← Cases
ChinaMedia and entertainment2023 – ongoing

Bilibili

In-house index model for search, moderation and video summaries

Bilibili said better AI understanding of content helped lift its click-and-conversion rate 19% in Q2 2026

In progressCompany's own claim

Problem

Bilibili needed to understand its videos and users well enough to improve search, content distribution and ad performance.

Approach

In July 2023 Bilibili released its own index model, built to understand its videos and users for internal uses such as search, content moderation and video summaries, rather than to compete as a general model. Management said it plans about RMB 1 billion in extra AI capex for 2026, mostly servers and computing power, focused on video understanding, distribution and creation.

Results

19%year-on-year rise in the combined click and conversion indicator (CTCVR) in Q2 2026, partly credited by management to AI understanding of content and user intentCompany's own claim[1]
doubledsearch ad revenue in Q2 2026 versus a year earlier, as reportedCompany's own claim[1]
about RMB 1 billionplanned extra AI capital expenditure for 2026 (a plan, not a result)Company's own claim[1]

Takeaway

Management tied the ad gains to AI content understanding only in part, and the report gives no separate measure of the model's effect.

Categories

Evidence
Company's own claim

Sources

Source summaries

[1]36Kr (AIX财经)15 Sep 2026

The article covers Bilibili's AI spending plans and the index model released in July 2023. It reports Q2 2026 search ad revenue doubling and CTCVR up 19%, partly credited by management to AI content understanding. It gives no measured result for moderation or summaries.

Full details are in the original sources.

Added 1 Nov 2026 · Updated 1 Nov 2026
Open case page ↗