Canonical & hreflang causing strange Google results

[rango]

We recently made a significant change to pages on our site that are translated into other languages, largely following advice from this help article: [support.google.com...]

But the result has been bizarre and I'm seriously considering reverting the changes.

This is roughly how our pages are structured

English page title might be: 'Widgets" - url is /widgets-en.html
Japanese page title might then be: "Widgets[in Japanese]" url is /widgets-ja.html

Large areas of content are translated. As much as possible basically - unfortunately some things like product descriptions aren't possible.

Both English and Japanese have the canonical set to /widgets-en.html

English page has a rel="alternate" hreflang="ja" pointing to the Japanese url
Japanese page has a rel="alternate" hreflang="en" pointing to the English url

As Google claims, this should "help Google serve the correct language or regional URL to searchers"

Now, here's the weird part: when I do a search for "Widgets" on Google.jp, it returns the url for /widgets-ja.html. But the title and snippet are from widgets-en.html !

What's more, if I do a search for "Widgets[in Japanese]" on Google.jp, the result isn't returned at all.

I should add, the Japanese search I am doing is quite specific - not a generic phrase that it might not rank for. It really should be there. I just find the whole thing bizarre. Like it's now only ranking based on the English phrases and entirely ignoring the translations.

Is my basic structure here correct? Should that canonical even be there always pointing back to English or is that just messing up any chance of ranking in other languages?

I should add, this is all part of attempts to de-pandalize the site. Unfortunately at the moment, it's only had the rather negative effect of losing a large portion of our foreign language traffic without any benefits at all on the English side. :(

[g1smd]

The canonical tag usage sounds wrong and I would remove it.

The hreflang stuff sounds OK.

Having some pages as mixed ja/en language doesn't sound like a good idea.

[rango]

Thanks for responding g1smd!

When there's product names involved, it's pretty much impossible to have everything fully translated. Operating on the premise that it's still better for the user to see *as much as possible* in their own language, I think this is a normal setup.

In fact, the Google help page on this is specifically discussing situations where the boilerplate is changing language, but not the content. It would appear they are making an attempt to handle this correctly, but apparently their advice is wrong.

The way we have implemented canonical seems to be pretty much exactly what Google is advising in their help article on the topic.

But hmm, just see they have this in that help article now:

"When Google discovers a cluster of pages with a single canonical URL, our algorithms will use the title and snippet from the canonical version in our search results. Therefore, it's a good idea not to include region-specific content in the title and meta description tags of the canonical URL. For example, use "Example Widget Inc" instead of "Example Widget USA Inc" or "Example Widget UK"."

So it seems like they will always show the English title. Really not getting how this is meant to be better for their multilingual users :?

[rango]

Hmm, darn. I see now that this canonical usage is more intended for "en-US" vs "en-GB" type situations. Not "ja" vs "en".

There's this line in that help file which I obviously didn't pick up on:

"Don't use rel="canonical" on the German page, because the pages are in different languages and therefore are substantially different."

[Swirleigh]

You said "Both English and Japanese have the canonical set to /widgets-en.html"

The canonical is telling Google to list widgets-en.html and not widgets-jp.html. Get rid of the canonical on the widgets-jp.html page, then Google will list in the language the person searched.

[rango]

Thanks Swirleigh, this is now fixed (see my follow up posts) and all results have returned back to normal. Phew :)

Just as a follow up - the way I was doing it is what is meant to be done to distinguish between locales of the same language.

For example, en-US pages should have a canonical pointing to en-GB or vice versa, whichever is deemed the original. At least, that's the way I read Google's help articles on this.

Somehow I got confused and thought this applied to all languages and Google would use the hreflang to still deliver the right language to the right person.