What If Comprehensible Input Included Output? (CI+O Explained)

I'm not in academia, and this isn't meant to be a theoretical debate about Krashen's work or the exact definition of comprehensible input. I'd like to share some thoughts I've had while learning with CI, explain a gap I kept running into, and outline an approach I call CI+O — Comprehensible Input + Output.

What I mean by comprehensible input

Within the community, “comprehensible input” is often used quite loosely. It can mean anything from grasping the vague overall meaning of a message to understanding perhaps 90% of what was said. That range is wide — and it matters, because not all of it is the same kind of learning.

Krashen's i+1 principle says that at each interaction we should add only one new unknown element (the +1). Taken seriously, comprehensible input is input where you understand everything except that single new piece — a word, a grammar pattern, a new conjugation of a familiar verb, and so on. This is what I call strict CI.

Why does that distinction matter? Because vaguely understanding a message is still valuable: you can follow the conversation, stay engaged, and avoid feeling completely lost. But it doesn't let your brain do the focused work that drives acquisition. When there is only one unknown element, you can absorb it while placing it in context within the current message — because every other piece is already known, you can direct your full attention to that new one.

Think of what happens in your native language when someone uses a word you haven't heard before. You understand the rest of the sentence without effort. You don't spend an iota of energy decoding the familiar parts — only on that new word. The same mechanism applies when learning a foreign language: strict CI recreates the conditions where one new element gets your undivided attention.

When the condition isn't met, input becomes mostly comprehensible rather than truly comprehensible — and my intuition is that unless CI is strict, it can't fully realize its potential.

For a deeper dive on strict vs approximate CI, see Strict Comprehensible Input Explained.

Why most CI content is only approximate

All the CI services I've personally used provide what I'd call approximate CI, for two structural reasons:

• Their content is static — not generated dynamically based on what the learner actually knows.
• ...because, they have no way of knowing exactly what the learner knows, in the first place. You would essentially need a continuously updated model of the learner's vocabulary and grammar knowledge.

The best you can do with static content is create a series of stories or videos that start from zero and gradually build on previously introduced vocabulary. Conceptually, that works. In practice, learners are almost never exposed to the target language through a single channel.

They'll talk to native speakers, watch videos, read books, use textbooks, study flashcards, and more. As they learn words outside the CI system, their knowledge gradually diverges from the assumptions made by the static content — making it less and less adapted to the individual learner.

Why static comprehensible input can't fully adapt to you goes into this in more detail.

CI + Output: the missing piece

This isn't a criticism of CI. After all, it's called Comprehensible Input. But I think it's unfortunate that the approach traditionally focuses only on input.

Personally, I find that words stick much more easily once I've actively used them, rather than simply being exposed to them. My hypothesis: CI is already much better than random input, but combining CI (passive) with output (active) could make it substantially more effective.

What would CI+O look like?

The most straightforward implementation is a conversation — through text or speech. You receive a strict CI message. You reply. You receive another strict CI message. And so on.

Every response reuses only vocabulary you've already encountered while occasionally introducing a single new element, allowing vocabulary to expand naturally at the i+1 pace. Grammar isn't explicitly taught; it's gradually acquired through repeated exposure in conversation — also one of Krashen's principles.

Why hasn't CI+O existed before?

At least as far as I'm aware, because it simply wasn't practical. To satisfy what I believe are the two conditions that unlock CI's full potential — strict CI, and CI combined with output — you need:

1. A continually updated model of what the learner knows, and
2. The ability to generate new content on the fly based on that knowledge.

Until generative AI, that simply wasn't possible at scale.

AI language learning for beginners explains why unconstrained chatbots miss this — and what works better.

One way to test CI+O: your flashcards

This led me to build Talk To Your Flashcards — a proof of concept that's free during public alpha. The goal is to test whether CI+O is actually as useful as I think it might be.

I personally used it to learn Mandarin Chinese and felt CI+O has a lot of potential. The app tracks the words you know through your flashcard deck and generates conversations that stay within that vocabulary, introducing new words one at a time — strict CI, with output built in.

It supports Spanish and Mandarin Chinese, and like traditional CI, I think it's best suited to beginners — especially complete beginners. If you use Anki or another SRS app, see Anki and speaking practice for how this fits into a flashcard workflow.

Other CI+O possibilities

A conversation app is just one implementation. I think there are many others:

• An RPG where every NPC speaks to you using strict CI — quests, dialogue trees, and exploration all bounded to your vocabulary.
• A collaborative story-writing game where the AI writes a few sentences, you continue the story, then the AI responds again — always following strict CI principles.
• Listening practice where every audio clip is generated for your known words, not a fixed curriculum.

I'm genuinely excited by the possibilities AI opens here. If this resonates with you, I'd love to keep discussing and experimenting — join the Discord community I created, dedicated to exploring CI+O and building new language-learning experiments together.