In brief
Cloudflare’s 16 October update improves Chinese, Japanese and Korean keyword queries in AI Search. Here is what bilingual businesses should check before adopting it.
A multilingual search update
Cloudflare announced an update to AI Search on 16 October 2026 that addresses a practical issue for multilingual websites: keyword queries in Chinese, Japanese and Korean. For businesses serving customers in more than one language, the release offers a useful reason to review how their own website search behaves.
What Cloudflare announced
The change applies to AI Search keyword search configured with its trigram tokenizer. Cloudflare says it now matches queries in these languages, including full natural-language questions. Previously, queries without spaces could be treated as one exact phrase and return no keyword results. See the official announcement dated 16 October.
This is a specific search-product update. It does not automatically upgrade every website using Cloudflare, translate English content into Chinese, or establish that an AI assistant will answer every multilingual question correctly.
Why bilingual businesses should pay attention
Our practical takeaway for Malaysian SMEs is to test the customer journey beyond the language selector. A website may display English and Chinese pages neatly while its search box still struggles with how customers actually ask questions.
Imagine a visitor looking for delivery information or a staff member searching an internal procedure. If the answer exists but cannot be found, they may contact the team again or leave the task unfinished. These are examples of workflows worth testing, not results measured or promised by Cloudflare.
For a bilingual catalogue, support centre or company knowledge base, start with a small set of genuine questions. Include complete sentences, short product names and the mixed English-Chinese wording customers use. Decide which document should appear for each query before evaluating a new search setup.
Configuration and limits matter
Cloudflare’s documentation recommends selecting trigram for Chinese, Japanese and Korean content instead of the default porter tokenizer. Trigram matches overlapping three-character sequences; terms shorter than three characters cannot match independently. Changing the tokenizer triggers a full reindex. Keyword search can also be combined with vector search in a hybrid configuration. These details are covered in the keyword-search documentation.
For a development team, that means planning a controlled test rather than switching a live search configuration and assuming the outcome. Compare English and Chinese results before and after the change, check short names separately, and confirm that the relevant content has been indexed. Review account limits and costs before choosing a production setup.
MIRA CODE’s perspective
We see this as a useful, narrowly scoped improvement to investigate when planning bilingual websites and knowledge-search tools. The business goal should remain straightforward: help customers and staff reach the right information with less effort.
A sensible next step is a search-quality checklist for the existing website, followed by a small technical trial where appropriate. There is no need to replace a working system simply because a feature is new. Choose the approach that fits the content, users and maintenance capacity of the business.
