Search for a Japanese writing practice app and the store will hand you two completely different products without telling you they are different.
One kind teaches you to draw characters. You trace 書 with your finger, the app grades your stroke order and proportion, and you get better at handwriting. The other kind takes a paragraph you wrote and tells you what is wrong with it. Both are filed under writing practice. They share almost nothing, they solve problems of wildly different value to an adult learner, and the more useful of the two is the one that is worse marketed.
I am a native speaker living in Tokyo, and this is the twelfth article in a series where I test Japanese learning products against what the language actually demands. This one has a clearer conclusion than most, because there is a specific rule that every Japanese child is taught by national curriculum at the age of eight or nine, that governs every piece of Japanese writing an adult produces, and that not one product in this comparison checks.
I will get to it. First, the split.
Two different things wearing the same name
Character writing is Skritter, Kakimashou, Write It, Kakitai. You practise producing kana and kanji by hand, with stroke order and stroke direction graded.
Composition correction is LangCorrect, Journaly, LangJournal, Nihongo Diary, Fix My Japanese, HiNative. You write sentences of your own and somebody or something tells you how they land.
If you are an adult learning Japanese for work, travel, reading or conversation, the second category is worth far more of your time than the first, and I want to be blunt about why.
You will type almost everything you ever write in Japanese. Typing runs through an IME, which means you produce the reading in kana and then choose the correct kanji out of a list. That is a recognition task. You need to know that 帰る and 変える are different and which one you meant, but you do not need to be able to draw either of them from memory. Japanese adults know this from the inside, because our own handwriting ability has quietly degraded for exactly the same reason — the phenomenon even has a name here, 書けない漢字が増えた, and it is a completely ordinary thing for an educated native to admit.
This is the same argument I made about handwriting in my guide to learning hiragana and katakana, where I said thirty to sixty minutes of handwriting total is enough for kana. For kanji the number is larger than zero — handwriting does help the shapes stick, and there are still forms to fill in at a ward office — but it is nowhere near a $14.99 per month habit for most people.
So this article is mostly about the second category. I will cover the handwriting tools honestly and briefly, and then spend the real time where the real problem is.
Why writing deserves more of your attention than you are giving it
Here is the argument for writing that nobody makes, and it is the reason I wanted to write this article.
Everything I have found broken across this series has the same shape: Japanese forces you to encode your relationship to the listener into the sentence, and no product makes you choose. I found it with casual and polite speech, where every product teaches both registers and none asks which one this listener gets. I found it with keigo, where the framework the products teach is two decades out of date against the official one.
The reason those choices are so hard to practise in conversation is time. Somebody is looking at you. You have about a second.
In writing you have unlimited time. You can stop, think about who is reading this, decide what your relationship to them is, and choose accordingly. Writing is therefore the one place where the register decisions of this entire series can be practised deliberately, slowly, with the pressure off, until they become the reflex you need when somebody is standing in front of you.
Writing is not a separate skill from speaking well. It is the training ground for the hardest part of speaking well. That is the case for taking it seriously, and no writing app makes it, because no writing app is looking at register.
The rule: 文体の統一
Now the thing I promised.
Japanese has two writing styles. 敬体, which ends sentences in です and ます. And 常体, which ends them in だ and である. Every English-language resource will tell you this much, usually as a note about politeness levels.
What they do not tell you is the rule that governs them:
Within a single piece of writing, you pick one and you stay in it. You do not mix them.
This is not a stylistic preference or an advanced nicety. It is taught in Japanese elementary schools as a matter of national curriculum. In the 学習指導要領 issued in 2017, the guidance for 中学年 — third and fourth grade, so children of eight and nine — includes writing while paying attention to the difference between 敬体 and 常体. In the previous 2008 version it sat explicitly under 書くこと, the writing strand. It was reorganised in 2017 into the knowledge and skills strand under 言葉遣い, but it did not go away. Japanese children are drilled on this for years.
And it is the single most common thing that marks a piece of learner Japanese as learner Japanese, in my experience reading it. Not particle errors. Not kanji. The text starts 今日は友達と映画を見ました, drifts into 面白かった, comes back with とても良い映画でした, and every one of those sentences is grammatically perfect. The document is still wrong, and a Japanese reader feels it immediately as a kind of tonal seasickness without necessarily being able to say why to a non-native.
There is a second layer that matters just as much: which style you choose is determined by what the document is, not by how polite you feel. A diary, an essay, a report, a news article, a novel — 常体. A letter, an email to somebody you do not know well, a speech, a school composition addressed to a reader — 敬体. Choosing 敬体 for an academic essay is not politer. It is wrong, in the way that writing a lab report in the second person is wrong in English.
So a learner keeping a Japanese diary to practise has already made a decision they do not know they made. Diaries are 常体 territory. If you are writing yours in です・ます because it feels safer, you are practising the wrong register several hundred times.
Why the AI correctors cannot catch this
This is the part I find genuinely interesting, and it is structural rather than a matter of the models being weak.
文体 consistency is a property of the document. AI correction tools operate on the sentence. You paste in a paragraph, the tool walks through it, and for each sentence it asks whether that sentence is correct. Every sentence can pass while the document fails, because the error does not live inside any sentence — it lives in the relationship between sentence three and sentence seven.
And then there is a second, more concrete obstacle. Look at the limits these tools ship with. Nihongo Diary allows 300 characters per entry on its free plan and 500 on the $9 per month plan. 500 Japanese characters is roughly one solid paragraph. Style consistency is a problem that appears at the length of a page. These tools are capped below the length at which the problem becomes visible. You cannot practise document-level style in a box that does not hold a document.
I want to be fair here: the AI tools are genuinely good at what they do target. Particles, conjugation, word choice, and the specific business of making a clumsy sentence sound native are exactly the things they handle well, instantly, at three in the morning, for free or nearly free. That is real value and I use similar tools myself for English. I am saying that there is a category of error they are architecturally positioned to miss, and it happens to be the category that most marks your Japanese as foreign.
How I judged these
1. Does a human being read the whole thing? Not a sentence. The whole thing.
2. Does it work at document length, or is it capped at a paragraph? Check the character limit before you subscribe.
3. Does it touch register at all — either by asking who the reader is, or by flagging 文体 drift, or by refusing to guess?
4. Is it teaching you the writing system or teaching you to write? Both are legitimate. They are not the same purchase.
The comparison
| Tool | Cost | Correction by | Whole document? | Touches register? | Can it pay me? |
|---|---|---|---|---|---|
| LangCorrect | Free, premium optional | Native speaker volunteers | Yes | Only if a corrector chooses to mention it | No |
| Journaly | Free tier, premium from $3.50 per month | Native speakers and peers | Yes | Same | No |
| HiNative | Free, Premium $5.68 per month | Native speakers, plus AI credits | No. Question sized | No | No |
| LangJournal | Free tier, Premium $7.99 per month | AI | Entry sized | No | No |
| Nihongo Diary | Free 1 per day at 300 characters, Plus $9 per month at 500 | AI, plus community | No. Hard character cap | No | No |
| Fix My Japanese | Free 3 per day, unlock ¥500, Pro ¥5,000 per year | AI | Paste sized, no stated cap | Partly. It has polite and formal modes | No |
| Skritter | $14.99 per month | Automated stroke grading | n/a. Characters | n/a | Yes, 15 percent — I am not in it |
| Kakimashou | Free | Automated stroke grading | n/a. Characters | n/a | No |
| italki | Roughly $10 to $30 per lesson | Human teacher | Yes | Yes, if you ask | Yes, about $10 — I am not in it |
| Bunpro, WaniKani, Rocket, JapanesePod101 | Various | n/a | No. None of them take composition | No | JapanesePod101 does. Rocket could, but I am not in it |
Read the last column again. Every tool I am about to recommend pays me nothing, because I am in none of their affiliate programmes. The one writing product here with a healthy commission on offer is Skritter, and Skritter solves the other problem — the less valuable one.
The tools, one by one
LangCorrect — the best thing here, and it is free
What it does: you write a journal entry in Japanese, and native speakers correct it. Corrections come from human volunteers, and the site is explicit that this is human rather than AI. Free to join, with an optional premium tier.
Why it wins: a human reads the whole entry. That is the entire argument. A person who grew up in the language sees your paragraph as a paragraph, and if you slide from ました into 面白かった they can see it, because they are not looking at your sentences one at a time in a loop.
The catch, and it is a real one: community correction runs on reciprocity, whether or not the site enforces it. If you write in Japanese and never correct anybody in English, you will find your entries getting thinner responses over time. Budget the time to give corrections back. This is a feature, incidentally — correcting somebody else in your own language is a surprisingly good way to notice what you do unconsciously.
Also: correction quality varies. A volunteer will fix what is wrong and will rarely explain the underlying rule. You will get this sounds more natural far more often than you will get a reason.
Journaly — the same idea, slightly different community
What it does: journal entries in your target language, corrections and suggestions from native speakers and other speakers, more than 70 languages. The free tier allows unlimited posts and feedback. Premium starts at about $3.50 per month.
Why it is here: it is the closest thing to a direct alternative to LangCorrect, and the two communities are different sizes on different days. Post the same entry to both, at least while you are working out which one has more active Japanese correctors this month. It costs you nothing.
The AI diary apps — LangJournal and Nihongo Diary
What they do: you write a diary entry, an AI corrects it instantly and explains what changed. LangJournal has a genuinely usable free tier with unlimited standard corrections, and Premium at $7.99 per month for advanced corrections and more AI features. Nihongo Diary gives you one AI correction a day at up to 300 characters free, and $9 per month for ten a day at up to 500 characters.
What they are good for: speed and volume. Instant feedback at any hour, no social cost, no reciprocity, no waiting. For a beginner producing their first hundred sentences, that loop is genuinely valuable, and the explanations are better than nothing by a wide margin.
Where they stop: the character caps and the sentence-level architecture described above. Nihongo Diary itself does not claim to address style consistency across an entry, and I would not expect it to.
How I would actually use them: as the first pass, not the only pass. Run the entry through AI to clear out the mechanical errors, then post the cleaned version for human correction. The humans then spend their attention on the things only humans notice, instead of on your particles.
Fix My Japanese — the practical one
What it does: an AI grammar corrector from a solo developer, David Mohl. Free with three corrections a day, a ¥500 unlock, and a ¥5,000 per year Pro tier. It checks grammar and particles, explains the corrections, and has polite and formal modes. The listing is unusually honest, describing the AI as a knowledgeable guide rather than a perfect oracle.
Why I am singling it out: those mode switches are the only acknowledgement of register I found in any of the automated tools. It is not the same as teaching you to choose — you set the mode, so the decision is still yours and unexamined — but it at least admits the axis exists. The listing also suggests pasting from the clipboard for fast email proofreading, which is honestly the most realistic use case in this entire article. If you live here and have to send Japanese email, this is the ¥500 that will actually get used.
HiNative — a scalpel, not a workshop
Free to use, with Premium at $5.68 per month and Premium Pro at $28.99. The relevant feature is the Does this sound natural template, which gets a short piece of text in front of native speakers.
Use it for one sentence you are unsure about, not for your essay. It is a question and answer service, and the format rewards short, specific questions. Paid tiers buy visibility for your question rather than better answers.
italki — the only one that will read a document properly and tell you why
A human teacher, at roughly $10 to $30 per lesson, will read a full piece of writing, tell you the rule you broke, and adapt to the fact that you keep breaking it. Nothing else in this article does that third thing.
I raise it knowing it is the most expensive option here, and knowing that italki runs an affiliate programme paying about $10 a referral — the smallest in this whole series and, notably, less than one lesson costs. I am not in that programme, so recommending it here earns me nothing. If you are serious about writing and can afford one lesson a month dedicated to written work you prepared in advance, it will outperform every subscription in this table. Send the work before the lesson. Do not spend the lesson watching somebody read.
Skritter and Kakimashou — the handwriting half
Skritter is $14.99 per month with a one week trial, and grades your handwriting by stroke order, direction and proportion. It is good at what it does. Its affiliate programme pays 15 percent lifetime commission, and I am not in it. I will note that the payout requires converting five paying members before the first payment, which tells you something about how these programmes are structured.
Kakimashou is free and does the core of the same job in a browser.
My honest position: for most adult learners, handwriting is a small fixed investment, not a subscription. If you want the shapes to stick, use Kakimashou for free. Buy Skritter if you have a specific reason — a handwritten exam, calligraphy, a genuine preference for learning through the hand — and not because it is the most polished product with the best marketing in a category you happened to search for.
The big courses do not do this at all
Worth stating plainly, because it is easy to assume otherwise. Bunpro makes you type Japanese, and I have praised it repeatedly for that, but you are typing into a blank in a sentence somebody else wrote — production at the fragment level, not composition. WaniKani takes no free writing. Rocket and JapanesePod101, the two products with the biggest commissions on this page, have no mechanism for you to write something original and have it corrected.
If you want composition feedback, you are leaving the mainstream course market entirely. That is the finding.
What actually breaks in learner writing, from the reading side
Three things, in the order I notice them.
1. 文体 drift, described above. The one to fix first, because it is invisible to your tools and to you.
2. Speech written down. Japanese has a real vocabulary split between 話し言葉 and 書き言葉, and learners write the way they have been taught to talk. In writing, でも becomes しかし, だから becomes したがって or そのため, ～てる becomes ～ている, ～ちゃう becomes ～てしまう, じゃない becomes ではない. None of this is about politeness — it is about medium. A learner essay full of でも and ～てる reads to me like a transcript of somebody talking, which is a different failure from being impolite and one that AI correctors mostly let through, because every one of those forms is correct Japanese.
3. 読点 placement. Japanese has no spaces between words. As I described in my guide to reading Japanese, the kanji themselves do much of the chunking work — but the 読点 does the rest, and it is doing a structural job rather than the rhythmic, breath-based job an English comma does. Learners place them where an English comma would go, and the result is a sentence a native reader has to re-parse. There is no rigid rule to hand you, which is precisely why a human reading your whole paragraph beats a machine checking your sentences.
Why the market looks like this
The same reason as every other gap in this series, and by now it is almost monotonous.
The JLPT has no writing section. At any level. From N5 to N1. It also has no speaking section, and every question is multiple choice, which I documented in detail in my article on what the JLPT actually measures.
So the only external standard in this market — the one employers ask for, the one products organise their syllabus around — asks you to recognise Japanese and never once to produce a paragraph of it. There is no exam pressure creating demand for composition feedback, which means there is no money in it, which is why the best tools for it are a free volunteer community and a solo developer charging ¥500.
The mainstream products are not being lazy. They are responding rationally to a scoreboard that does not count this.
What to actually do
- Decide your 文体 before you write anything. Diary, essay, report, article: 常体. Letter, email to a stranger, anything addressed to a reader: 敬体. Write it at the top of the page if you have to.
- Write a short entry three times a week. Short and frequent beats long and occasional, because the point is the decision you make at the start, and you want to make it often.
- Run it through a free AI corrector first to clear the mechanical errors — LangJournal free tier or Fix My Japanese free daily allowance.
- Then post the cleaned version to LangCorrect or Journaly for human eyes, and correct somebody else in English while you wait.
- Once a month, if you can afford it, take one italki lesson built entirely around a piece of writing you sent in advance.
- Handwriting: use Kakimashou, free, in whatever gaps you have. Do not subscribe unless you know why you are subscribing.
- Every so often, write something over 800 characters. It is the only way the style problem becomes visible to you, and every tool in this article except the human ones will actively prevent you from doing it.
The bottom line
The best writing correction available to you is a free website staffed by volunteers, and the second best is another free website staffed by volunteers. The most useful paid thing in this article is a ¥500 unlock from a solo developer. The most expensive option, italki, is also genuinely the best if you can afford it, and its affiliate programme — which I am not in — would pay me less than a single lesson costs.
The one product here that would pay me properly is a handwriting app, in the half of this category that matters least to most adults, and I have told you to use the free alternative instead.
And the rule that will do more for how your Japanese reads than any subscription in this table is the one Japanese eight year olds are taught by national curriculum and no app has ever mentioned to you: pick 敬体 or 常体, and stay in it to the end.