Japanese online game companies are all-in on AI

Every Japanese online game company covered by an annual industry survey is now using generative AI tools, according to the 2026 edition of a research report published by Japan's Online Game Association and Kadokawa ASCII Laboratories.

The JOGA Online Game Market Research Report 2026 has been produced annually since 2004 and tracks the state of Japan's domestic online game market alongside corporate trends among its developers. Recent editions have added questions on generative AI covering both companies and players. A preview of the findings published by Famitsu puts generative AI adoption among the surveyed companies at 100%.

The tooling breakdown is the more interesting figure.

• Google's Gemini was the most widely used model at 94%, followed by
• Anthropic's Claude at 84% and
• GitHub Copilot at 76%.

The percentages sum well past 100, indicating that most respondents run several models in parallel rather than standardizing on one vendor.

That represents a substantial reshuffle in a single year. In the 2025 edition of the same survey, the most-used tool was OpenAI's ChatGPT, at a 59% utilization rate. ChatGPT does not appear in the headline figures for 2026. Whether it has fallen out of the top three or simply out of the summary reporting is not clear from the preview.

The shift in what companies are using AI for is equally notable. In the previous year's survey, content planning ranked alongside user analysis as the most commonly cited application. In the 2026 edition, the tasks companies were most eager to hand to generative AI tools were user preference analysis and user behavior prediction — analytics functions rather than creative ones.

This direction of travel runs against the way generative AI in games is usually discussed. The pitch has generally been about content: art assets, dialogue, level layouts, NPC behavior.

What the JOGA respondents say they most want automated is the live-ops analytics layer; working out what players like and predicting what they will do next. For online and mobile operators running long-tail live services, that is where the commercial value sits, and it is a category where the output can be checked against real player data rather than judged on taste.

Players surveyed for the same report did not share the enthusiasm. Their most frequently voiced concerns about generative AI in games were potential copyright infringement, and the prospect of games converging on a single look.

The two concerns are related. Both describe the same failure mode from different angles: models trained on existing work reproducing that work, either closely enough to raise a legal question or loosely enough to flatten everything into an averaged house style. Neither concern maps onto what the Japanese companies say they are actually using the tools for, which suggests the visible disagreement is partly a mismatch of assumptions about where AI is being applied.

Nevertheless, the 100% figure needs framing against the survey's scope. JOGA's membership and the report's remit cover the online game market specifically, operators of mobile and browser titles, live-service businesses with running telemetry and ongoing content pipelines. It is not a broad cross-section of Japanese game development, and the segment it does cover is the one where analytics-oriented AI has the clearest business case.

Other surveys give lower numbers. The Computer Entertainment Supplier's Association found in September 2025 that 51% of Japanese game companies were using AI in some capacity, with the two most common applications being creative — visual asset and image generation, followed by story and text generation. A more recent survey of Japanese creative professionals in corporate roles found 59% of companies using AI, and of those, 71% did not actively disclose that they were doing so.

That non-disclosure figure is worth sitting with alongside the 100% headline. If roughly seven in ten companies using AI are not telling anyone, published adoption rates are measuring willingness to answer a survey question as much as they are measuring actual practice. It also means the player concerns recorded in the JOGA report are being formed largely without visibility into which games are using generative tools.

The preview reporting does not state the sample size for the 2026 survey, the response rate, or how "use" was defined for the purposes of the adoption question — whether a single engineer using Copilot counts the same as a studio running models in its production pipeline. The full report has been published by JOGA; the figures circulating so far come from Famitsu's preview of it.

Yet for anyone tracking AI adoption claims across the games sector, the JOGA numbers are most useful read as a vendor-share snapshot rather than an adoption headline.