Multilingual Content Strategy

A multilingual content strategy determines how a site adapts its content for each target market, which markets to prioritise, and how to maintain quality and consistency as the number of languages grows.

Translation vs localisation

The most important distinction in multilingual content is between translation and localisation.

Translation converts text from one language to another with equivalent meaning. A translated page uses the correct words in the target language but may still feel foreign to a local audience. Cultural references may not land. Product examples may be irrelevant. Pricing or measurements may use the wrong convention. The page is linguistically correct but contextually mismatched.

Localisation adapts the content for the cultural context of the target audience. It goes beyond word choice to address:

• Local idioms and natural phrasing
• Culturally appropriate examples and case studies
• Local currency, units, and date formats
• Local legal or regulatory requirements (particularly in finance and healthcare)
• Locally relevant calls to action and contact details
• Images and visual content appropriate to the market

For SEO purposes, localisation consistently outperforms translation. Localised content matches what local audiences actually search for, uses the phrasing they use, and answers the questions they ask. International keyword research for each target language is a prerequisite: search volume, dominant phrasing, and user intent differ significantly across markets even for the same underlying topic. Translated pages often rank for the wrong queries or fail to rank at all because the keyword targeting reflects the source market rather than the destination market.

Machine translation

Google’s position on machine translation moved in 2025, and a lot of SEO advice still reflects the older line. Its multilingual documentation used to tell site owners to block automatically translated pages with robots.txt, on the grounds that automated translations “could be viewed as spam”. Google removed that section on 11 June 2025, describing the edit as aligning the docs with the March 2024 spam policy update and “a docs-only change, no change in behavior”.¹ Asked about it directly, Google stated that its “policies do not strictly define content that has been translated by AI as spam”, and that the scaled content abuse policy “mentions automated transformations, including translations, as part of the overall warning against creating large amounts of unoriginal content that provides little to no value to users”.²

So the operative test is value, not production method. The spam policy applies “no matter how it’s created”,³ which cuts both ways: machine translation is not disqualifying, and hand-translation is not a safe harbour if the result is thin.

That said, the practitioner case for reviewing translations before publishing is unchanged, and it rests on quality rather than on policy risk:

• Keyword mismatch: machine translation converts what you wrote, not what local users search for. If the source page was optimised for British English queries, the translation optimises for the literal translation of those queries, which may have different or zero search volume in the target language. This is the strongest argument and it is not solved by better translation quality.
• Unnatural phrasing: stiff, literal translations read badly, and readers leave. Be careful how you justify this, though: Google has repeatedly denied using analytics engagement metrics as ranking signals, with John Mueller stating there is “a bit of misconception here that we’re looking at things like the analytics bounce rate when it comes to ranking websites, and that’s definitely not the case”.⁴ The cost of bad phrasing is lost readers and lost conversions, not a ranking penalty.
• Tonal register: formality conventions vary by language. German business communication tends to be more formal than British English; French more formal than American English. A machine translation may apply the wrong register for the market.

The workable middle path has a name in the translation industry: machine translation post-editing, where a native speaker edits machine output with the latitude to change phrasing substantially rather than only correcting grammar. For most content that is the right balance of cost and quality. Reserve full human translation or transcreation, where the message is rebuilt for the market rather than converted, for high-value commercial pages and anything where a misjudged tone carries real cost.

How does Google decide what language a page is in?

From the visible content, and from nothing else you might expect. Google’s documentation is explicit: “Google uses the visible content of your page to determine its language. We don’t use any code-level language information such as lang attributes, or the URL.”⁵

Two practical consequences, both of which catch sites out:

Don’t mix languages on a page. Google advises “using a single language for content and navigation on each page, and by avoiding side-by-side translations”.⁵ A page with English navigation around German body copy is harder to classify than one that commits.

Half-translated pages are a named failure mode. Google warns that “translating only the boilerplate text of your pages while keeping the bulk of your content in a single language (as often happens on pages featuring user-generated content) can create a bad user experience if the same content appears multiple times in search results with various boilerplate languages”.⁵ This is the exact pattern a partial rollout produces, which makes it a content-parity question as much as a technical one.

Note that the lang attribute still has real uses for accessibility and for screen readers. It just isn’t how Google works out the language.

Google may translate your pages for searchers

Worth knowing before deciding whether a market needs its own content: Google sometimes machine-translates results itself. “Sometimes Google may translate the title link and snippet of a search result for results that aren’t in the language of the search query”, and if a user clicks a translated title link, “they’re presented with a page that’s been machine translated”.⁶

The feature covers a defined set of languages, requires no action to enable (“you don’t need to do anything to opt in”), and can be opted out of with the notranslate rule. Performance from it is visible in Search Console’s Performance report via the Search Appearance filter.⁶

This does not remove the case for properly localised content, since a machine-translated snippet of source-market content still targets source-market keywords. It does mean the choice in a low-priority market is rarely between “translated presence” and “no presence at all”.

Content parity

Content parity means having the same pages available in all target languages. It sounds like the obvious goal, but it is not always the right one.

For product pages, support content, and legal documentation: content parity is important. Users who land on a French version of a site and find half the content only in English are likely to leave. There is a search cost as well as a user cost here, since a page whose navigation is translated but whose body copy is not is the boilerplate-only pattern Google names as a problem.⁵ Better to publish fewer pages fully localised than many pages half-done.

For editorial content, blog posts, and thought leadership: parity may not be achievable or worthwhile, especially for smaller sites. A better approach is to create a core set of high-priority pages in all languages and build additional content markets-first, based on local keyword research rather than as translated versions of English content.

Market prioritisation

Adding a new language to a site is a significant ongoing commitment. Every content update needs to be reflected across each language version; every new page needs evaluating for inclusion in each market. Before adding a language, consider:

• Search demand: is there meaningful search volume for your topics in the target language?
• Market size: is the potential revenue worth the content and maintenance investment?
• Competitive conditions: are local competitors already well-established? What does competing effectively require?
• Internal capability: do you have native-language review capacity, or will you rely on machine translation?

Starting with one or two markets and doing them properly produces better results than spreading content thinly across many markets.

How do you manage content updates across languages?

The most common failure in multilingual content management is update drift: the English version of a page is updated, translated versions are not, and the content gradually diverges. This creates inconsistencies that confuse users and can introduce factual errors in secondary markets.

Practical approaches for managing updates:

• Track which pages exist in which languages and when each was last updated.
• Prioritise update propagation for product pages, pricing information, and any legally sensitive content.
• For editorial content, decide in advance which updates require re-translation and which are minor improvements that do not need mirroring.
• An outdated translated page is not always better than no translated page. For factually time-sensitive content, it often is not.

Duplicate content between language versions

Pages in different languages are not treated as duplicate content by Google, even when the underlying information is the same. Hreflang annotations signal that these pages are related but serve different audiences. Properly annotated multilingual content does not require additional canonicalisation across languages.

The duplicate content risk that does apply: pages in the same language with minor regional variations (British English vs American English, for example) where the content is near-identical.

Google documents a remedy for this case. Where similar or duplicate same-language content sits on different URLs, such as example.de/ and example.com/de/, its advice is to “pick a preferred version and use the rel="canonical" element and hreflang tags to make sure that the correct language or regional URL is served to searchers”.⁵ Note that this is canonicalising within a language, not across languages, and the two mechanisms do different jobs: hreflang says which version suits which audience, canonical says which version is the one to index. There is more on that distinction in hreflang best practices.

The practitioner alternative is to make the regional versions genuinely different enough to justify separate URLs, which is usually the better outcome where the market warrants the effort. Where it does not, take Google’s documented route rather than publishing near-identical pages and hoping hreflang sorts it out.

Multilingual content and AI search

One finding is worth carrying, with its limits stated precisely. The language of a query changes which sources AI systems cite, not merely which language those sources are in. Profound’s March 2026 analysis of 3.25 billion AI citations across seven models and 14 countries, filtering prompts to each country’s native language, measured this most clearly in social citations: in Google AI Overviews, YouTube moved from 38% of social citations on English prompts to 65% on Brazilian Portuguese ones while Reddit fell from 21% to 7%, and TikTok rose from 3.1% to 15.9% on Spanish prompts, described as roughly a fivefold increase over the English baseline.⁷ ChatGPT behaved differently again, with Reddit accounting for 51 to 76% of its social citations regardless of locale.

Two limits on what that supports. The measured effect is on the social-platform mix specifically, not a general claim about all cited sources. And it is vendor research, so treat the direction as more reliable than the precise figures.

The defensible conclusion is narrow but useful: a language’s citation pool is largely composed of sources in that language, so content existing only in English is largely absent from answers generated for other languages. That is an argument for localised content over and above the ranking argument. It is not evidence about how retrieval selects sources, so don’t extrapolate a mechanism from it.

Frequently asked questions

Should the English version always be the primary version?
Not necessarily. The primary version for SEO purposes is the one with the strongest link authority and the clearest canonical signals. If the German market is larger than the English market for your business, the German version may function as the primary version. Base the decision on business reality rather than English-first assumptions.

How do you handle idioms and colloquialisms?
They generally do not translate. Replace the idiom with a locally natural equivalent in the target language, or rephrase to avoid idiomatic language entirely. A professional translator will handle this automatically; machine translation will not.

Can you SEO-optimise translated content?
Only partially. Translation produces content optimised for the source market’s keyword patterns. To optimise for the target market, independent keyword research in the target language is needed, followed by revisions that reflect local search behaviour. This is the primary reason translated content underperforms against locally created equivalents.

Is AI-translated content against Google’s guidelines?
No. Google has stated that its “policies do not strictly define content that has been translated by AI as spam”,² and in June 2025 it removed the older documentation advising site owners to block automatically translated pages.¹ What does apply is the scaled content abuse policy, which targets large amounts of unoriginal, low-value content “no matter how it’s created”.³ The question to ask is whether each translated page is genuinely useful to a reader in that market, not which tool produced the text.

Footnotes

1. Search Central documentation updates, June 2025 — Google Search Central ↩ ↩²

2. Google Softens Its Stance On Automated Translations — Search Engine Roundtable ↩ ↩²

3. Spam policies for Google web search — Google Search Central ↩ ↩²

4. Is bounce rate a Google ranking factor? — Search Engine Journal ↩

5. Managing multi-regional and multilingual sites — Google Search Central ↩ ↩² ↩³ ↩⁴ ↩⁵

6. Translated results in Google Search — Google Search Central ↩ ↩²

7. How query language reshapes AI citations — Profound ↩