Multilingual AEO means organising a multilingual website with separate URLs, a sound canonicalisation strategy, reciprocal hreflang, and natural translation with market adaptation. That's the documented base for Google. Other AI tools need separate technical checks, and there is no single published rule for how they choose language versions or sources.
A Greek traveller searches:
"ξενοδοχείο με θέα στη Σαντορίνη για πρόταση γάμου"
A foreign visitor might search:
"boutique hotel in Santorini for a proposal"
The need is similar. The language, the phrasing, and often the criteria by which the choice is judged aren't exactly the same.
For a business addressing more than one market — in tourism, exports, or international B2B, for example — it isn't enough to have a language-switch button.
For a properly organised multilingual presence, the recommended base is for each language version to:
- have its own stable URL,
- be correctly linked to the other versions,
- be accessible to Google Search,
- be naturally written for its audience,
- and be consistent in the details and content it presents.
That's the practical meaning of multilingual AEO. It isn't a separate "algorithm for AI". It's the sound organisation of a multilingual website, on which international SEO, search engines, and AI tools that use public content can work.
None of these techniques guarantees an appearance, a mention, or a specific position in an answer.
What does multilingual AEO mean?
Multilingual AEO means organising a bilingual or multilingual website so that each real language version is a normal page and is correctly linked to its corresponding versions.
The base is international SEO.
The recommended base includes:
- a separate URL for each language,
- correct
canonical, i.e. a correct declaration of the main version when identical or very similar pages exist, - correct
hreflang, so Google links the corresponding language or regional versions, - real content in the page's language,
- sound translation and market adaptation where needed,
- and structured data that agrees with the content.
You also need to separate two concepts that are often confused:
- multilingual website: it serves more than one language,
- multi-regional website: it addresses different countries or geographic areas.
A website can be both.
For example, a Greek and an English version form a multilingual structure. Two English versions, one for the United Kingdom and one for the United States, can target different markets even though they use the same language.
This distinction also matters in hreflang, because you can declare either just the language or a combination of language and region.
You don't need to set up a different website or a different technical structure for each AI tool. First, there has to be one well-organised public website.
Separate URLs and correct canonical
The recommended base is for each language version to have its own stable URL.
For example:
https://example.com/hotel
https://example.com/en/hotel
or:
https://example.com/el/hotel
https://example.com/en/hotel
Google recommends using different URLs for the different language versions rather than relying only on one shared address that changes language depending on the visitor's settings. The crucial thing is that each version can be found and processed independently.
And canonical?
canonical and hreflang don't do the same job.
canonical is used to declare which URL you prefer as the main one when identical or very similar versions of a page exist. hreflang is used to declare the relationship between different language or regional versions.
In a typical multilingual site, a real Greek and a real English version shouldn't be treated as if they were just two technical copies. A self-referencing canonical is usually the right base for each genuinely translated version.
For example:
Greek version
<link rel="canonical" href="https://example.com/hotel" />
English version
<link rel="canonical" href="https://example.com/en/hotel" />
You don't mechanically declare all languages as canonical pointing to the Greek version. If an English page is a real English version you want to appear normally in search, it has to be able to stand as a separate page. Different language versions aren't considered duplicate content just because they carry the same meaning in a different language. Very similar regional variants in the same language — for example en-GB and en-US with nearly identical main content — can be treated as duplicates. In that case, canonicalisation and hreflang need to be planned together, with the canonical kept in the same language or the best available substitute language.
hreflang: link the language versions correctly
hreflang helps Google understand that specific URLs are language or regional versions of corresponding content.
It isn't used to "protect" a sound translation from duplicate content. It helps Google link the different language versions and choose the most suitable one for the user.
A simple EL + EN example:
<link rel="alternate" hreflang="el" href="https://example.com/hotel" />
<link rel="alternate" hreflang="en" href="https://example.com/en/hotel" />
The same set of references has to exist on both pages. That is:
- the Greek page references the Greek and the English,
- the English page references the Greek and the English.
The references have to be bidirectional: the related language versions point to each other. Each page also has to declare itself. The addresses inside hreflang have to be full URLs.
Language and region codes
hreflang can declare just a language or a language together with a specific region. For example:
el→ Greek,en→ English in general,en-GB→ English for the United Kingdom,en-US→ English for the United States.
We don't use a region code alone, like GB, without a language code.
When several versions of the same language exist for different markets, it can be useful to also have a general language version. For example:
<link rel="alternate" hreflang="en" href="https://example.com/en/" />
<link rel="alternate" hreflang="en-GB" href="https://example.com/en-gb/" />
<link rel="alternate" hreflang="en-US" href="https://example.com/en-us/" />
But you don't need to create separate versions for different countries if there are no real differences in the market, the content, or the experience you offer.
What about x-default?
x-default declares the fallback URL when none of the specific language or regional versions is a suitable match. It can point to a language or country selector, a general page, or a version that has been chosen as the default.
For example:
<link rel="alternate" hreflang="el" href="https://example.com/el/" />
<link rel="alternate" hreflang="en" href="https://example.com/en/" />
<link rel="alternate" hreflang="x-default" href="https://example.com/" />
It isn't mandatory on every multilingual website. Google recommends it as a fallback when none of the specific language or regional versions is a suitable match. Use it when there's a real reason to declare a fallback URL for users who don't match the other versions.
Does Google determine language from hreflang?
No.
Google determines a page's language from the visible content itself. hreflang isn't the mechanism it uses to decide which language a page is in. It does a different job: it links the different language pages to each other. That's why each page really has to be written clearly in the language it serves.
Translation and market adaptation — not word-for-word copying
On a multilingual site, not every text needs to be written from scratch. Sound translation is a perfectly normal part of the process. The issue is quality and adaptation.
There's a difference between a sound translation and a word-for-word rendering that carries the syntax and expressions of the first language into the second.
For example:
❌ EL: "Το ξενοδοχείο μας είναι ιδανικά τοποθετημένο για να απολαύσετε μια αξέχαστη εμπειρία διαμονής." ❌ EN: "Our hotel is ideally located to enjoy an unforgettable stay experience."
A more natural version could be:
✔ EL: "Είμαστε πέντε λεπτά με τα πόδια από την Οία, σε ήσυχο σημείο με θέα στην καλντέρα." ✔ EN: "A five-minute walk from Oia, in a quiet location overlooking the caldera."
What matters isn't whether the English text started as a translation or was written from a blank page. What matters is that the final result is natural, accurate, suitable for the specific audience, and useful for the market it addresses.
Adaptation isn't only about language
Two markets can have different questions and different selection criteria. A Greek and a foreign traveller may care about different information before they book. The same happens in B2B and exports.
That's why we don't mechanically translate:
- search words and phrases,
- frequently asked questions,
- titles,
- calls to action,
- examples,
- or real user searches.
We check how each audience really searches for and evaluates the service.
The right logic is: translate where that's enough, adapt to the market where needed, and differentiate more when the audience's need genuinely changes.
Structured data and language
Each language page's structured data has to agree with its real visible content.
If a page is in Greek:
- the title in the structured data should agree with the Greek title,
- the description should correspond to the Greek page,
- and any FAQs described should be the ones that actually appear on that version.
The same applies to the English page.
We don't mechanically copy the JSON-LD from one language to another, leaving titles or descriptions in a different language.
And inLanguage?
inLanguage is a Schema.org property for declaring content language on CreativeWork types such as Article and WebPage, using language codes based on BCP 47. It shouldn't be treated as a mandatory general field for every element on the page.
The key is that the structured data: uses the appropriate type, describes real content, and agrees with what the user sees.
Which schema types fit each case, we develop in the schema markup guide.
How does all this connect to search and AI tools?
We shouldn't treat Google Search, ChatGPT Search, and other AI tools as if they worked in one common way across all languages.
Google Search
For Google, the multilingual structure rests first on the known principles of international SEO: separate URLs, real content per language, correct canonical, correct hreflang, and a clean structure.
For Google's AI features, the important pages first have to be indexed and eligible to appear in Google Search with a snippet. Since 31 August 2026, the Search generative AI control has been available worldwide in Search Console, so also make sure the site is included in those features. Even when these conditions are met, Google doesn't guarantee that a page will appear or be used in a specific AI feature.
ChatGPT search
ChatGPT Search works differently. If you want your site's public content to be available for normal inclusion in ChatGPT Search, allow OAI-SearchBot in robots.txt and make sure the host, CDN or firewall doesn't block OpenAI's published searchbot IP ranges. That doesn't guarantee that a page will be used or mentioned. OpenAI says sites that block OAI-SearchBot won't be shown normally in ChatGPT Search answers, although a URL may still appear as a navigational link.
And what about the query's language?
We shouldn't assume that an AI tool will use only sources in the language of the query. There's no common public rule that explains how each platform selects sources by language.
So we don't build the strategy on assumptions like "the AI always prefers content in the query's language" or "English content always has a greater chance of being used".
We plan something more stable: that each real audience has a correct, accessible, and useful version of the information it needs.
What are the common mistakes?
The first mistake is having multiple languages on the same URL and changing the content only according to the visitor's language settings. Google recommends different, stable URLs for each language version.
The second mistake is automatic redirects without a real user choice. If you guess the visitor's language wrong and force them elsewhere, you may hide the version they actually want. Better to have a clear, visible language-switch option.
The third mistake is wrong canonical. A real English version shouldn't be mechanically declared canonical pointing to the Greek one just because the two pages describe the same service.
The fourth mistake is incomplete hreflang. It isn't enough for the Greek version to reference only the English. Each page has to correctly include the corresponding language versions and declare itself too.
The fifth mistake is wrong language or region codes. We use correct language codes and, when there's real targeting of a specific country or region, the correct combination, like en-GB or en-US.
The sixth mistake is treating x-default as mandatory. It's used when there's a real need for a fallback URL for visitors who don't match the other languages or regions.
The seventh mistake is believing that hreflang prevents duplicate content. That isn't its function.
The eighth mistake is low-quality translation. Translation isn't a problem in itself. The problem is text that's unnatural, inaccurate, or unsuitable for the second audience.
The ninth mistake is mechanically translating search words. How and why a user searches can differ by market and language. One audience may use other expressions, different place names, or care about different information.
The tenth mistake is structured data that doesn't agree with the language and real content of the page.
The eleventh mistake is treating hreflang, schema, or any other technical setting as a guarantee of a presence in AI tools. It isn't.
Where do you start?
Start from a simple question:
which languages and markets do you have a real audience in?
You don't need five languages just because "it looks more international". You need the languages that serve real customers.
If you're starting with Greek and English:
1. Create separate URLs
For each important page, the recommended base is to have a stable URL for the Greek and for the English version.
2. Check the canonical
Each genuinely translated version should be able to stand as a separate, indexable page and will usually use a self-referencing canonical. Don't mechanically declare all languages as canonical pointing to one shared version. For very similar regional variants in the same language, review canonicalisation together with hreflang.
3. Set up correct hreflang
Each language version should: reference itself, reference the corresponding other versions, and the references should be bidirectional. Use language and region codes correctly. x-default is added only when there's a real need for a fallback URL.
4. Translate and adapt to the market properly
Don't translate mechanically, word for word. Look at which information, questions, and expressions genuinely make sense for each market.
5. Check the structured data
The structured data has to agree with the real content of each language page. inLanguage is used where it's appropriate for the specific content type.
6. Check whether the content is accessible to Google and ChatGPT
For Google: check that the important pages are indexed and eligible to appear in Search with a snippet, and that the site is included in the Search generative AI control in Search Console, which has been available worldwide since 31 August 2026. For ChatGPT Search: allow OAI-SearchBot in robots.txt and make sure the host, CDN or firewall doesn't block OpenAI's published searchbot IP ranges.
7. Measure each market separately
Don't look only at total traffic. Check: which language brings visits, which landing pages perform, which searches are used, which market generates prospects or sales, and whether behaviour differs substantially by language or region.
Where Praion comes in
You can organise the first steps yourself. Clarify which languages and markets you really need. See whether separate URLs exist. Check whether the core translations are natural and useful.
When a multilingual website has to work systematically, though, it needs an overall design: organising the information and pages, the URL structure, correct canonical, correct hreflang, targeting the right languages and markets, translating and adapting the content, understanding how each market searches, structured data, technical accessibility to search engines and AI tools, and measurement by language and market.
That's the logic of multilingual AEO. Not creating "one Greek page and four copies of it", but giving each real audience the correct language — and where needed, local — version, without losing the company's shared identity and business logic.
See our approach, the schema markup guide, or, if you target an international tourism audience, AEO for hotels and tourism.
If you don't yet have a clear picture of AEO, start with what AEO is.
Multilingual AEO can improve how clearly and correctly a website's different language and regional versions are organised. It doesn't guarantee appearance, mention, recommendation, or ranking in Google or in AI tools. How each platform discovers, selects, and uses content can differ and change over time.