AI use is close to universal in PC and console game development, but most studios are still only using it at the edges of their pipelines, according to new research from Vivrato.
The findings come from VAIA, a structured interview study of AI use in the games industry carried out between Q2 and Q3 2026. Researchers held 101 confidential conversations with senior leaders and decision makers, mostly at studios headquartered in North America and Europe. According to the snapshot, 97% of those interviewed had used AI in some form.
Of the full sample, 94% reported current production use in at least one of nine stages of the game development pipeline, the sequence of steps a studio moves through from early concept to a live game. Depth was another matter. Asked to rate how deeply AI was embedded across that pipeline, participants gave a median answer of 3/10.
Where studios are using AI
Code and engineering was the most commonly reported use at 86%, followed by general productivity work on 83% and concept and pre-production on 79%. Design and prototyping came next on 68%, with art, assets and animation at 60% and quality assurance (QA) and testing at 57%.
Reported use thinned out towards the commercial end of the pipeline. Narrative and worldbuilding appeared in 55% of interviews, analytics and player data in 49%, and LiveOps and monetisation, the ongoing updates and revenue work that keeps a released game running, in 37%. Vivrato noted that the stage rates were calculated against all 101 interviews, while the number of known answers varied by stage from 86 to 99.
Integration depth remains low
Participants rated their pipeline integration on a scale of zero to 10, and 52% put themselves at three or lower. Some 17% sat in the lowest band of zero to one, with a further 35% on two to three. Just under a third, 32%, landed in the middle band of four to six.
Deeper integration was rare by comparison. Nine% of those interviewed rated themselves at seven to eight and 7% at nine to 10, meaning fewer than one in six placed their studio in the top half of the scale. The depth figures were based on 100 of the 101 interviews.
There was also a regional split in the results. Leaders at European studios, including the UK, reported an average integration depth of 4.2/10, against 3.5/10 in North America. The regional bases were 46 interviews in Europe and 49 in North America, with the remaining six not broken out.
Behind the research
Vivrato described VAIA as a structured interview study rather than an online survey. Each conversation followed the same 44-question guide and covered the same nine production stages. Answers were anonymised and reported only in aggregate, with no individual or company identified.
Every interview also separated six types of AI, so that a studio’s view of generative AI was not treated as its view of AI overall. Those were generative AI, covering large language models (LLMs) and image and audio generation; classical machine learning, used for prediction, classification and clustering; rules-based automation such as scripted logic and decision trees; and retrieval-augmented (RAG) systems, which pull in existing documents or data to answer a query. The list was completed by fine-tuned or custom-trained models adapted on a studio’s own data, and what the snapshot called autonomous AI workflows, or agentic systems.
Vivrato is offering two paid follow-ups to the research for studios that want to go further. A studio report benchmarks a studio against similar-sized peers and includes up to one hour with the VAIA team, while an enterprise report covers up to four confidential team interviews, tailored analysis, recommendations and an executive readout.
