# WeSearch now handles all WeChat Search traffic

> The retrieval system is fully deployed in WeChat Search. Its authors report more article clicks, lower resource costs and faster graphic-text retrieval.

Oossa · 2026-11-03 · https://oossa.com/en/wesearch-now-handles-all-wechat-search-traffic

WeSearch, a new retrieval system for WeChat Search, is now serving 100% of the service’s production traffic, its authors say. In a one-month, full-traffic A/B test, they report a 0.68% increase in click-through rate (CTR) and 2.27% more article clicks per user. The system also cut retrieval resource costs by 41% and reduced graphic-text retrieval latency by 19 milliseconds.

The system is designed to help WeChat Search find relevant material across collections ranging from hundreds of millions to billions of documents. Rather than assigning every document one fixed route through a hierarchy of topics, WeSearch first identifies groups of potentially relevant documents for a query, then ranks individual documents in those groups. The authors say this helps with documents that cover several topics, which can be a poor fit for a single fixed route.

## How does it fit into search?

WeSearch is the semantic retrieval path in WeChat Search, working alongside a lexical syntax tree. In simple terms, it helps find content by meaning, while the other component handles word and structure-based matching. The authors describe the two parts as operating together, rather than presenting WeSearch as a replacement for every part of search.

The framework also uses compact document representations and training that accounts for relevance. Its index can be updated incrementally, allowing the authors to refresh the model weekly, or daily when needed. That matters for a search service whose content keeps changing: updates need not wait for a complete rebuild of the index.

The authors report consistent gains in one-week A/B tests across collections of different sizes, but the abstract does not give the figures for those tests. It also does not specify how the CTR change was calculated or provide an independent evaluation. The reported one-month results therefore describe WeSearch’s measured performance within WeChat Search, not a general comparison with other search systems.

## The facts

- WeSearch is reported to serve 100% of WeChat Search production traffic.
- The authors report a 0.68% CTR increase and 2.27% more article clicks per user in a one-month full-traffic A/B test.
- The same test reported 41% lower retrieval resource costs and a 19-millisecond reduction in graphic-text retrieval latency.
- WeSearch can refresh weekly, or daily when needed, and was tested on collections ranging from hundreds of millions to billions of documents.

## Why it matters

For people searching in WeChat, the authors’ results suggest that the system may help surface articles people are more likely to click, while using fewer retrieval resources. The reported gains are specific to WeChat Search; the abstract does not show whether users will notice the changes in every search or how the CTR figure was calculated.

## Sources & references

1. [WeSearch: A Unified Model-Centric Retrieval Framework for WeChat Search](https://arxiv.org/abs/2611.00530) – arXiv cs.IR, 2026-11-03

Last updated: 2026-11-03