Criteria & Method

What counts as AI-native, how the index is classified, and where the data comes from.

Why this exists

A new kind of game, with no shelf to sit on

Since 2022, a small but fast-growing body of games has made a language model part of the playable system itself — the witness you interrogate, the referee that rules on your spell, the dungeon master that improvises the world. Storefronts have no category for them; algorithm-driven discovery buries them.

They also share an unsolved economic problem: inference costs money per turn, so pricing has fractured into subscriptions, credits, API keys and local models. This index exists to make the field discoverable in one place — and to document, plainly, how each game charges.

Inclusion criteria

Three tests, from the survey

We adopt the operational definition of “AI-Native Games: A Survey and Roadmap” (arXiv:2607.00527). A game is AI-native when all three hold:

1
Generative AI is present at runtime
The game calls a generative model while you play. Enemy AI, pathfinding, finite-state NPCs, difficulty scaling and offline asset generation do not count.
2
The core loop depends on it
Model output directly interprets your actions, produces playable content or consequences, adjudicates semantic rules, drives essential characters, or evolves world state.
3
It is not trivially substitutable
The counterfactual test: replace the model with finite authored content or deterministic rules, and the core experience collapses or becomes a fundamentally different game.

The boundary matrix

Two questions separate four tiers. The index labels every entry with its tier; only AI-Native and AI-Boundary titles are catalogued.

	Strict game structure? — Yes	Strict game structure? — No
AI constitutive of the core loop? — Yes	AI-Native AI is constitutive of the core loop, inside a strict game structure. e.g. Vaudeville	AI-Boundary AI is central, but goals, rules, win/lose or state progression are thin. e.g. AI Dungeon
AI constitutive of the core loop? — No	AI-Augmented AI enriches a layer, but the loop survives its removal. Not indexed. e.g. Tenure-Track Chili	Out of scope No AI core and no game structure: plain chatbots, tavern role-play, decoration.

Strict game structure? — Yes

Strict game structure? — No

AI constitutive of the core loop? — Yes

AI-Native

AI is constitutive of the core loop, inside a strict game structure. e.g. Vaudeville

AI-Boundary

AI is central, but goals, rules, win/lose or state progression are thin. e.g. AI Dungeon

AI constitutive of the core loop? — No

AI-Augmented

AI enriches a layer, but the loop survives its removal. Not indexed. e.g. Tenure-Track Chili

Out of scope

No AI core and no game structure: plain chatbots, tavern role-play, decoration.

The dual-axis taxonomy

Each entry carries one G code (the form you play) and one N code (the AI function the loop cannot live without).

G-axis · Game form

G1
Narrative Adventure×23
A branching story or case advanced through free-form exchanges with characters or a narrator.
G2
RPG×9
Assume a character; quests, conflict and party play advance through role-played actions.
G3
Puzzle×7
Discrete language challenges — guessing, decoding, persuading past a constraint.
G4
Strategy / Management×1
Manage factions, resources or institutions toward explicit objectives.
G5
Simulation×4
Open-ended modeling of a social, historical or everyday situation.
G6
Sandbox / Creation×2
Freely combine and create; no fixed win condition.
G7
Social Deduction / Party×3
Multiplayer deduction, voting and comedic competition.
G8
Relationship / Companion×2
The loop centers on building a long-term bond with an AI character.
G9
Hybrid / Experimental×2
Novel forms that refuse a single conventional genre.

N-axis · AI mechanic

N1
Epistemic Interaction×19
Elicit hidden information from the model: interrogation, investigation, lie detection.
N2
Social Influence×11
Persuade, deceive, negotiate with or manipulate AI characters to progress.
N3
Generative Narrative / AI GM×12
Open-ended input drives the plot; the model hosts and continues the story world.
N4
Semantic Adjudication×6
The model rules on open actions — translating language into effects, outcomes, win/lose.
N5
Multi-Agent Simulation×4
Observe, intervene in, or manage a society of autonomous model-driven agents.
N6
Generative Construction×1
Play revolves around artifacts the model generates — images, objects, worlds.

Sources & data

The seed corpus is the 53 dated artifacts of the survey (Table 2), extended and re-verified against store pages and official sites. The dataset lives in the open on GitHub; every entry records its verification status. the survey (arXiv:2607.00527)

Submit a game

Built or found an AI-native game that passes the three tests? The index is a public dataset — add it via pull request, or open an issue with a link and we'll do the digging.

Submit on GitHub ↗