Your Japanese Site Says a Different Company Makes Your Product.

▶ 60-Second Slide Summary Prefer the 8-slide version of this article? Tap to read in 60 seconds. View slides →

Quick Answers

Our Japanese site was produced by a professional agency. What could actually be wrong with it?

In the work behind this article, the findings that mattered were almost never translation quality in the sense a linguist would flag. They were ownership and structure failures: a Japanese page naming a different company where the English says “we”; one item out of ten in a partner FAQ carrying entirely different content in the two languages; a page served as lang="ja" with an English title and description. A good agency translates the strings it is given. None of these are failures of the strings.

How do I check any of this if I do not read Japanese?

By comparing structure rather than wording. Company names are the same characters in both languages, so you can search for them. Item counts in an FAQ can be counted. A page title lives in the page source. Every check in this article was chosen because someone who reads no Japanese can run it and get a yes or no. What none of them tell you is whether the Japanese reads well — that still needs a person who reads Japanese, and it is a different job.

TL;DR

Six patterns survived re-checking across teardowns of live corporate sites, and the strongest one is not about translation at all: the Japanese page names a company that does not make your product, in sentences where the English says “we”. Below it sit a correct technical term sitting three lines above an incorrect one on the same page; a ten-item FAQ where the tenth item is a different question in Japanese and a contractual obligation therefore exists in only one language; pages served as Japanese with English titles, sitting beside pages that are properly Japanese, which is what proves it is an oversight rather than a policy; and sites carrying two different qualities of Japanese at once, which a site-wide average hides completely. The last section is the one worth your time if you are choosing a vendor: five findings we discarded, why, and why a list of findings that all point the same way is a sales document rather than an audit.

Key Takeaways

• The worst failures are subject errors, not word choices. A sentence that says a different company makes your product is factually false about your own property, and no amount of fluency fixes it.
• “It is our house style” stops working when the correct term is on the same page. Several findings were a correct term and an incorrect one a few lines apart, which means the right answer already exists inside the company.
• Equal item counts do not mean equal content. A ten-item FAQ matched ten-for-ten in both languages, and the tenth item was a different question entirely — so a reporting obligation existed in English and nowhere in Japanese.
• A machine count of “English strings remaining” contains zero instances of bad translation. In one case the count matched other locales exactly, which means it was not a Japanese problem at all.
• An audit that finds nothing to discard is not an audit. In the same period we dropped five findings, including the most visually dramatic one, because they did not survive being checked a second time.

Your Japanese Page Names Someone Else’s Company

This is the strongest thing we have found, and it is the one that takes the least Japanese to verify.

On a full sweep of 1,135 product and service pages on one manufacturer’s site, nine sentences across seven pages — eleven occurrences in total — had replaced the English “we” and “our” with the name of an unrelated company. The findings crossed three different business areas. The English original had none. Neither did the four other language versions. It existed only in Japanese.

The heaviest instance was on a spare parts page. One sentence read, in effect, “〈the other company〉, as the manufacturer of your equipment…”, and the sentence immediately beside it referred to “〈the site owner〉 genuine parts”. Two adjacent sentences, two different owners, on a page whose entire commercial purpose is to persuade a maintenance engineer to buy from one of them.

Several of the other instances sat inside regulatory language — references to FDA, USP, ISO and EU regulations — which means the party named as guaranteeing compliance was a company with no relationship to the product.

It is worth being precise about why this is serious, because it is easy to file it under “translation errors” and move on. It is not a translation error. It is a subject error. The sentence is grammatical, fluent, and false about the reader’s supplier. A reviewer checking whether the Japanese reads naturally would pass it, because it does.

The Correct Term Is Already on the Page. Three Lines Up.

A second pattern appeared repeatedly: a correct technical term sitting a few lines above an incorrect one, on the same page, in the same section.

In one case the two words sounded nearly identical but used different characters, and the difference changed the meaning. In another, the English was printed alongside the Japanese within the same two lines, and the two katakana spellings adjacent to it disagreed with each other. In a third, a manufacturing term that had originated in Japanese, travelled into English, and come back again had returned as a coinage that does not exist in the language.

The reason this class of finding is worth more than its size suggests is that it removes an answer. When a page contains only the wrong term, a vendor can say it is a house convention. When the right term is three lines above the wrong one, the correct answer demonstrably already exists inside the organisation, and what failed was the process that was supposed to apply it consistently.

One FAQ Item Out of Ten Is a Different Question Entirely

On a marketplace site, a partner-facing FAQ was built as an accordion with ten items in both English and Japanese. Items one through nine corresponded. The tenth did not.

The English tenth item explained an obligation: which transactions a partner is required to declare, including ones that did not close on the platform, and the fact that declared transactions appear on the next invoice. The Japanese tenth item was a generic “Need help? Contact customer support.”

So the notice of a contractual obligation, with direct consequences for fees, billing and disputes, did not exist anywhere in the Japanese version of the page. This is not a question of how well something was translated. A piece of the agreement is present in one language and absent in the other.

The structural lesson is the useful part: the item counts matched. Ten and ten. Any check that compares counts would have passed this page, and any check that compares the two languages heading by heading would have caught it in under a minute.

The Page Is lang="ja". The Title Google Shows Is English.

On the same site, three pages told the whole story between them. One served a Japanese page with both the <title> and the meta description still in English. A second had the title in English and the description in Japanese. A third had both properly in Japanese.

The third page is what makes this a finding rather than a preference. If every page were in English, you could argue it was a deliberate choice about global consistency. Because some pages are done properly, what remains is an oversight — and it landed, as these things tend to, on the page selling a paid certification service.

The cost here is specific and easy to underestimate. Search engines display the title and description. A Japanese buyer searching in Japanese sees an English snippet for a Japanese page and forms a judgment before the page is opened. You can have a perfectly translated page and still lose at the touchpoint that decides whether anyone reads it.

Two Different Japanese Languages on One Site

On a ten-page close read of one site, the pages did not behave as a single population. Pages that read as though they had been written in Japanese rather than converted into it carried no literal-translation sentences at all. A page that read as a literal conversion ran at roughly 100 per cent — on a base of four sentences, which is small enough that the number is an indication rather than a measurement. The pages that also carried the other company’s name ran somewhere around sixty to seventy per cent.

What that spread means in practice is that a site-wide average is close to useless. It will land somewhere in the middle, describe no actual page, and reassure everyone. The pattern behind it is ordinary enough: a company employs people who write good Japanese, they write the pages they own, and a separate body of pages arrives from a product database or an outsourced pipeline and never passes the same desk.

This is also why we report quality per page rather than per site, and why a proposal that opens with a single site-wide percentage should be read carefully.

What We Threw Away — And Why That Matters More Than What We Kept

If you are using this article to decide whether an outside review is worth commissioning, this is the section to read.

In the same body of work, five findings were discarded because they did not survive a second look:

• “The Japanese news listing is empty.” Every other language version was also empty, including English. A site-wide bug, not a Japanese one.
• “The sitemap has no entries.” Our own request had been stopped by bot verification. The sitemap later turned out to contain 3,263 entries. The zero was ours, not theirs.
• “Japanese has far fewer pages than the other languages.” Japanese had 3,263, Korean 3,422, French 3,391. Comparable.
• “Four sentences where the Japanese reverses or contradicts the meaning.” The English original and the German version said the same thing in each case. The Japanese was a faithful translation of a source that read that way.
• “The testimonial carousel is offset by one in Japanese.” This was the most visually dramatic finding of the set. Re-checked in a real browser after rendering, card by card, every name matched its image. The offset was an artefact of extracting text in DOM order from a flat parse.

Two things follow from that list. The first is a warning about counting. In one of these audits, a machine count of English strings remaining in the Japanese pages matched the count in another language almost exactly, which is a reliable sign that the strings are a platform behaviour rather than a Japanese failure. More generally, a total of “N English strings remaining” contains, by construction, zero instances of a sentence being translated badly. The two are different measurements, and only one of them is what you are buying.

The same applies to the small stuff we found and did not lead with: extra spacing inside a word, image alt text left in English where it never renders on screen, a single dropped character, a misspelled font name in a stylesheet. All real. All fixable by the client in an afternoon. None of them a reason to commission anything, and presenting them as though they were is how a findings list becomes a sales document.

The second is about direction. If every finding in a report points the same way — everything is broken, nothing is fine — that is not what a real site looks like. Sites are uneven. A review that comes back with nothing discarded has not been checked twice.

Running This on Your Own Site This Week

Assembled, the checks above take about an hour on a mid-sized site and need no Japanese:

1. Search your Japanese pages for company names that are not yours. Compare any hit against the English sentence.
2. Search your five or six core technical terms and look for two spellings on one page.
3. Put your terms and partner FAQs side by side in both languages and compare headings from the top.
4. View source on your three most commercially important Japanese pages and read the <title> and meta description.
5. Read one hand-written page and one database-fed page together and decide whether they sound like the same company.

Be clear about the limit. None of this tells you whether your Japanese is good. It tells you whether your Japanese site says true things about who you are, whether it contains the same commitments as your English site, and whether it presents itself to a Japanese search engine as a Japanese page. Those are the failures that cost money quietly, because everyone who could notice them is reading the other language.

The findings described here come from localization QA work on live corporate sites. Company names, product names and URLs are withheld deliberately. Figures are as measured on the dates that work was done, and sites change — the counts are offered as an illustration of scale rather than as a current description of anyone’s website.

Frequently Asked Questions