Discussion
  • AI Search

Do AI assistants cite the same page for a topic across languages, or does each language version compete separately? Any multilingual citation data?

Published 1 reply

Discussion starter: an AI search question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.

Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.

1 reply

  • Niraj Raut
    Niraj Raut
    Original poster

    There is no robust public data on how assistants choose between language versions of the same content, so any answer rests on how retrieval works and on limited observation. Assistants usually answer in the user’s language and tend to prefer sources in that language when good ones exist, which means each language version competes within its own language, much as it does in search. But they do not always stay within one language: for technical topics, niche subjects or questions where the best source exists only in English, assistants often draw on English sources and answer in the user’s language. For assistants that search the web, the language version retrieved depends on the searches the assistant runs and how each version ranks for them. The practical implication is that each language version needs to be strong in its own right, and facts need to be consistent across all of them.

    How language affects which page gets cited

    Situation Likely behaviour
    Strong local-language sources exist Local-language version more likely to be cited
    Only English sources are strong English sources cited, answer given in the user’s language
    Your localised version is thin or machine-translated May lose to local competitors or your own English page
    Search-based assistants Retrieval follows how each version ranks for the searches run in that language

    How to find out for your content

    1. Build a prompt set in each language you publish in, covering the same topics.
    2. Run prompts repeatedly across the assistants your audiences use.
    3. Record the language and URL of each cited source, including whether your English or localised version was cited.
    4. Compare citation rates by language and topic, and identify where your localised versions lose to local competitors or to English sources.

    What to do with the findings

    • Strengthen weak localised versions where they lose to local competitors: native-quality writing, local examples, local data.
    • Keep facts consistent across languages, so whichever version is cited, the answer is right.
    • Use hreflang correctly so search-based retrieval surfaces the right version for each language and market.
    • Localise the pages that matter most rather than translating everything, focusing on topics where local-language demand and competition justify it.

    Hypothetical example:

    A software company publishes documentation in English, German and Japanese. Prompt tracking shows that German prompts usually cite its German pages, while Japanese prompts often cite its English pages, and sometimes a Japanese community blog, because the Japanese documentation was machine-translated years ago and is out of date. The company updates the Japanese documentation with native writers and current screenshots, and checks hreflang. Over the following months, Japanese prompts cite the Japanese pages more often, and the community blog’s outdated instructions appear less frequently in answers.

    Watch for this:

    Inconsistent facts across language versions, such as different prices, different feature lists or outdated translations, mean assistants can give different answers in different languages. Users notice, and it undermines trust in the brand. Keep a single source of truth for facts and propagate changes to all versions.

    Global facts and local pages

    Some content, such as global research, company facts and product specifications, is the same in every language, and assistants may cite whichever version they find. Make sure each version states the facts identically, with dates, and that localised versions are updated at the same time as the original. For data-heavy content, consider publishing the core findings in each language even if the full report exists only in English, so assistants answering in other languages have a local-language source that attributes the data correctly.

    Language-specific AI platforms

    In some markets, locally popular assistants and search engines may rely on different sources and indexes from the global platforms. If a large share of your audience in a market uses a local assistant, include it in prompt tracking for that market. The fundamentals are the same: strong local-language content, consistent facts and presence in the sources those systems use.

    When English sources are fine

    For audiences comfortable reading English, or for highly technical topics where English is the working language, being cited through an English source may be perfectly acceptable. The investment in localisation should follow where local-language demand and competition make a difference, not an assumption that every language needs every page.

    Assume each language version competes within its language, strengthen localised versions where they lose to local sources, keep facts consistent across languages, use hreflang correctly, and track citations by language to see where localisation investment actually changes what assistants say.

    Need help with this on your own site?

    Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.

    See SEO services

Add a reply

Replies are open to approved members. Log in or apply to join.