AI graded readers that match your vocabulary

Graded readers have been a staple of language learning for decades. The idea is simple: you read stories written at your level, picking up new words from context instead of a dictionary. It works. Research on extensive reading (Day & Bamford, 1998) consistently shows that learners who read a lot at the right level acquire vocabulary faster and develop better intuition for grammar.

The problem is finding enough material. If you're learning Spanish or French, you have options. If you're learning something less widely studied, you'll run out fast. And even for popular languages, most graded readers are written for a generic learner, not for you specifically. They don't know which words you've already mastered and which ones you're still working on.

Langadoo takes a different approach. Instead of publishing a fixed library of graded readers, we generate them on demand using your actual vocabulary data. Every story is built around words you already know, with a handful of new ones woven in. You never run out, and you never waste time reading material that's too easy or too hard.

For the big European languages, the graded reader market is decent. Oxford, Cambridge, and a few independent publishers put out leveled fiction and non-fiction for Spanish, French, German, and Mandarin learners. You can usually find something at your level, at least for the first year or two.

For less common languages, the situation is bleak. When I started learning Serbian, I found maybe three graded readers for beginners. I finished them in two weeks and hit a wall. Finnish was worse. Vietnamese had essentially nothing. You're left trying to read native material that's way above your level, which is discouraging and inefficient. Or you give up on reading altogether and stick to flashcards, which is its own kind of dead end. I've written more about this in our journey creating Langadoo.

Even for popular languages, there's a second problem: the readers aren't personalized. A "B1 Spanish reader" assumes a generic B1 learner. But your B1 might look completely different from someone else's. Maybe you've spent three months watching cooking videos and know every kitchen word in Spanish, but you've never studied politics or sports vocabulary. A generic graded reader can't account for that.

When you use Langadoo, you build up a vocabulary profile over time. Every word you save from a video, every word you review with spaced repetition, every word you click on in a story gets tracked. The system knows not just which words you've seen, but how well you know them based on your review history.

When you request a graded reader, the AI uses this profile. It writes a story that's composed primarily of words you've already demonstrated knowledge of, then introduces a small number of new words in context. This follows what linguist Stephen Krashen called the i+1 principle (Krashen, 1982): input should be just slightly above your current level. Not so far above that you're lost, but enough to stretch you. The result is a story you can actually read with maybe 90-95% comprehension, where the remaining 5-10% are new words you can figure out from context or click to translate.

Each generated story includes comprehensible input features built in. You can click any word for an instant translation. If a new word seems worth learning, you save it with one tap and it enters your SRS review queue. The story itself becomes a source of sentence mining material, just like a YouTube video would be.

Paul Nation's research (2001) on vocabulary acquisition keeps coming back to one finding: words learned through meaningful encounters in context transfer to productive use far more reliably than words learned from bilingual lists. When you read a word in a story, you absorb its collocations (what words tend to appear near it), its register (formal vs. casual), and its grammatical behavior (what prepositions follow it, how it conjugates in practice). A flashcard that says "courir = to run" gives you none of that.

Graded readers give you repeated exposure to target words across a sustained narrative. You might encounter the same word five or six times in a single story, each time in a slightly different sentence. That repetition in varied contexts is what moves a word from "I recognize this" to "I can use this." Pair that with spaced repetition review after reading and you get both depth of encoding and long-term retention.

You can absolutely ask ChatGPT to write you a story in your target language. I've done it myself. The output is usually grammatically correct and readable. But ChatGPT doesn't know what words you've learned. It can't check your review history or adjust difficulty based on your actual progress. You have to manually specify your level in the prompt, and even then it's guessing. The stories tend to either be too simple (because you asked for "beginner") or randomly difficult (because it threw in vocabulary you've never seen).

Langadoo's story generation is connected to your learning data. It doesn't guess your level from a label. It looks at the specific words you know, how recently you reviewed them, and how confident your recall is. The stories that come out are calibrated to your actual vocabulary, not to an abstract proficiency band. As you learn more words, the stories naturally get more complex. You don't have to tell the system you've progressed. It already knows.

Reading alone isn't enough. Listening alone isn't enough. Flashcards alone definitely aren't enough. What works is combining them so each method reinforces the others.

On Langadoo, a typical loop looks like this: watch a YouTube video, save words you don't know, those words feed into your SRS queue, then the AI generates a graded reader that reinforces what you just learned while introducing a few more. Words from the story also go into your review queue. Each piece feeds the next. You can read more about how sentence mining ties these methods together.

Day and Bamford (1998) argued that extensive reading works best when it's paired with other forms of input and review. They were writing before YouTube and SRS apps existed, but the principle holds. Graded readers give you sustained, comfortable reading practice. Videos give you listening and pronunciation exposure. SRS makes sure the vocabulary sticks. The three together cover more ground than any single method can on its own.

Sign up for free and generate your first AI graded reader. No credit card needed.

References: Krashen, S. (1982). Principles and Practice in Second Language Acquisition. Pergamon Press. Nation, I.S.P. (2001). Learning Vocabulary in Another Language. Cambridge University Press. Day, R. & Bamford, J. (1998). Extensive Reading in the Second Language Classroom. Cambridge University Press.