for something with more shape. A powerful addition to the AI drilling and exercises.

Jessy tried StoryLearning* first, the intuitive language course her friend had mentioned weeks earlier. Once she realized the daily small talk had a ceiling she couldn’t drill her way past.

It builds a whole course around one long story instead of scattered exercises, with native audio and no invented grammar rules along the way.

Andrew tried the same approach for Spanish a few weeks later, after Jessy told him about it over a drink.

Neither of them dropped AI.

They just stopped asking one tool to do a job it wasn’t built for. AI still runs its morning drills and vocabulary quizzes.

The story-based course gives them the part AI can’t fake: something worth reading past the point where it gets hard, which is apparently the exact condition under which a language actually sticks.

Before you open another app

Split your AI threads by function instead of dumping everything into one conversation.

Use AI to quiz yourself, not just to explain things. Cross-check anything grammar-specific against a second source occasionally, since confident and correct aren’t the same thing with AI.

And if drilling alone has started to feel like moving icons around a screen, that’s usually the sign you need something with a story in it, not another app.

StoryLearning’s* summer sale (50% off) runs through July 22, if the story-shaped gap is the one you’re missing to learn the language you’ve always wanted to.

Cheers,

Mark
The AI Learning Guy
👋⚡😎

The AI Learning Guy newsletter 🤖 🧠💡

AI learning hacks and mega prompts delivered to your inbox.

Interesting Sources

Note: No single website has all the answers. This list serves as a starting point for those who want to explore or satisfy their curiosity about AI.
Links: Links with * are affiliate links. See disclosure below.

Affiliate disclosure: To cover the cost of my email software and the time I spend writing these emails, I sometimes link to products. Please assume these links are affiliate links. If you choose to buy through my links, a big THANK YOU – it will make it possible for me to keep doing this.