Japan Ranks 96th of 123 in English. Its Weakest Skill Is Writing, and Every Interface Is Writing
EF publishes an English proficiency ranking every year and Japan's position in it gets reported every year with the same tone of national embarrassment. The 2025 edition put Japan 96th of 123, with a score of 446 against a global average of 488. It was the eleventh consecutive year the ranking fell.
I do not think the rank is the interesting number, and I will explain why I distrust it in a moment. The interesting numbers are the four underneath it, on the Japan fact sheet.
| Skill | Score |
|---|---|
| Reading | 454 |
| Listening | 437 |
| Writing | 394 |
| Speaking | 393 |
Sixty points between the best skill and the worst. Reading is comfortably the strongest thing Japanese test takers do in English, and writing is close to the weakest.
Every string in a product interface is writing.
Why I trust the shape and not the rank
The methodology deserves a caveat before I build anything on it, because a lot of commentary skips this.
The EF index is not a national sample. It is compiled from people who chose, unprompted, to take a free online English test. That group is not representative of Japan and is not representative in the same way in every country. A country where taking an English test online is a normal thing an ambitious person does will produce a different sample from one where it is a niche activity. Comparing Japan's 446 to another country's number carries all of that with it.
So I would not use the rank to conclude much about Japan. Eleven consecutive years of decline is more likely to be telling you something about who is taking a free online test in 2025 versus 2014 than about the country's classrooms.
What survives the caveat is the internal comparison. The same people, on the same day, on the same test, scored 454 on reading and 394 on writing. Whatever selection effect brought that sample together applies equally to all four skills. The spread is real.
The spread is exactly the wrong way round for building products
Consider how English enters a Japanese software team.
Documentation comes in English and gets read. Library docs, RFCs, error messages from a package, a Stack Overflow answer, an API reference. Reading is the strongest skill and it gets used constantly, so it stays strong.
Then the team ships an English version of the product and somebody has to write the interface. Button labels. Empty states. Error messages. The one sentence explaining what a permission does. Onboarding copy that has to be understood by someone who is annoyed and in a hurry.
That is production writing, in the weakest skill, usually done by whoever on the team is considered good at English, which in practice means whoever reads it best.
Then it gets reviewed by people who read English better than they write it. Reading comprehension is exactly the ability that lets you understand text that is technically wrong. You look at "Please input the correct information" and you understand it perfectly, so it passes.
The result is not gibberish. Japanese product English is almost never incomprehensible. It is comprehensible and slightly off, in a way that costs trust without ever causing a support ticket. "Failure of registration." "Please wait a moment for a while." "This service is now suspending." A native speaker reads those, understands them and quietly downgrades the whole product.
The job function scores
The fact sheet also breaks scores down by job function, and this is where I stopped and read it twice.
| Function | Score |
|---|---|
| Strategy and project management | 578 |
| Teacher | 561 |
| Marketing | 551 |
| Accounting and finance | 538 |
| Admin and clerical | 530 |
| R&D | 519 |
| Sales | 516 |
| Human resources | 510 |
| Customer service | 505 |
| IT | 502 |
| Technicians and maintenance | 493 |
| Student | 477 |
IT sits tenth of twelve. Below accounting. Below admin and clerical. Seventy six points below strategy and project management.
I want to be careful here, because this is a self selected sample and the IT category is enormous and vague. But the direction is worth sitting with. In the sample, the professionals who physically type the strings that appear in an exported product score lower in English than the accountants.
And the group that scores highest, strategy and project management, is generally not the group writing button labels.
What I actually do about this
I work in four languages and I am confident in two of them for professional writing. I know what it is like to be fluent enough to be dangerous, which is the state where you produce text that is fine and does not sound like anyone.
A few things that have worked on real projects.
Stop treating interface copy as translation. Translation is a task where you receive Japanese and return English, so the output is measured against the input rather than against the user. Interface copy is a design decision with a language attached. When I have got teams to write the English string first, in a doc, next to a description of what the user is trying to do, the quality moves immediately, even when the writer's English is unchanged.
Give the strings a single owner. Distributed authorship is how you end up with "Sign in", "Log in" and "Login" in one product. The owner does not need to be the best English speaker in the company. They need to be the only one who can commit a string.
Write down the register, and be specific. Not "friendly and professional", which means nothing. Actual rules. Errors say what happened and what to do next. Buttons are verbs. No exclamation marks. Never apologise twice in one message. A team with weak writing and a clear specification will beat a team with better writing and no specification.
Use a native reviewer for the twenty strings that matter rather than a proofreader for all four hundred. Onboarding, payment, errors, permissions and deletion. That is where a wrong register turns into a lost customer, and it is maybe an hour of somebody's time.
And be suspicious of machine output for exactly the same reason the review process fails. Modern translation produces grammatical English with no point of view, which is precisely the failure mode nobody on a Japanese team is equipped to catch, because it reads correctly.
The thing the ranking hides
Japan does not have an English problem in the way the annual headline suggests. It has a writing problem, sitting inside an English problem, at exactly the point where the country's products meet everybody else's customers.
A country can export excellent hardware with weak English on the box. Software is different, because with software the English is not on the box. It is the product. Every screen is a paragraph somebody wrote.
The 394 is not a national embarrassment. It is a specific, addressable, comparatively cheap gap sitting between a lot of good work and the people who might pay for it.