Multilingual news website: translation, URLs and hreflang

In 2005 the phpWordPress forum announced new front-end languages as volunteers supplied them, and a reader asked in 2006 for Arabic, Urdu and Persian. The question has not changed. Translating the interface is the easy part, and a multilingual news site is mostly a content and URL problem.

What does “multilingual” mean in a CMS?

A CMS can translate three different things, and vendors often only mean the first.

Drupal’s User Guide describes exactly this split: user interface text, configuration text, and content text and files, handled by the Interface Translation, Configuration Translation and Content Translation modules. Content can be made translatable per field, so an article’s headline and body are translated while a shared photo credit is not. Other systems reach the same result through plugins or dedicated fields. According to each vendor’s documentation, check which layers are built in and which need an extra component.

How should translations be stored?

There are two common models. In the first, each translation is a separate article linked to its siblings. In the second, one article holds several language fields. A separate article per language is easier to give its own URL, its own slug, its own publish date and its own editor sign-off. For news that matters, because a Spanish version may be published hours after the English one, or never.

Ask before you choose:

• Can a translation be unpublished or corrected without touching the original?
• Is there a visible link from each article to its siblings, and does it handle a missing translation?
• Do slugs, meta descriptions, image alt text and structured data translate too?
• Can a translator work without access to everything else in the admin?

Which URL structure should a news site use?

Google’s guidance compares four options.

Subdirectories keep everything on one host, one analytics property and one set of redirects, which suits a small team. Separate domains make sense when you run distinct editorial operations per country with their own staff and legal entity. Whatever you pick, use it consistently and keep each page in one language, including navigation. Google warns that translating only the boilerplate while leaving the bulk of the content in one language can create a bad user experience.

How does hreflang work?

Hreflang is an annotation that says “these pages are the same content in different languages or regions”. Google’s documentation gives three equivalent ways to supply it: HTML link tags, HTTP headers, or your XML sitemap. A news site publishing many articles a day usually finds the sitemap easiest to keep correct, because the CMS can generate it, while page-level tags suit sites with a handful of languages.

The rules that trip people up:

1. Every version lists every version, including itself. Google says each language version must list itself and all others, and that if two pages do not point to each other, the tags are ignored.
2. Use full URLs. Alternate URLs must be fully qualified, with https, not a path such as /foo.
3. Use valid codes. A language code in ISO 639-1 format, optionally followed by an ISO 3166-1 Alpha 2 region code.
4. Add x-default for the fallback. It applies when no other language or region matches the reader’s browser setting. A language-chooser page or your main edition usually fills this role.

Because the tags only work if two pages agree, a missing translation is a common failure. If the Spanish version of a story does not exist, the English page should not list one. A CMS that generates the tags from actual translation relationships avoids this; hand-edited tags drift.

Hreflang does not tell Google what language a page is. The documentation says Google does not use hreflang or the HTML lang attribute for that and relies on its own algorithms. Set the lang attribute anyway, because screen readers and browsers use it, but treat it as separate from hreflang.

Should readers be redirected by browser language?

Google advises avoiding automatic redirects between language versions. A reader following a link to the English article should land on the English article. Offer a clear language switcher that links to the matching translation of that article, and fall back to the language home page only when no translation exists.

What does a translation workflow look like?

A workable newsroom process has four steps.

1. Decide which stories get translated. Translate by audience demand and editorial value, not everything by default.
2. Translate with a named reviewer. A fluent editor checks headlines, quotes, names and numbers. Translators need to know whether to keep local spelling and place names.
3. Publish with links to siblings. The translation goes live with hreflang relationships in place and a note of the original’s publication date.
4. Handle corrections in both. When the original is corrected, the CMS should flag translations as out of date.

Machine translation can speed up a first draft. Google’s spam policies treat using generative AI or similar tools to generate many pages without adding value for users as scaled content abuse, so bulk, unreviewed output is a risk to the whole site, not only to those pages.

Other things that differ by language

• Search. Test your CMS’s site search with real queries in each language you publish.
• Encoding. Use UTF-8 end to end, in the database, templates and feeds.
• Right-to-left languages. The theme needs direction-aware layout, which is a themes question as much as a content one.
• Feeds and newsletters. Publish one feed per language.
• Dates, numbers and currency. Format per locale, not per site.

If you are choosing a system now, the feature checklist lists multilingual support alongside the other requirements to compare, and the CMS guides cover the wider decision.