Comprehensible input is language you can mostly understand. It is reading or listening where you follow the message even though some words are new, and the new parts make sense from context. The term comes from linguist Stephen Krashen, whose Input Hypothesis claims this is how we acquire language: by understanding messages slightly above our current level, not by memorizing rules and hoping they turn into speech.
The harder questions are practical: how much do you need to understand, how much volume does acquisition take, and where do you find material at your level that you actually want to read? This guide covers all three.
Why input works
Krashen's formula is i+1: "i" is your current level, "+1" is the next step just beyond it. When input sits there, you understand the message and absorb the new pieces through context. In his words, "if input is understood, and there is enough of it, the necessary grammar is automatically provided".
Two words in that sentence carry the whole method. "Understood": input you cannot follow is just noise. And "enough": a few minutes a day does not approach the volume acquisition needs.
Krashen later added a refinement that matters for self-taught learners: input should be compelling, so interesting you forget it is in another language. Mildly interesting material rarely survives a busy week; compelling material carries you through the months acquisition requires.
What "mostly understand" means in numbers
Researchers have measured this. Hu and Nation (2000) tested learners on the same story at different levels of known-word coverage: most needed about 98 percent of words known, roughly one unknown per fifty, to read comfortably without help. Laufer and Ravenhorst-Kalovski (2010) set 95 percent as a minimal threshold with some assistance, and 98 percent for independent reading. Listening is more forgiving: van Zeeland and Schmitt (2013) found about 95 percent coverage appears adequate by ear.
Hold these numbers loosely. A 2023 replication by Kremmel and colleagues could not fully reproduce Hu and Nation's results, so 95 to 98 percent is a rule of thumb, not a law. The direction is solid, though: nearly everything known, unknowns sparse enough to infer.
This is also why native content fails beginners: novels and newspapers need roughly 8,000 to 9,000 word families for 98 percent coverage (Nation, 2006). With 1,500 words you are nowhere close, and "just read books and watch shows" turns into dictionary archaeology.
A worked example
Say you are an A2 Spanish learner and you read:
Ana abre el paraguas porque empieza a llover.
You know abre (opens), porque (because), empieza a (starts to), and llover (to rain). The only unknown is paraguas. But someone is opening something because it starts to rain, so your brain supplies "umbrella" before you could reach for a dictionary. That is acquisition: a new word arrived inside a message you understood.
One unknown in an eight-word sentence is well below 98 percent, but the coverage thresholds describe running text, not single sentences; in a full story at the right level this moment arrives every few sentences. Now give that sentence five unknowns: no context is left to infer from, nothing sticks, and reading turns into decoding. That gap is the difference between acquiring and studying.
The volume problem
Acquisition also needs far more input than most plans assume. Nation (2014) modeled the reading volume needed to meet the most frequent 9,000 word families often enough to learn them; at a typical reading speed of around 150 words per minute, his figures work out to roughly 1,200 hours, an hour a day for three years.
That leaves two practical needs: a steady supply of material at your level that you actually want to read, and a way to handle unknown words without breaking flow.
Where to find input at your level
The non-app options are worth knowing, each with a tradeoff.
- Graded readers are professionally leveled and reliable, but the catalog picks your topics, and outside the biggest languages the selection is thin.
- Learner podcasts offer slow, clear speech, but levels are self-declared and hard to match to your vocabulary.
- Simplified news sites cover real news in plain language, but exist for few languages, usually at one level band.
All three are useful, and all share one gap: you rarely get your topic and your level in the same text.
That combination is what LingoUnify's generator is for. Pick a topic from the preset categories (daily life, travel, work and school, stories, factual topics) or write a free prompt about anything, from your hometown's history to how espresso machines work. Choose a difficulty from A0, a pre-beginner starter level, up to C2 (the targeting is an estimate, not a certified level) and a length of roughly 350, 650, or 1,000+ words, and the text streams in. Generation respects regional varieties, so a learner of Dominican Spanish gets Dominican Spanish (see our guide to choosing a Spanish variety). Because you chose both topic and level, the result is self-selected and compelling by construction, the two qualities the research keeps pointing at. Texts worth keeping can be saved to your Library.
The pen tool: fast lookups without leaving the text
The coverage research has a practical loophole: 95 percent was adequate "with some assistance". Assistance is the lever. If a lookup costs a dictionary tab and a search, flow dies and reading becomes study; if it costs one click, texts slightly above your comfort level become readable.
That is the pen tool's job. With the pen on, any word you click opens a small popover: save it as a learning word (stored with its surrounding context), mark it learned, hear it spoken, or pronounce it into the mic for feedback. Optional automatic lookups (translation, definition, grammar, etymology) use the surrounding words to disambiguate, so banco resolves to "bench" in a park sentence and "bank" in a money one; they draw on a daily AI allowance that varies by plan. Dragging across a phrase translates it or reads it aloud.
None of this replaces inference; it backs it up. You still read for meaning and guess from context, reaching for the pen when a word blocks the meaning or is worth keeping.
The word overlay: your saved words in everything you read
A saved word means little if you never meet it again. The overlay highlights your saved words, yellow for learning and green for learned, across everything you read on the platform: generated texts, your Library, AI and friend chat, imported articles in Browse, deck stories, and external websites through the browser extension. With the pen off, clicking a highlighted word shows your saved translation. Every text becomes a fresh context for words you chose, and saved words can later be turned into flashcards, a separate step you take, not something automatic.
One limit worth knowing: the embedded YouTube player in Browse does not support the pen or overlay. LingoUnify does not process or store video audio or captions, so its word tools cannot run inside the player.
Listening input
The same coverage logic applies by ear, and you cannot slow down a native radio host, so build listening on material you already understand. Re-listening to a text you have read is the easiest win: Library texts play aloud with word-by-word highlighting at 0.5x to 1.5x speed, turning a text you have already understood into listening practice. At the harder end, live radio can help you test your ear against fast, natural speech. If the vocabulary is far beyond you, though, radio is still exposure, not yet acquisition.
Input from conversation
Reading is not the only channel. The AI chat replies in your target language at your chosen level, A0 to C2. With feedback off it simply converses, no corrections, and can weave one or two of your current learning words into its replies. When you are stuck, a hint button suggests replies with translations.
Friend chat adds real people. Incoming messages can be translated as-is or rewritten at any level from A0 to C2, shown alongside the original by default, with text-to-speech playback. The pen and overlay work in both chats, and messages from a friend who writes above your level become input you can actually use.
Honest limits and common mistakes
Comprehensible input is necessary. The research does not say it is sufficient.
- Output still matters. Merrill Swain's studies of Canadian French immersion students (Swain, 1985) found that years of comprehensible input still left gaps in speaking accuracy. Producing language forces you to notice what input lets you coast over.
- Some explicit study helps. A meta-analysis of 49 studies by Norris and Ortega (2000) found explicit instruction produces larger gains than implicit instruction. Input does most of the work; deliberate review and a little grammar speed it up.
- Do not look up everything. Extensive reading research says to read for general meaning, not 100 percent comprehension (Day and Bamford, 2002). Use the pen for blockers and repeats, and let the rest go.
What to do next
- Start one level below where your ego wants to. Move up when nothing is unknown, down when you are decoding.
- Generate or import a text about something you would happily read in your native language.
- Read with the pen on. Save the words that block meaning or keep recurring, and skip the rest.
- Re-listen to texts you have already read, and treat harder audio as ear training until you can follow the main message.
- Aim for 20 to 30 minutes daily rather than occasional long sessions. The volume research is blunt: this works in months and years, not days.
- Add conversation a few times a week, AI or human, and read your partner's messages as carefully as a text.
The method is not complicated. Understand messages, lots of them, about things you care about. Everything else is logistics.