I thought my language redirect hid my Japanese pages from Google. The numbers said otherwise

The morning of 13 October

My lab site, pw.migurei.ai, has Japanese and English versions. The Japanese pages carried a little JavaScript: if the browser's first language was not Japanese, the page switched to the English version.

On the morning of 13 October 2026 I inspected the 13 URLs in my sitemap with Search Console. Six were indexed: five English pages and /ja/scanner, the only Japanese page with no English twin. Every Japanese page that had an English twin was missing. Most sat at "Crawled - currently not indexed": Google had fetched them and chosen not to keep them.

One page was stranger still. The HTML of /company declared itself as canonical. Search Console reported the user-declared canonical as /en/company.

The logs told a neat story

I opened the logs. On 12 October Googlebot fetched /company with no Accept-Language header. A second or two later it fetched /en/company twice, along with the JS and CSS files, this time sending Accept-Language: en-US.

It looked exactly like Google's renderer running my script, seeing an English-speaking browser and getting redirected.

Google's documentation pointed the same way. Its guide to managing multi-regional and multilingual sites (updated 10 December 2025, read on 13 October 2026) says to avoid automatically redirecting users from one language version to another, and that the crawler sends requests without Accept-Language. The page on locale-adaptive pages warns that Google might not crawl, index or rank all of your content for different locales.

So I wrote the claim: automatic language redirects keep the Japanese version out of Google's index. I removed the redirect that same day and waited to measure.

The turn

The baseline reading came on 15 October. All five Japanese pages in the experiment were already on Google, even though the copy Google held was from before the fix and still pointed its canonical at the English page.

So the redirect had not kept them out of the index. The core of my claim was wrong.

By the next two readings, on 17 and 19 October, Googlebot had recrawled all five and Google had chosen each page as its own canonical. That fits the idea that the redirect muddled canonicals. But I had no control group, so I cannot separate the fix from everything else that happened at the same time: sitemap lastmod dates changing, my "Request indexing" clicks, and Google simply coming back. The detailed numbers are on the pw research pages.

What to do with your site

Follow Google anyway, because it is the official guidance, not because of my experiment:

1. Do not redirect visitors to another language based on their browser or location.
2. Keep a language menu; if you like, show a one-line hint that an English version exists, without switching.
3. Declare hreflang for each pair of pages, and give each language version a canonical that points to itself.

For a site serving Japan this matters: as I understand it, Gemini and AI Overviews can only cite your Japanese page if it is in Google's index.

How sure is this

The claim is refuted, and I am keeping it public rather than deleting it. The limits section below was written while it was still a hypothesis. The experiment covered one site, five pages and one Search Console account. And being "on Google" does not yet mean a page shows up for Japanese queries or in AI Overviews.