"Learn Japanese from YouTube" is advice that is technically true and practically useless, because the difficulty of Japanese short-form video runs from "a person slowly naming the vegetables they bought" to "two comedians doing rapid-fire wordplay", and nothing on the thumbnail tells you which one you are about to open.
This is the version of the advice with the missing parts filled in.
Do the kana first. All of it. Before anything else.
There is no route into Japanese that goes around hiragana and katakana, and every week spent postponing them is a week of learning material you will have to relearn.
Both syllabaries together are 92 characters plus a handful of modifications. Two to three weeks at twenty minutes a day gets you reading them slowly; a couple of months of seeing them in context gets you reading them at speed. It is the single highest-return fortnight in the language.
The reason it cannot wait: romanised Japanese misrepresents the sound system. It hides that Japanese is timed in morae rather than syllables, which is why Tokyo is four beats and not two, and why a long vowel is a different word from a short one. Learners who spend six months on romaji develop a pronunciation and a listening model that then has to be dismantled.
Kanji is a different question and it does not need to be answered yet. You can get a long way with kana plus readings shown above the words — which is what furigana is for, and what a transcript with readings above each word does automatically.
Why Japanese listening is specifically hard
Three properties, none of which is "Japanese people speak fast" in any simple sense.
There are no spaces, and there are no word boundaries in the sound either. Japanese writing runs continuously, and so does the speech. In a language with spaces you at least get the segmentation for free in writing; in Japanese you have to find the joins in both channels. A transcript that splits the words for you is doing real work.
It is counted in morae, not words. Our reference rates for Japanese are about 3.2 characters per second for a deliberate, clear delivery and about 5.6 for ordinary conversation. That ratio — a bit under 1.8 — is roughly what every language shows between careful and casual speech, and it is why textbook audio feels manageable and a real clip does not.
Pitch accent exists and is not taught. Japanese distinguishes words by pitch pattern, and most courses ignore it entirely. You do not need to study it formally, but you do need enough listening that your ear absorbs it, because it is part of how native listeners segment the stream.
Particles do the work that word order does in English. は, が, を, に, で and the rest carry the grammatical relationships, they are short, unstressed and easy to miss at speed — and missing one changes who did what to whom. Early listening practice is very largely particle practice, whether or not you think of it that way.
What to actually watch, in order
Phase one: anything with a strong visual anchor. Cooking, unboxing, get-ready-with-me, building things, cleaning, pets. The requirement is that the speaker is doing the thing they are describing, so the meaning is carried by the picture and the words attach to it. You are not translating; you are matching sound to an event you can see. This works from week one, before you know any grammar.
Phase two: one person talking to camera about an everyday subject. Routines, opinions about food, explaining their job, a tour of their flat. Single speaker, no overlap, no cuts mid-sentence. This is where you start needing a transcript, because the meaning is now in the words.
Phase three: two people in conversation. This is a real step up and it is the one that matters, because overlapping speech, interruption and reaction are most of how people actually talk. Interview clips are the gentle end of it.
Phase four: comedy and anything fast. Save it. Jokes require both speed and cultural reference and they are the last thing to come.
Anime is not on this list deliberately. It is fine as a reward and it is a poor training set: the register is stylised, the speech is often either unnaturally slow or unnaturally fast, and the vocabulary is skewed towards things nobody says.
The three traps
Watching with English subtitles on and calling it study. With English on the screen you read English. Your eyes go to the thing you can read instantly and your ears switch off; this is well attested and it happens whether or not you intend it. If you want the full argument it is in subtitles: which language, and when. The short version is that English subtitles are for enjoying something, not for learning from it.
Watching a lot, once each. Getting through fifty clips once is far less useful than getting through ten clips five times. The second and third pass is where segmentation happens, and it is the part that feels like it is not working.
Looking up every unknown word. Stopping on every word turns a forty-second clip into a twenty-minute grammar exercise and destroys the flow that listening practice depends on. Watch it through, note the two or three words that actually blocked you, look those up, watch it again.
A session that works
Twenty minutes, one clip of about thirty seconds, and a transcript.
- Watch it once, no transcript, no subtitles. Register how much you got. You will get less than you expect and that is the baseline, not a verdict.
- Watch it again. You will get noticeably more, purely from knowing what is coming.
- Read the transcript without the audio. Now the words are separated, the readings are above the kanji, and you can see what was actually said. This is usually the moment where you discover that the incomprehensible bit was three words you know.
- Look up the two or three words that were genuinely new. Not all of them. The two or three that mattered.
- Watch it twice more with the transcript. You are now hearing the boundaries.
- Watch it once more without. This is the rep that counts.
Then do the same clip again tomorrow before starting a new one. Six passes across two days on one short clip will do more for your listening than an hour of new material.
Where the words should come from
One thing to be deliberate about: do not let your vocabulary be only the words you chose to save.
The words that stand out are the unusual ones, so a self-assembled deck fills up with interesting rare vocabulary while the words carrying most of the language sit half-known. In Japanese this is especially costly, because the high-frequency items are particles, auxiliaries and the handful of verbs that do everything — exactly the words that do not feel like words worth saving, and exactly the words you are missing at speed.
Learn the commonest few hundred deliberately, in frequency order, and let the interesting words arrive on their own. The reasoning is in how many words you need to understand a language.
What a year of this looks like
Japanese is in the hardest FSI band for English speakers, roughly three times the hours of Spanish. Realistically, at twenty minutes a day with kana done early:
- Month 1–2: kana fluent, a few hundred words, following visually-anchored clips about a third of the time.
- Month 3–6: grammar basics, particles starting to land, following single-speaker everyday clips with a transcript.
- Month 6–12: single-speaker clips without a transcript most of the time, conversation clips with one.
- Year 2: conversation without a transcript, and the long climb through the part where progress stops being visible week to week.
That is not fast, and no honest version of it is. What makes it survivable is that the material is stuff you would have watched anyway — which is the only reason anybody does twenty minutes a day for two years. The full arithmetic is in how long it really takes.