What does the game universe look like by experience?
297 games. 30 attributes. No genre tags.
Find your next game →Color shows mean distance to all other games in the full 30-dimensional space — computed independently of the 2D layout. A game's color is meaningful. Its exact position is approximate.
Findings
What the data shows
A lot of AAA blockbusters are the same game experientially
The ten most “average” games in the library — closest to everything else in the space — are almost exactly what any experienced gamer would call prestige third-person action-adventure: Horizon Forbidden West, Star Wars Outlaws, Deathloop, Rise of the Tomb Raider, Avowed, Dragon Age: The Veilguard, Control.
The system found this cluster from 30 first-principles attributes, with no knowledge of their genre or theme labels. These games are experientially interchangeable: moderate combat, moderate exploration, moderate narrative, similar session structures. When recommendations within this cluster feel accurate but generic, the data explains why.
The hardest games to recommend are the most isolated points
The most isolated games — furthest from everything else in the space — are Animal Crossing: New Horizons, Unpacking, A Short Hike, Return to Monkey Island, Noita, Coffee Talk, Factorio.
These are also the games players most consistently report being unable to find good recommendations for. Their profiles are genuinely rare: near-zero tension, unusual cognitive styles, session structures unlike anything in the dense center. They occupy experiential territory the industry has barely explored — and the absence of neighbors around them is an accurate map of why those recommendations have always been hard to find.
Competitive multiplayer games form their own island
A distinct cluster sits apart from everything else: Rocket League, League of Legends, Guilty Gear: Strive, Marathon, Rematch. What unites them isn't genre — it's a specific experiential profile almost nothing else shares: human competition. Match-based sessions against other players. Analytical thinking to outwit the opponent. Tension driven by the desire to win and the fear of loss.
Methodology
How it works
#1 – 30 attributes scored by AI
All 297 games in the catalog are evaluated across 30 attributes describing what it demands from you as a player: how much sustained attention it requires, whether it tolerates being put down mid-session, tension level, session length before something feels satisfying, cognitive style of the core loop.
Attributes are scored using AI with calibrated reference anchors — closer to sentiment analysis applied to player experience than to genre classification. The scores capture what the game asks of you, not what it looks like from the outside.
#2 – Each game becomes a point in space
Scores are stored as a vector in pgvector. Similarity between two games is L2 distance — Euclidean distance in that 30-dimensional space. No black box: the distance between any two games is a number you can reason about.
The map above is a UMAP projection of that space into 2D. Color encodes each game's mean distance to all others in the full 30-dimensional space, computed independently of the 2D layout. Position is approximate. Color is meaningful.
#3 – Search finds games that make the same demands
Enter games you love and the system finds others with similar experiential profiles. For multiple inputs, candidates are ranked by their average distance to all your games — so results score highest when they're a close neighbor to each of your inputs, not just geometrically central between them.
The result is recommendations based on what playing actually feels like, not what the game looks like on a store page.