Internationalization support for AI Search

  • Release version: Yokohama
  • Updated May 21, 2026
  • 6 minutes to read
    • Summarize
    Summarized using AI
    This content was generated using new OpenAI-powered functionality. Results are provided on an as is basis and are not guaranteed to be accurate or complete.

    Summary of Internationalization support for AI Search

    AI Search in ServiceNow supports indexing and searching in multiple languages offered by the ServiceNow AI Platform®. This internationalization support enables customers to perform accurate search operations across diverse language content automatically, without requiring manual configuration. Supported languages include Arabic, Brazilian Portuguese, Czech, Danish, Dutch, English, Finnish, French (including French - Canada), German, Hebrew, Hungarian, Italian, Japanese, Korean, Norwegian (Bokmål), Polish, Portuguese, Russian, Simplified and Traditional Chinese, Spanish, Swedish, and Turkish.

    Show full answer Show less

    After activating a new language plugin, customers must reindex all relevant content to enable search functionality in that language.

    Indexing and Tokenization Behavior

    AI Search uses language-specific tokenization settings to process text segmentation differently for each supported language. For example, while most languages use spaces and punctuation for tokenization, languages like Chinese and Japanese require contextual interpretation for accurate word and sentence breaks. Additionally, Japanese text recognizes the nakaguro (middle dot) as a word separator.

    Language identification is performed automatically during indexing for Task table records, non-Task records (using the record’s language or instance default), and external documents. Content in unsupported languages defaults to English tokenization.

    AI Search can identify and index embedded regions of Japanese, Simplified Chinese, or Traditional Chinese text within documents primarily in other languages, applying the appropriate tokenization for those regions.

    Search Query Processing

    Search queries are tokenized based on the current user's session language in the ServiceNow AI Platform. However, terms in Japanese, Simplified Chinese, or Traditional Chinese within queries are processed using their respective language tokenization regardless of the session language, ensuring precise matching across multilingual queries.

    Matching between search queries and indexed content is most reliable when both are in the same language, as differing tokenization can lead to unpredictable results.

    Language-Dependent Search Features

    • Genius Results: Evaluated only when NLU models share the query language; LLM or heuristic models have language support restrictions.
    • Language Identification and Tokenization: Applied consistently across supported languages for indexing and queries.
    • Lemma and Unicode Normalization: AI Search performs language-specific normalization and decompounding for certain languages to improve search accuracy.
    • Result Improvement Rules: Activated only if matching the search query language or set to All Languages.
    • Stop Words and Synonyms: Applied based on dictionaries matching the search query language.
    • Typo Handling: Supported for most languages except Japanese, Korean, and Chinese variants.

    Managing Multilingual Content and Search

    ServiceNow customers can enhance multilingual search recall by assigning Knowledge article search languages by user country, allowing user searches to match articles in multiple languages relevant to their location and session language.

    Additionally, globally searchable knowledge articles can be configured with encoded queries for multilingual accessibility. Customers can also control facet translations for Knowledge article references, choosing whether to display them in the user's session language or the instance default language.

    Fallback Language Support

    AI Search supports fallback languages for user sessions, enabling searches across both the session language and the fallback language. This includes use of appropriate stop word, synonym, and typo dictionaries for the fallback language. Customers can configure a global fallback locale to extend search coverage across translated content effectively.

    AI Search supports indexing and search in all languages offered by the ServiceNow AI Platform®. Search linguistic features are supported in Arabic, Brazilian Portuguese, Czech, Danish, Dutch, English, Finnish, French, French - Canada, German, Hebrew, Hungarian, Italian, Japanese, Korean, Norwegian (Bokmål), Polish, Portuguese, Russian, Simplified Chinese, Spanish, Swedish, Traditional Chinese, and Turkish.

    Internationalization support is automatically enabled and isn't configurable.

    To view the full list of languages offered as ServiceNow AI Platform plugins and supported in AI Search, see Activate a language.

    Note:
    After you activate a new language plugin, you must reindex all indexed source content that you want to make searchable in the new language. For details on reindexing, see Perform a full table index or reindex for a single AI Search indexed source.

    Language settings determine how AI Search separates the text of indexed content and search queries into individual terms. This process, called tokenization, is handled differently for each supported language, using language-specific settings. For example, most languages use spaces and punctuation to separate words and sentences, but when tokenizing Chinese or Japanese text, AI Search instead uses contextual interpretation to correctly identify word and sentence breaks. When tokenizing Japanese text, AI Search additionally recognizes the nakaguro (middle dot) as a word separator.

    Note:
    If you indexed content in any of the following languages before the specified date, reindex that content to benefit from improved tokenization and feature support added in the AI Search back end.
    Table 1. Languages requiring reindexing
    Date Languages to reindex
    August 2024 Brazilian Portuguese, Dutch, Italian, Japanese, Korean, Portuguese, Swedish
    August 2025 Finnish
    December 2025 Polish
    June 2026 Arabic, Czech, Danish, Hebrew, Hungarian, Norwegian (Bokmål), Russian, Turkish

    Indexing behavior in supported languages

    When indexing content and metadata from a ServiceNow AI Platform source record or an external document, AI Search uses tokenization settings for the language of the record or document, as shown in the following table.

    Table 2. Indexing behavior by source
    Record or document Tokenization settings
    Source record from the Task [task] table or one of its child tables AI Search performs language identification and uses tokenization settings for the detected language to index the record's content and metadata.
    Note:
    Language identification only identifies Arabic, Brazilian Portuguese, Czech, Danish, Dutch, English, Finnish, French, French - Canada, German, Hebrew, Hungarian, Italian, Japanese, Korean, Norwegian (Bokmål), Polish, Portuguese, Russian, Simplified Chinese, Spanish, Swedish, Traditional Chinese, and Turkish. Content in other languages is identified and treated as English.
    Source record from a non-Task table AI Search uses tokenization settings for the record's language to index its content and metadata.

    If the record has no language specified, the ServiceNow AI Platform treats it as being in the instance's default language. In an English instance, for example, AI Search indexes records without specified languages using tokenization settings for English.
    External document AI Search performs language identification and uses tokenization settings for the detected language to index the document's content and metadata.
    Note:
    Language identification only identifies Arabic, Brazilian Portuguese, Czech, Danish, Dutch, English, Finnish, French, French - Canada, German, Hebrew, Hungarian, Italian, Japanese, Korean, Norwegian (Bokmål), Polish, Portuguese, Russian, Simplified Chinese, Spanish, Swedish, Traditional Chinese, and Turkish. Content in other languages is identified and treated as English.
    Note:
    When indexing content and metadata, AI Search recognizes regions of Japanese, Simplified Chinese, and Traditional Chinese text embedded within text in other languages. These text regions are indexed with the appropriate language tokenization settings regardless of the surrounding text's language. As an example, suppose you index an English-language knowledge article that includes a paragraph of Simplified Chinese. AI Search indexes this paragraph's content as Simplified Chinese and the rest of the record's content as English.

    Search query behavior in supported languages

    When processing search query text, AI Search uses tokenization settings for the language of the current user's ServiceNow AI Platform session.

    Note:
    AI Search recognizes Japanese, Simplified Chinese, and Traditional Chinese terms in search queries. These terms are processed with the appropriate language tokenization settings regardless of the user session's language. As an example, if a user in a French user session searches for remplacement ordinateur 笔记本电脑, AI Search applies Simplified Chinese settings for the 笔记本电脑 term and French settings for the other search terms.

    AI Search compares your search query terms with terms from indexed content and metadata, returning search results for indexed records or documents that contain matches. When your search terms are in the same language as the indexed terms, AI Search processes both sets of terms with the same tokenization settings, producing predictable matches and search results. If your search terms aren't in the same language as the indexed terms, AI Search processes the two sets of terms with different tokenization settings and matching may be unpredictable.

    Language dependence for search features

    The following search features are language-dependent and supported only for the listed languages.

    Table 3. Search feature language dependence
    Feature Language dependence and supported languages
    Genius Results AI Search only evaluates Genius Result configurations with NLU triggers if the linked NLU model has the same language as the search query.

    Genius Results configurations that use an LLM or a heuristic model, such as Now Assist Multi-Content Response or Q&A, are limited to languages supported by the model.

    Supported languages: Arabic, Brazilian Portuguese, Czech, Danish, Dutch, English, Finnish, French, French - Canada, German, Hebrew, Hungarian, Italian, Japanese, Korean, Norwegian (Bokmål), Polish, Portuguese, Russian, Simplified Chinese, Spanish, Swedish, Traditional Chinese, and Turkish.
    Language identification and tokenization During indexing, AI Search identifies supported languages in Task table records and external documents. Text processing for the indexed content uses tokenization settings for the identified language.

    Supported languages: Arabic, Brazilian Portuguese, Czech, Danish, Dutch, English, Finnish, French, French - Canada, German, Hebrew, Hungarian, Italian, Japanese, Korean, Norwegian (Bokmål), Polish, Portuguese, Russian, Simplified Chinese, Spanish, Swedish, Traditional Chinese, and Turkish.
    Lemma and Unicode normalization AI Search performs language-specific lemma normalization for terms in indexed content and search queries.
    Supported languages: Arabic, Brazilian Portuguese, Czech, Danish, Dutch, English, Finnish, French, French - Canada, German, Hebrew, Hungarian, Italian, Japanese, Korean, Norwegian (Bokmål), Polish, Portuguese, Russian, Simplified Chinese, Spanish, Swedish, Traditional Chinese, and Turkish.
    Note:
    For German, Danish, Hungarian, Korean, Norwegian (Bokmål), and Swedish, AI Search performs term decompounding in addition to lemma normalization. For Finnish, AI Search uses algorithmic stemming to identify lemmas.

    AI Search performs Unicode normalization for all terms in indexed content and search queries.

    For more information on normalization of lemmas and Unicode forms in indexed content and search queries, see Lemma and Unicode normalization.
    Result improvement rules AI Search only evaluates activation for result improvement rules that have the same language as the search query or that have All Languages specified.

    Supported languages: All languages activated in your instance. For the list of languages you can activate, see Activate a language.
    Stop words AI Search only considers stop words from dictionaries that have the same language as the search query.

    Supported languages: All languages activated in your instance. For the list of languages you can activate, see Activate a language.
    Synonyms AI Search only considers synonyms from dictionaries that have the same language as the search query.

    Supported languages: All languages activated in your instance. For the list of languages you can activate, see Activate a language.
    Typo handling AI Search derives a separate list of auto-correction terms for each supported language found in search source indexed content. Auto-correction only replaces search query terms with terms from the list that has the same language as the search query.

    Supported languages: Arabic, Brazilian Portuguese, Czech, Danish, Dutch, English, Finnish, French - Canada, French, German, Hebrew, Hungarian, Italian, Norwegian (Bokmål), Polish, Portuguese, Russian, Spanish, Swedish, and Turkish. Typo handling isn't supported for Japanese, Korean, Simplified Chinese, or Traditional Chinese.