Comprehensible input is language you understand without decoding it. You read or hear a sentence and the meaning arrives; you do not assemble it word by word. That is the whole definition, and it is the mechanism underneath almost everything on this site.
The term is Stephen Krashen's. He set it out in Principles and Practice in Second Language Acquisition (1982) and developed it in The Input Hypothesis: Issues and Implications (1985). His claim was that people acquire a language by understanding messages in it, and that conscious study of rules produces something different and less useful: knowledge about the language rather than command of it.
Krashen's stronger claims have been argued over for forty years and some of them have not aged well. The core observation has aged very well indeed.
An example
Take a beginner in Spanish who already knows ayer (yesterday), mi hermana (my sister), compró (bought), pequeño (small) and lo llamó (called it). They see a photo of a small dog and hear:
Ayer mi hermana compró un perro pequeño y lo llamó Toby.
They have never met perro. They do not need to. Everything around it is understood, the photo does the rest, and the word arrives with its meaning attached. That is comprehensible input, and it is also the clearest illustration of i+1: the "i" is what you already have, the "+1" is the one new thing, and the context carries you across the gap.
Play the same sentence to someone on their first day of Spanish and it is not comprehensible input. It is a pleasant noise. Nothing in it can be inferred because nothing around it is known.
Krashen's five hypotheses
Comprehensible input is one part of a larger theory, usually called the Monitor Model. It has five parts:
- The acquisition-learning hypothesis. Acquiring a language (subconsciously, the way children do) is a different process from learning about it (consciously, from rules). Only acquisition produces fluent use.
- The monitor hypothesis. Consciously learned rules act only as an editor, a "monitor", on what acquisition produces, and only when you have time, know the rule and are thinking about form. Which is why you can recite the subjunctive triggers and still not use the subjunctive in conversation.
- The natural order hypothesis. Grammatical features are acquired in a broadly predictable order, largely regardless of the order they are taught in. This drew on the morpheme-order studies of the 1970s.
- The input hypothesis. We acquire by understanding input a little beyond our current level (i+1). Speaking emerges on its own once enough has been acquired; it is a result of acquisition, not a cause of it.
- The affective filter hypothesis. Anxiety, low motivation and low self-confidence raise a mental "filter" that stops input from being acquired even when it is understood.
The input hypothesis is the one the whole comprehensible input movement rests on, and it is the one this article is mostly about.
What the evidence supports, and what it does not
This is where most articles on the subject either cheerlead or dismiss. The evidence sits in between.
Well supported: that large amounts of meaningful, understood input are necessary for acquisition. The extensive-reading literature, from the Fiji "book flood" study of Elley and Mangubhai (1983) onwards, consistently finds that learners given lots of readable, interesting material improve across the board, including in areas nobody taught them. Nobody serious in second language research now argues you can skip input.
Contested: that input is sufficient. The most famous problem came from Canadian French immersion programmes. Merrill Swain found that immersion pupils, after years of rich comprehensible input, understood French at close to native level but still made persistent grammatical errors when they spoke and wrote. Her output hypothesis (1985) argued that having to produce language forces you to notice the gap between what you want to say and what you can say, which input alone does not.
Also contested: that input just needs to be comprehensible, however it gets that way. Michael Long's interaction hypothesis argued that the most useful input is the kind you negotiate: asking for repetition, getting rephrased, checking you understood. Richard Schmidt's noticing hypothesis (1990) argued that input only becomes intake when you consciously notice a feature of it, which gives explicit instruction a real, if modest, role: it tells you what to look for.
Weakest: the claim that consciously learned knowledge can never turn into acquired knowledge, and some of the hypotheses' testability. Barry McLaughlin's critique (1987) was that terms like "acquisition" and "comprehensible" were defined loosely enough that the theory could explain almost any result, which is a weakness in a theory, not a strength.
The practical upshot is simple enough. Input should take most of your hours. Speaking, being corrected and a little explicit grammar make the input work harder. None of that rescues input you cannot understand.
The number that actually matters
Strip away the theory and comprehensible input reduces to a coverage figure: what proportion of the words in front of you do you already know?
The research converges on a narrow band:
- 95% is the threshold Batia Laufer identified for reasonable comprehension of a text.
- 98% is the figure Hu and Nation (2000) put on comfortable, unassisted comprehension of fiction: reading without help and without it feeling like work.
- Below 95%, comprehension collapses fast. You spend your attention decoding, and decoding crowds out the acquisition you are reading for in the first place.
The distance between 95% and 98% is larger than it looks. At 95% coverage you meet an unknown word roughly every twenty words, which is about one per line. At 98% it is one in fifty. The first is study. The second is reading.
This is the number the whole method hangs on, and it has an uncomfortable implication that input purism tends to skate past.
The bottleneck nobody puts on the poster
At zero words, nothing is comprehensible.
That sounds trivial. It is not, because it means the input method is at its weakest exactly where beginners are told to rely on it hardest. The first stretch of pure input is not 95% coverage, it is closer to 5%, and 5% coverage is not language, it is weather. Specially made beginner material helps, with pictures, gesture and a speaker who slows right down, but there is a limit to how much of a language can be carried by mime.
This is where adults quit. Not because they lack an ear, but because they were told the input would click and it did not click on the timescale anyone promised. The 1,000 to 1,500 hour estimates that circulate in input-only communities are real, and a good share of those early hours are spent on material the learner cannot yet understand.
There is a faster way through, and it is not a shortcut. It is arithmetic.
Engineering comprehensibility: the frequency lever
Word frequency in every human language follows a steep curve. A small number of words do most of the work, and the drop-off is severe.
Paul Nation's coverage work on English puts the most frequent 1,000 word families at roughly 80% or more of ordinary spoken English, and the same Zipfian shape holds across the major languages. The practical consequence is one of the most useful facts in language learning:
Learning the thousand most frequent words of a language takes you to roughly 80% coverage. Learning a thousand words chosen at random takes you almost nowhere.
Eighty per cent is not yet the 95% threshold. But it is the difference between weather and signal, and it is the point where context starts doing the work for you. At 80% coverage an unknown word sits in a sentence you otherwise understand, which is exactly the i+1 condition. Below that, an unknown word sits in fog.
Coverage also explains why picking up vocabulary from input alone is slow. Research on incidental learning from reading suggests a word usually needs several encounters, often put at ten or more, before it sticks, and the words just outside your level are, by definition, the ones you meet least often. Deliberate study front-loads the frequent words so the input can spend its encounters on the rarer ones.
So the position this site takes is not that comprehensible input is wrong. It is that comprehensibility is a property you engineer rather than a stage you wait for, and the fastest lever is deliberate, frequency-ordered vocabulary work. See why the first 1,000 words matter for the full arithmetic.
The loop
In practice the method is a loop, and the order matters:
- Drill a frequency tier. Top 100, then top 500, then core 1,000, and on. Use active recall, not rereading, and let spaced repetition decide when each word comes back.
- Read and listen inside that tier until it is genuinely easy. This is where the acquisition happens. Graded readers are written to exactly this constraint: every content word sits at or below a frequency ceiling, so a learner who has drilled the top 100 can read a top-100 story end to end with no lookups. That is comprehensible input by construction rather than by luck.
- Move up a tier and repeat.
Each half feeds the other. The drill is what makes the next tier of input comprehensible; the input is what turns drilled words into reflexive recognition and teaches you the grammar you were never explicitly taught. Skip the drill and the input is fog for a very long time. Skip the input and you end up with a large passive vocabulary you cannot use at conversational speed, which is the classic outcome of a good school languages education.
The Kilo Lingo curriculum is built as this loop, in Spanish, French and Mandarin. Each chapter drills a band of the frequency list and then hands you a graded story written to that exact band, and a build check fails if a story strays above it. The methodology page explains how.
Where to get it, by stage
Absolute beginner (0 to ~300 words). Almost nothing authentic is usable yet. Use material engineered for the stage: graded readers at the lowest tier (Spanish stories, French stories, Mandarin stories), beginner comprehensible input video where the speaker draws and gestures, and the word drill to build the floor you are standing on.
Early intermediate (~300 to 1,500 words). The range opens up fast. This is where dedicated CI libraries earn their reputation:
- Spanish: Dreaming Spanish is the reference video library, built explicitly on Krashen's model. Español con Juan is the standard podcast step up. See also the Spanish podcast round-up and the Spanish reading list by CEFR.
- French: InnerFrench is the closest French equivalent and is pitched beautifully at B1. Français Authentique suits slightly earlier. More in the French podcast round-up.
- Mandarin: Du Chinese and Mandarin Companion handle graded reading, which matters more in Mandarin than anywhere else because the character barrier keeps authentic text out of reach for far longer. See the Mandarin podcast round-up.
Upper intermediate and beyond (2,000+ words). Authentic material becomes viable, and the job changes from finding comprehensible input to finding input you actually enjoy: television with target-language subtitles, novels, native podcasts. Be warned that the bar is higher than it feels. Nation (2006) estimated that 98% coverage of written English takes something like 8,000 to 9,000 word families, and of spoken English 6,000 to 7,000. The coverage threshold is doing the work for you now; your remaining task is volume.
Five ways people get this wrong
Treating background noise as input. Audio you cannot follow is not comprehensible input, it is audio. Exposure alone does not build vocabulary or grammar. If you cannot follow the gist, the material is not doing the job you think it is.
Choosing material that is far too hard. The instinct is that harder material pulls you up faster. It does the opposite. A text at 80% coverage teaches less per hour than a text at 96%, because at 80% you are decoding rather than reading. Pick the level where you can move forward without stopping.
Choosing material that is too easy for too long. The mirror mistake. Once a tier is comfortable, the new-material rate has dropped and you are maintaining rather than acquiring. Move up.
Refusing to speak for a thousand hours. Krashen described a silent period and argued it should not be forced. Some online communities have hardened that into a rule against speaking at all until some large hour count. The evidence does not support anything that strict, and the immersion research suggests the cost: you understand well and speak badly, and never find out which is which.
Believing that vocabulary study is cheating. It is not cheating and it is not opposed to input. It is the fastest available method for reaching the coverage level at which input works at all.
Where that leaves you
Krashen was right about the mechanism. Language is acquired by understanding things, not by memorising rules about them, and hours of genuinely comprehensible input are worth more than the same hours of grammar drill.
Where the popular version of the argument goes wrong is in treating comprehensibility as something that arrives on its own if you are patient enough, and input as the only thing that counts. Comprehensibility arrives much sooner if you go and get it, and the most efficient route is the least glamorous one: learn the most frequent words first, deliberately, then spend as much time as you can reading and listening at the level those words unlock, and open your mouth earlier than the purists would like.
That is the entire method. The rest of this site is an implementation of it.
Related reading
- Why the first 1,000 words matter: the coverage arithmetic in full.
- What are graded readers?: the format that makes input comprehensible by construction.
- How polyglots learn languages: input, output and where the two camps disagree.
- How long does it take to learn a language?: what the hour estimates mean in practice.