In its Q3 2026 shareholder letter, Roblox, the game creation and distribution platform, said that more than 80% of the top 100 experiences by total playtime over the past month used AI to streamline development. It also said that more than 70,000 games have been published through Build, a tool that generates games from text prompts, since the alpha launched.
These figures measure different things, however. The first shows how widely AI is used in the production of popular games. The second shows how many new games have been published through Build. In a separate evaluation example the company published for Build, a game scored 98 for technical quality but only 31 for game design. As AI lowers the barrier to making games, the example highlights that building a game that works and building one players want to return to are separate challenges.
What the 80%+ and 70,000 Figures Actually Count
First, the top 100 experiences were selected by total playtime. According to Roblox's Q3 2026 shareholder letter (PDF, 1 page), more than 80% of them used AI over the past month to streamline development. This means AI was used in some part of the production process; it does not mean that more than 80% of top games were made almost entirely automatically.
The Build figures, on the other hand, come from an alpha running in four countries: New Zealand, Singapore, Serbia, and Thailand. According to the company, more than 500,000 people have used it since launch, and more than 70,000 games have been published. This is a cumulative total from a trial limited to certain regions, not a measure of overall creation activity on Roblox.
Roblox also highlights two games, "Break and Steal an Egg" and "Huss Valley," as examples built almost entirely with AI. Both reportedly ranked among the top 20 by concurrent players in Q3 2026. However, that ranking is based on the number of people playing at the same moment, a different criterion from the top 100 selected by total playtime.
The main figures from the shareholder letter are summarized below.
| Figure | What it covers and how it is measured | What it shows |
|---|---|---|
| 80%+ | Share of the top 100 experiences by total playtime over the past month that used AI to streamline development | How widespread AI assistance is in producing popular games |
| 70,000+ | Cumulative games published since the Build alpha launched in four countries | The volume of games supplied through Build |
| 2 games in the top 20 | Q3 ranking by concurrent players of two games said to be built almost entirely with AI | Examples of AI-assisted games drawing large numbers of concurrent players |
In this way, 80%+, 70,000+, and top 20 represent, respectively, AI usage rate, number of published games, and rank by concurrent players. Because what each measures differs, combining them cannot yield the probability that a game published through Build becomes a hit. Nor is it clear which stages humans were involved in, or how much they edited, for the two games that reached the top 20. These results alone do not support the conclusion that AI produces hits, nor do they provide grounds for judging the quality of the 70,000-plus published games as a whole.
An Official Example: 98 for Technical Quality, 31 for Game Design
How should a game's technical polish be weighed against how enjoyable it is to play? One clue comes from Roblox's official article on how it evaluates Build, published on October 8. Build is not simply a tool that converts text into code. It defines a game specification from the user's instructions, generates code and assets, and combines game mechanics, user interface (UI), controls, and more to construct a finished game.
It is difficult to capture the quality of a game with so many components in a single score. So Roblox uses a system that not only checks whether the specified requirements are met but also evaluates static quality such as code, behavior during actual play, and visual and audio presentation separately. This is meant to verify, from different angles, whether what was requested was implemented, whether the game can be operated without problems, and whether it is appealing as a game.
The official article uses an obstacle-course game, "Skyforge Ascent," as a generated example. In the company's internal automated evaluation, the game scored 98 out of 100 for technical quality but only 31 for game design. In other words, it was rated highly on technical grounds, but judged to have considerable room for improvement as a play experience.
According to Roblox, the game lacked mechanisms that make players want to play repeatedly, as well as a design in which the experience deepens as the game progresses. Meanwhile, in a test based on actual play, it earned a perfect 100, which also showed that checking whether a game runs is not enough to detect game design problems. Resolving the technical issues that keep a game running does not automatically produce a mechanism that encourages players to try again.
In game development, testing whether generated code works correctly and evaluating why players keep coming back play different roles. If a game is judged complete simply because it scored high on technical measures, problems remaining in its play design could be overlooked. Roblox's use of multiple evaluation axes is likely intended to capture these differences.
That said, the 98 and 31 are neither ratings from ordinary players nor results measured by a third party. They are internal automated evaluations of a single generated example that Roblox itself published, and the two scores measure different qualities. It would not be appropriate to interpret the gap as the average performance difference across Build, or as an objective quantification of how fun a game is. Even so, it is a case that shows the need to evaluate technical completion and playability separately.
As Games Multiply, Discovery and Retention Matter More
Even if AI makes games easier to build, they cannot be called a success unless players find them and keep playing. In its shareholder letter, Roblox also says that as the supply of content grows, it becomes harder to capture user attention and create connections between people. The idea is that in an era when games can be produced in large numbers, what matters is not only the performance of creation tools but also the systems that deliver those games to the right players.
Roblox cites existing connections between users and the behavior of inviting friends to games as its strengths in distribution. Besides searching for games on their own, players have a path to join new games in order to play with friends. Linking game discovery to personal relationships is a characteristic of the company's platform. As AI increases the number of games, such mechanisms are expected to become even more important.
Indeed, the company's recommendation system actively surfaces new games. According to Roblox, about 40% of the games in the top 100 by impressions in its weekly recommended-for-you feature, RFY, had been created within the past three months. However, this figure is the share of new games among the 100 with the most recommendation impressions. It does not represent the share of new games among all published games, nor the rate of AI usage.
Also, a new game appearing in recommendations is a separate matter from players continuing to play it afterward. Roblox says it will improve its recommendation system to emphasize long-term retention. The challenge is whether it can guide players not merely to games that draw them in, but to games they keep using. To evaluate the spread of AI-assisted creation as a success, usage data needs to be checked as well as the number of published games.
Andy Chalk of the games outlet PC Gamer has voiced concern about an industry direction that treats development speed and game count as key measures of success. Still, the fact that more than 70,000 games have been published does not by itself mean that most of them are low quality. What matters is whether players can find games they enjoy among the growing number of titles, and whether those games are appealing enough to keep playing. Judging the success of Roblox's AI strategy requires separately examining the supply of games and their actual usage.
Distribution Beyond Roblox and What Lies Ahead
Besides combining AI-assisted creation with its recommendation system, Roblox is pursuing plans to extend where its games are distributed beyond the platform. One is "Roblox Everywhere." The plan is to distribute Roblox games as standalone apps on external stores, with a first phase targeting selected top games expected to begin in the first half of 2027. Not all creators will be able to use external distribution right away; the rollout will start with a limited set of games.
Conversely, "Universal Player" is a concept for bringing outside games into Roblox. It aims to use AI to make existing PC, console, and mobile games playable on Roblox as well. Whereas Roblox Everywhere widens where Roblox games can be distributed, Universal Player widens the sources of games that can be brought onto the platform. However, specific conditions for Universal Player, such as when it will be generally available, which titles will be supported, and how rights will be handled, have not yet been disclosed.
If Roblox Everywhere actually launches, the key question will be whether games that gained popularity within Roblox can attract players on external stores as well. For Universal Player, which games can be brought in and under what conditions will help determine how feasible the concept is. Both are still at the planning stage and should be considered separately from actual distribution results.
The spread of AI-assisted game creation may reduce the burden of making games and widen opportunities for people who previously could not take part. But as the Build evaluation example shows, being able to make a game that works technically does not guarantee an appealing one. Whether Roblox can turn the growing number of games into sustained player engagement, and whether it can also succeed with distribution on external stores, will be important measures of how well its AI-driven game creation and distribution strategy works.