Where studios actually use AI — and where they don't
97% of studio leaders have used AI in some form during development. The interesting finding is which stages they use it in, and how badly they rate the integration.
Consultancy Vivrato interviewed 101 studio leaders across Europe and North America between April and September 2026. The headline number is that 97% have used AI in some form during development, and 94% have used it in at least one of nine production stages.
That's a lot less interesting than the breakdown. Here it is:
| Stage | Studios using AI |
|---|---|
| Code and engineering (engine, tools, gameplay code) | 86% |
| General productivity (docs, email, meetings, planning) | 83% |
| Concept and pre-production (ideation, brainstorming) | 79% |
| Design and prototyping (mechanics, systems, features) | 68% |
| Art, assets and animation (characters, environments, mocap) | 60% |
More than half of studios also use AI for QA and testing and for narrative and world building. The lowest adoption is in analytics and player data, and in LiveOps and monetisation.
The honest part
More than half of respondents rate their integration of AI into the full production pipeline at three or lower out of ten. European studios report deeper adoption than North American ones — a self-rated depth of 4.2 out of 10 against 3.5.
Vivrato founder and CEO Adam Boyes made the point that the framing has been too narrow:
We've noticed this year, public discourse and definitions about AI in games usually defaults to generative AI: chatbots, image generation, and synthetic voices. However, we know that studios use AI in a variety of ways, from machine learning that predicts player churn to rules-based automation in testing, with very different results.
The interesting question isn't around which game studios are using AI anymore. The focus is now where they're using it, what is changing, what's actually helping, and where teams are still running into challenges.
Why this slice matters
The public argument about AI in games is mostly about generated art and voices — the visible stuff. Vivrato's data says the heaviest real-world usage is in code, documentation and ideation, and that the art pipeline is actually the least AI-saturated of the five stages. The loudest argument and the actual adoption curve are not the same shape.
The other useful signal: a self-rated 4.2/10 for pipeline integration is not the score of a solved problem. It's the score of something being bolted on.
SOURCES
Every claim above is drawn from these reports.