Websites Covering a Topic in 2 Languages Get Nearly Twice the AI Citations

We analyzed 1.1 million AI citations across Google AI Overviews, ChatGPT, and Claude.

When someone asks an AI assistant a question, the answer usually comes with sources: links to the websites the AI used to build its reply. These are AI citations, and for a growing number of searches, they’re the new page one.

So we wanted to know whether speaking your customers’ language helps you earn them. With AI search agency Ellipsis, we analyzed 1,136,972 AI citations across Google AI Overviews, ChatGPT, and Claude, asking the same questions in English and Spanish in Spain and Mexico. Then we checked whether each website covered the topic in both languages.

The short answer is that it helps a lot, as long as the content is there in both languages.

Key Findings

• Websites covering a topic in both English and Spanish had a 98% higher AI citation rate than websites covering it in only 1 language. That’s nearly 2x, on average across all 3 platforms.
• ChatGPT showed the biggest gap, citing bilingual coverage almost 3x as often.
• A multilingual website with thin translation coverage had no overall advantage. The topic itself has to exist in both languages.
• AI strongly prefers sources in the question’s language: 95% of links from Spanish questions led to Spanish-language content.

Covering a Topic in 2 Languages Comes With Nearly Double the Citation Rate

To set a baseline, we looked at how often websites with a topic in only 1 language were cited at all, in either English or Spanish. The answer was not very often: 11.6% on average across the 3 platforms.

When a website covered the same topic in both English and Spanish, that rate rose by 98%. The boost varied by platform:

• ChatGPT: almost 3x higher
• Google AI Overviews: 1.8x higher
• Claude: 1.6x higher

A Multilingual Website Isn’t the Same as Multilingual Coverage

If you remember one finding from this study, make it this one. Websites with a multilingual setup but only a handful of translated pages had, overall, no advantage in AI search.

ChatGPT won’t cite you on a topic because your website has a second language. It cites you because you answer that specific question, in that language. A language switcher on your homepage is a good start, but AI looks at what’s behind it.

AI Prefers Content in the Question’s Language

We also checked the language of the pages AI linked to. We read the page text itself rather than guessing from the domain, subdomain, or subdirectory, since a .com can host Spanish content as easily as a country domain can host English.

Among links where we could identify the content language, 95% of links from Spanish questions led to Spanish-language pages. For English questions, 77% led to English-language pages.

That might sound obvious, but it’s an important data point. Large language models (LLMs) are multilingual, yet they still clearly favor sources in the reader’s own language.

The English questions gave us a useful cross-check: websites covering the topic in both languages were cited 19.7% of the time, compared with 3.1% for websites with Spanish content and no English version.

More surprisingly, bilingual coverage helped even when language match wasn’t the difference. On ChatGPT, for questions asked in Spanish, websites that also had English coverage were cited 2.5x as often as Spanish-only websites (21.8% versus 8.7%), although both groups had the Spanish content.

How This Compares With Our 2025 Study

Our 2025 study, also run by Ellipsis on the same websites, covered Google AI Overviews and ChatGPT. Its headline figures came from Google AI Overviews: comparing websites labeled as translated or untranslated, it found up to 327% higher visibility for second-language searches and 24% more citations per question for translated websites.

This year, that website-wide label didn’t tell the same story on its own. It still showed a clear positive link on Claude, but no clear advantage on Google AI Overviews or ChatGPT. So we asked a more specific question: does the website cover the relevant topic in both languages? That’s where the nearly 2x gap appeared.

Our read is that AI search platforms are getting better at judging which content deserves a citation. The takeaway has sharpened too: having a multilingual website still matters, but what counts is making the content your customers need available in their languages.

What This Means for Your Multilingual GEO Strategy

Generative engine optimization (GEO) is the practice of making your content easy for AI platforms to find and cite. Based on this study, multilingual GEO comes down to depth. A website that’s mostly in 1 language with a few translated pages won’t see the benefit, while full coverage of the topics you want to be cited for will.

A practical way to start:

1. List the questions customers ask about your products or services.
2. Check which pages answering them already exist in each of your languages.
3. Fill the gaps on your priority topics first.
4. Track whether AI platforms start citing those pages.

That way, your translation work goes to the content that earns citations, rather than to having a second language for its own sake.

Weglot can help with the translation side. Our AI translations learn your tone, glossary, and messaging, and you can review and refine them before they go live, so every new language still sounds like you. Our built-in multilingual SEO (translated metadata, hreflang tags, and language-specific URLs) then helps each translated page get discovered.

Ready to cover your key topics in your customers’ languages? Start your free trial.

How We Built the Dataset

• Websites: the same panel of websites in Spain and Mexico used in our 2025 study.
• Prompts: we took each website’s ranking keywords, which point to topics it was likely to perform well on, and turned each one into equivalent questions in Spanish and English. That gave us 11,664 test prompts.
• Platforms: we ran every prompt through Google AI Overviews, ChatGPT, and Claude, keeping the topic and country consistent between languages.
• Results: 116,574 usable responses containing 1,136,972 citations.
• Analysis: we checked whether each website had relevant content answering each question in both languages, not whether it offered a language switcher. We compared websites on the same questions, counted each website once per response, and gave each AI platform equal weight.

This study was researched exclusively for Weglot by AI search agency Ellipsis, who also ran our 2025 study.