The theory of comprehensible input — that we acquire language by understanding messages slightly above our current level — has been kicking around for forty years. Stephen Krashen called it i+1: input where i is what you know, and the +1 is the small stretch that lets you pick something new up without drowning. Simple in principle. Genuinely hard in practice, because "slightly above your level" is a moving target that shifts every couple of weeks. A July 2026 paper in Education Sciences is one of the first to test what happens when AI generates that input at a calibrated CEFR level — and the results are worth paying attention to.
What comprehensible input actually means
Krashen's claim is narrower than the phrase suggests. Not just any input you can mostly follow — specifically input that stretches you by a small amount. Read something two levels above you and you're guessing; you spend your session decoding grammar instead of absorbing it. Read something two levels below and nothing new lodges in memory. The sweet spot is a narrow band, and it moves as you move.
That's what makes the idea hard to apply on your own. You can't easily judge whether a text is at your i+1 — most learners either coast on easy material or drown in native-level articles and blame themselves for not understanding enough.
The problem: most reading you find online is at the wrong level
Type "Spanish news for beginners" into a search engine. You'll get:
- Actual A2 material written for classrooms, which reads flat and childlike
- B2 news articles labeled "for beginners" because someone glossed the hard words
- C1 native content with a translation crutch stapled to the side
None of these are i+1 for the average A2-to-B1 learner. The classroom material sits beneath them; the "glossed" articles are above them; the C1 stuff turns reading into a translation exercise. This is the gap CEFR-leveled reading was designed to close. When a text is genuinely calibrated to A2 or B1 — vocabulary distribution, sentence length, tense range, grammatical structures — you get input you can process, with a manageable percentage of new words per page.
Comprehensible input isn't a hack. It's the least glamorous, most consistent finding in second-language acquisition.
What the 2026 research actually shows
The Education Sciences paper compared two reading pathways using AI-generated texts calibrated to learners' CEFR levels:
- Extensive reading — lots of text, slightly below or at level, minimal lookup
- Intensive reading — shorter texts at or just above level, with active work on unknown vocabulary
Both approaches showed measurable gains. The interesting finding is that calibration mattered more than pathway. Learners who received texts pitched at their actual level — whether extensive or intensive — outperformed learners working with uncalibrated material. When the level was right, both modes worked. When the level was wrong, neither did.
That's a useful result for anyone building a reading habit: the pathway you pick matters less than getting the level honestly right.
Extensive vs intensive reading — and why you probably need both
Extensive reading is what most people picture when they imagine "reading in a foreign language." You read a lot of text you can mostly follow, without stopping much. Volume matters more than depth. This is where reading fluency comes from — the ability to process the language in real time instead of decoding it word by word.
Intensive reading is the opposite. Shorter passages. Slightly harder. You stop, look up, notice, take a word or a construction and pull it into your vocabulary. This is where new items enter the language you actually produce.
Most learners do one of these poorly and skip the other entirely. Sentence-level exercises in most apps skip extensive reading almost by design — a single sentence isn't reading. On the other end, learners who decide to "just read Kafka" do a version of intensive reading so brutal that nothing sticks and the book gets abandoned by page ten.
The study's practical implication: pick both, and calibrate both. Twenty minutes of extensive reading at your level for fluency; ten minutes of intensive reading a step above for growth. Do that most days and you'll outpace anyone doing an hour of drills.
How to apply this without an app
If you're doing this on your own, three habits go a long way:
- Know your CEFR level honestly. Take a short self-assessment or a graded reader placement test. Guessing high is the most common mistake, and it's the one that quietly kills reading habits.
- Choose one graded reader series and stay with it. Penguin Readers, Cambridge Readers, Olly Richards' short story collections — these exist precisely because uncalibrated native material is too hard for almost everyone below C1.
- Track new words in a way that reappears. Not a static list you'll never reopen. Something that resurfaces each word in a new sentence a few days later. That's what spaced repetition is for.
The manual version of this workflow is completely viable. It's just slow to set up, and easy to abandon when the graded reader runs out and you have to decide what to read next.
Where Lex fits
We built Lex's reading around exactly this idea. Texts are generated at your chosen CEFR level from A1 through C2, so the +1 stretch is already sized. You can request a reading text on any topic — a short article about Basque cuisine, a fictional dialogue in a Berlin café — and get something calibrated rather than approximated. Words you save while reading reappear in later texts, exercises, and writing prompts, so the same items surface in different contexts on a spacing you don't have to manage yourself.
The point isn't that Lex is the only way to do calibrated reading. Graded readers have worked for decades. The point is that when the level is right, reading does the work that no amount of drilling can. That's what the 2026 research is starting to show, and it's what we've spent our time building for.
The takeaway
What's new in 2026 isn't the theory. Comprehensible input has been the consensus in second-language acquisition for a long time. What's new is that we finally have tools that can generate calibrated input on demand at the right level — and early empirical evidence that they measurably help.
Pick a level. Read at it. Read a little above it. Do that most days. That's the whole method.