Technical · 6 min read

Multilingual websites and AI search: hreflang, translation and local engines

AI engines answer in the user’s language and prefer native sources. Here is how to build multilingual sites that get cited in every market.

By the AEO GEO Marketing team · Updated

Key takeaways
  • AI assistants answer in the language of the question and favour native-language sources.
  • Hreflang helps engines show the right language and country version.
  • Professional, localised content performs far better than raw machine translation.
  • Each market may use different engines, platforms and trust signals.
  • Local authority signals matter in every language you target.

When someone asks an AI assistant a question in Spanish, Arabic or Japanese, the assistant typically answers in that language and tends to draw on sources written in it. A business with excellent English content but nothing in the local language may be invisible to local buyers, even if it serves them well.

For international businesses, multilingual content is not just translation; it is the foundation of visibility in each market.

Choose markets and languages deliberately

Start with business priorities: which countries and languages generate or could generate the most revenue? Consider where you have customers, staff, partners and legal ability to operate. It is better to do two languages well than six poorly. Some markets need several languages, such as Switzerland, Belgium, Canada or the UAE.

URL structures

Common options for multilingual sites are:

  • Subdirectories: example.com/es/, example.com/fr/ (easy to manage, shares domain authority).
  • Subdomains: es.example.com (more separation, slightly more complex).
  • Country-code domains: example.es, example.fr (strong local signal, more expensive to manage).

Subdirectories are a practical choice for most businesses. Whatever you choose, keep it consistent.

Hreflang explained

Hreflang attributes tell search engines which language and region each page is for and which pages are equivalents. For example, an English page for the UK, an English page for the US and a Spanish page for Mexico can be linked so each audience sees the right version. Hreflang can be added in the HTML head, HTTP headers or XML sitemaps. Each page should reference all its alternates, including itself, and include an x-default for users who do not match any version.

Common hreflang mistakes

  • Missing return links (page A references page B, but B does not reference A).
  • Wrong language or country codes.
  • Pointing to redirected or non-canonical URLs.
  • Using hreflang without truly equivalent content.
  • Forgetting to update hreflang when pages are added or removed.

Validate hreflang with SEO crawling tools and monitor indexing in each market.

Translation versus localisation

Translation converts words; localisation adapts content to the market: currency, units, examples, regulations, cultural references, local terminology and search behaviour. A page about mortgages translated word for word from US English will be wrong in Germany. Localised content answers the questions local buyers actually ask, with local facts.

Avoid raw machine translation

Machine translation has improved enormously, but publishing it without review risks errors, awkward phrasing and missing local context. Readers notice, and engines may judge the content as lower quality. Use machine translation as a starting point if you like, then have native speakers with subject knowledge review and adapt it.

Keyword and question research per market

People in different markets ask different questions and use different terms. Research each market separately: local search data, local AI prompts, local competitors and local forums. Do not assume that translating your English keywords captures how people search in another language.

Local engines and platforms

Most markets rely on Google, but some have important local engines and platforms. South Korea has Naver; mainland China has Baidu and domestic AI assistants; Russia has Yandex; Japan has strong use of Yahoo! JAPAN alongside Google. Social and messaging platforms also vary: LINE in Japan and Thailand, WeChat in China, WhatsApp in many regions. Adapt your strategy to the platforms that matter in each market.

Local authority signals

Authority is local too. Local-language media coverage, reviews on local platforms, local directory listings and local partners strengthen visibility in each market. A brand well known in the US may have little authority in France until French sources mention it.

Structured data in multiple languages

Add structured data to each language version, with localised names, descriptions and inLanguage properties. Keep organisation details consistent, and use the same @id for the organisation across languages. For local offices, use LocalBusiness schema with local addresses and phone numbers.

Show prices in local currency where possible, explain taxes and shipping clearly, and include legally required information for each market, such as company registration details, privacy notices and consumer rights. AI assistants answering cost questions in a local language need local pricing to cite.

Content parity and priorities

You do not need to translate every page. Prioritise pages that matter most in each market: home, core services or products, pricing, contact, key guides and location pages. Clearly link users to available languages, and avoid automatically redirecting based on IP address alone, which can block crawlers and frustrate users.

A worked example: a SaaS company entering Germany

A UK-based SaaS company wants to grow in Germany but receives few German leads. AI assistants answering German questions about its category cite German competitors. The company creates a German subdirectory, localises its core pages with a native-speaking product marketer, researches German questions and terminology, adds hreflang, publishes German case studies from local customers, and gains coverage in German trade media. It also lists itself on German software review platforms. Over the following months, it begins to appear in German-language AI answers.

Measuring international visibility

Track rankings, citations and AI mentions separately for each market and language, using prompts written natively in each language. Monitor traffic and conversions by country in analytics, and indexing for each language version in Search Console.

Language selectors and user experience

Make it easy for users to find their language. Use a clear language selector, typically in the header, showing language names in their own language (Deutsch, Français, Español) rather than flags, which represent countries rather than languages. Remember users’ choices with a cookie, and never force redirects that prevent access to other versions, especially for crawlers.

Right-to-left languages

Arabic, Hebrew, Persian and Urdu read right to left. Supporting them properly requires layout changes, not just translated text: mirrored navigation, aligned text, correct number formatting and fonts that render well. Poorly implemented right-to-left pages look unprofessional and reduce trust, which matters in markets such as the Gulf where bilingual Arabic and English content is expected.

Working with translators and local experts

Brief translators with context: who the audience is, the tone of voice, key terminology, which terms should not be translated and the purpose of each page. Provide a glossary and style guide. For specialist topics such as legal, medical or financial content, use translators with subject expertise, and have a local professional review for accuracy and compliance.

Local social proof in each language

Reviews, testimonials and case studies in the local language are especially persuasive. Collect reviews from customers in each market on the platforms they use, and feature local case studies on the relevant language pages. AI assistants describing your brand in a local language can draw on these local-language signals.

Domain authority across languages

New language sections inherit some authority from your main domain when they sit in subdirectories, but they still need their own local signals. Link to new language sections from relevant pages in your main language, earn links and mentions from local-language sites, and build local profiles and reviews. Over time, each language section develops its own standing in its market.

Testing language versions with AI assistants

Ask AI assistants your key questions in each target language and note which sources they cite. If they cite competitors or your English pages instead of your local version, review whether the local page answers the question as directly, whether hreflang is correct, and whether local authority signals are strong enough.

Managing multilingual content at scale

As sites grow, keeping language versions in sync becomes harder. Use a CMS or translation management system that tracks which pages have changed and need updates in other languages. Prioritise updates for high-value pages. Mark clearly in your workflow when a page’s source content has changed, so translations do not quietly fall out of date.

Key terms in this guide

  • RAG (Retrieval-Augmented Generation): A technique where an AI system retrieves relevant documents, usually from a search index, and uses them to write a grounded answer. Most AI search engines use a form of RAG, which is why being retrievable matters.
  • Citation: A link or reference to a source shown inside an AI-generated answer. Citations are the main unit of success in GEO.
  • Hreflang: An HTML attribute that tells search engines which language and region a page is for, so the right version is shown to the right users.
  • Prompt: The question or instruction a user types or speaks into an AI assistant. In GEO, prompts are the equivalent of keywords.
  • @id: A unique identifier used in JSON-LD to refer to the same entity consistently across pages and schema blocks.
  • Canonical URL: The preferred URL for a page, declared with a canonical tag, which consolidates signals when similar pages exist.

How AEO GEO Marketing can help

If you would rather have specialists handle this, these services cover the work described in this guide:

AI SEO & Generative Engine Optimization

Get cited in Google AI Overviews, ChatGPT & Perplexity. The new search visibility standard. It includes geo / aeo full audit, ai overview citation strategy, entity authority building and more, from $800/mo.

AI-Citation Content & Copywriting

AI-citation-ready content engineered for both humans and language models. It includes answer-first articles, pillar & cluster pages, conversion copywriting and more, from $900/mo.

Web Design & Development

Performance-first websites built for conversion, speed and AI search visibility. It includes ux & visual design, core web vitals engineering, schema-ready architecture and more, from $1,500/mo.

Not sure where to start? The free 48-hour AI snapshot shows where AI engines cite you today.

Frequently asked questions

Do AI assistants translate English content for local users?

They can, but they generally prefer sources in the user’s language and may favour local competitors with native content.

Is hreflang required for AI search?

Hreflang is a search engine signal. It helps engines show the right version, which supports AI features that rely on search indexes.

Should I use separate domains for each country?

Not necessarily. Subdirectories work well for most businesses and are easier to manage.

Can I use AI to translate my site?

Yes, as a first draft, but native-speaking experts should review and localise it before publishing.

Continue with these guides:

Or see how we put this into practice:

Want this done for you?

Get a free 48-hour GEO audit that shows exactly where AI engines cite you — and where they cite your competitors.