Generative AI appears to have become a routine part of the workflow for a large share of game developers surveyed in Japan. A preliminary summary from the Computer Entertainment Supplier’s Association (CESA) reports that 85.8% of respondents said they actively use generative AI in their work. The survey drew responses from 1,349 developers at CEDEC, a Japanese game-developer conference.
That headline figure is striking beside the prior year’s reported result, when 51% of surveyed developers said they used the technology. But it does not mean that nearly nine out of 10 Japanese game companies are putting AI-generated art, dialogue, or code into shipped games. The survey measures individual respondents, and its definition of use covers a wide range of activity: generating assets as well as using tools such as ChatGPT and Copilot for general administrative work.
The distinction matters. A developer using an AI assistant to organize routine work is participating in generative-AI adoption, but that is fundamentally different from a studio incorporating generated visual material into a game. CESA’s separate company-focused findings paint a more qualified picture, particularly for creative production. Together, the two sets of figures show both a rapid expansion in day-to-day tool use and a business still working out where its boundaries should be.
What the 85.8% figure actually measures
The preliminary result concerns developers rather than companies. That makes it useful for understanding how commonly individual professionals encounter or use generative AI, but less useful as a clean count of studio-level adoption. Several respondents can come from the same employer, follow the same internal rules, and use the same approved tools. A company with a formal policy permitting one narrowly defined use could therefore be represented by many affirmative responses.
There is also no company-by-company breakdown within this individual survey. It cannot tell readers whether usage is spread evenly across the industry, concentrated at a subset of organizations, or shaped primarily by particular job disciplines. It does establish that the tools are familiar enough to be active workplace technology for a substantial portion of the respondents.
In practical terms, generative AI refers here to systems that produce or help produce material from prompts or other inputs. That material can include images and other assets, while tools such as ChatGPT and Copilot can also assist with broad office, communication, and task-oriented work. Grouping those uses together explains why the developer-level adoption number can be high even when use in sensitive parts of game production remains comparatively limited.
Frequency is notable as well. Among respondents using the technology, 63% said they used it daily. Another 22.8% said they used it occasionally. These figures suggest that, for many participants, generative AI is not merely something tested once out of curiosity. It has entered recurring work habits. Still, recurring use does not answer what work is being delegated to a tool, how much human review follows, or whether any output reaches players.
Productivity and creativity lead the reported benefits
The responses make a sharp distinction between internal gains and player-facing gains. Many developers said the technology reduced production time and costs. A total of 837 respondents said it expanded creativity, while 738 said it lowered technical barriers. Only 147 said it benefited the end-user experience.
That gap is one of the most important findings in the preliminary summary. It suggests developers are chiefly seeing generative AI as an aid to the process of making games, rather than as a direct feature or quality improvement that players readily experience.
Lowering technical barriers can be understood as making certain tasks more approachable for people who might otherwise need additional specialized knowledge or time. Expanding creativity, meanwhile, reflects the respondents’ view that these systems can support ideation or help them approach work differently. Neither phrase proves that an AI tool independently creates better games. They describe reported benefits to the people using it during development.
For players, that difference is consequential. Faster completion of internal tasks could potentially help teams manage production, but it is not automatically visible in a game’s writing, art direction, localization, stability, or design. The small number linking AI to end-user benefit does not establish that player experience is harmed; it does indicate that respondents were much more confident discussing workflow advantages than finished-game advantages.
This is also where policy and transparency become central. If studios use generative AI for internal administrative tasks, the player-facing stakes may be different than when it is used to create visual assets, story material, localization, quality assurance, or programming. Those areas can influence the game itself and raise different questions about consistency, oversight, creative authorship, and audience trust.
Company survey shows a narrower production picture
CESA’s separate survey of member companies provides an important counterweight to the individual-developer result. In that company sample, 74% said they had policies governing the use of generative AI. The reported split between firms not using generative AI and firms using it in some manner was nearly even: 27 companies said they were not using it, while 26 said they were.
Use for visual assets was especially limited in that company data. Twelve companies said they used generative AI for visual assets, and fewer said they used it for story, localization, quality assurance, or programming. The results do not provide enough detail to rank every department or explain each company’s rules, but they demonstrate that “AI use” is not a single, uniform production practice.
Quality assurance, often abbreviated as QA, is the work of checking a game for issues and verifying that it works as intended. Localization is the adaptation of a game for another language or market, commonly involving translation but also cultural and contextual adjustments. Both are areas where errors can directly affect players, which helps explain why companies may approach their use of generative systems differently from simple administrative assistance.
Formal policies are significant because they define more than whether a tool is allowed. They can establish who is authorized to use it, which tasks are acceptable, and what forms of oversight are expected. The preliminary results do not spell out the contents of individual companies’ policies, so it would be premature to assume they all impose identical restrictions. Yet the high share of companies reporting some form of policy signals that governance is becoming part of the adoption story.
Why asset generation remains particularly sensitive
Generated visual material has already attracted scrutiny in the Japanese games space. A presentation from Level-5, the developer associated with Dark Cloud and Professor Layton, drew fan suspicions that it included AI-generated images. That reaction illustrates the challenge facing companies: even when productivity arguments are compelling internally, the use of generated assets can become a public issue if audiences believe it is inconsistent with a project’s creative identity or with their expectations of how the work was made.
The available survey results do not establish what any particular studio will do next, nor do they prove why each firm has adopted, restricted, or avoided particular uses. They do show that visual assets are a much narrower category of company use than the broad 85.8% developer figure might imply.
That is why broad claims about AI in games need careful wording. Saying a developer uses an AI tool may refer to assistance with general work, not generated artwork in a release. Saying a company has an AI policy does not reveal whether the policy is permissive, restrictive, or primarily procedural. And saying that AI can reduce time or cost does not demonstrate an improvement in the final player experience.
Player response will shape the implementation debate
The preliminary findings arrive amid mixed feedback from players and ongoing online concern about generative AI. The data suggests support for the technology is stronger in Japan and South Korea than in some other markets, but international reaction can still shape how game companies present and deploy it. Backlash connected to Crazy Taxi: World Tour is a reminder that audience response can become part of the calculation, especially where players believe AI affects the creative work they are paying for.
For studios, the practical takeaway is not simply that adoption is rising. It is that the type of adoption matters. A tool used to assist a developer with routine work may be evaluated very differently from one used for visual assets, narrative material, localization, QA, or code. Clear internal rules can help establish review and responsibility, while clear public communication may become equally important when the output has a visible role in a finished game.
Player trust is already a wider industry concern, as shown by policy discussions around virtual-currency pricing and player protections. Generative AI is a separate issue, but it similarly places attention on whether companies explain consequential choices plainly and protect the quality of the experience people receive.
CESA is due to release its full report in December. Until then, the preliminary figures support a measured reading: generative AI has moved quickly into the everyday working lives of many of the developers surveyed, but studio-wide usage is more fragmented, and use in player-visible production areas is far from universal. The next question is less whether game developers will encounter these tools than how companies set limits, supervise their use, and demonstrate value to the people playing their games.
