Public Steam disclosure research27,544 disclosures tracked
Where do games disclose AI use?
Search developers’ Steam disclosures, compare classified AI use across
games, and follow recorded changes.
This index adds cross-game classifications and statistics to
statements published on individual store pages.
Groups reflect automatic analysis of public disclosures; mature records are excluded. These counts do not measure AI use across all of Steam.
Missing or unclear statements cannot establish whether a game uses AI.
AI tools were used during development as a coding assistant: editing, reviewing, testing and debugging source code. The English in-game text and store text were translated from the developer's Japanese originals with the help of AI tools and then reviewed by the developer. All in-game art, music and sound effects are created by human artists without generative AI.
Game assets were created with the help of generative AI tools, then selected and edited by the development team. No content is generated live during gameplay.
During development, we used generative AI tools to assist with programming and to create or refine a small portion of the game’s art assets. The development team reviewed, edited, tested, and integrated all code and art assets produced with AI assistance. We also used AI tools to help create some of our Steam marketing materials.
The game does not use AI to generate content during gameplay and does not connect to any external AI services.
Localisation was created with assistance from generative AI.
Generative AI was used to prototype early musical concepts.
No content is generated during gameplay.
Generative AI tools assisted with some 2D visuals, promotional images, icons, title lettering and text localization. This content was created in advance and integrated into the game. The game does not generate AI content during gameplay.
Some of the visual assets, animations, sound effects and music in Tome of Ruin were created with the help of generative AI tools. Every asset was reviewed and edited by hand to meet the game’s quality standards, and many were remade by hand from scratch. Some store page translations were also produced with AI assistance. The game does not generate any content with AI during gameplay.
Many of the game's 3D models, textures and illustrations, and the translations of its text, were made with generative AI tools and then reviewed and edited by the developer. An optional bug-report helper on our servers uses a small text model to ask players up to two follow-up questions about a report they send.
Connects to 3rd-Party Service for AI Content Generation: Google Gemini 2.5 Flash-Lite via OpenRouter (allowed alternatives: OpenAI GPT-4o mini, Meta Llama 3.1 8B Instruct, Mistral Nemo)
Generative AI was used to create some images and audio,which were then manually edited and refined by our team.
No AI content is generated while the game is running.
Pawlosseum uses generative AI as a support tool for creating visual assets and some sound effects. The assets were developed following our art direction, with later editing and adjustments.
The game design, story, character design, mechanics and balance are original. No content is generated by AI while you play.
AI is used only in (as human retouched starting point):
- Some background/location images
- 2D animation (of human made characters)
- Character voices, and narration
- Some unique player sprites (by supporter request)
(No AI content is generated during gameplay. All normal character sprites, UI icons, and item sprites are human made. Willing to replace all AI assets with human made assets with enough support of the game, voice acting and 2D animation are expensive and I'm solo dev.)
Symloquy uses its official AI service to generate character dialogue and voice in real time. Some background images were created using generative AI.
Connects to 3rd-Party Service for AI Content Generation: Symloquy AI
We use generative AI tools to assist in the creation of 2D visual assets, 3D models, texts, and code. This includes certain in-game graphics, 3D models, UI elements, promotional artworks, and community items such as achievement icons. All AI-generated content serves merely as a foundation. It is heavily edited, adjusted, and manually finalized.
AI has mostly been used for programming & scripting. The places where it has been used for pre-rendered assets is the sky boats which was taken from the art I had drawn for Emplacement, the other was the under tiles so it could be generated into 3D models.
Some of the game's art was made with the help of generative AI (Google Gemini): character portraits, menu and screen backgrounds, some objects (chests, portals, crypt entrances) and icons. The pixel art drawn in code, the texts and the translations were made with the help of an AI assistant (Claude). All of this content was prepared in advance and selected and refined by the developer. The game does not generate anything with AI during gameplay.
Generative AI was only used for the game's title logo and app icon. The store screenshots were captured directly in the game. No AI content is generated during gameplay.
We used generative AI for game and promotional art, audio (including voices), and parts of the story text, then edited the results ourselves. The game doesn’t generate AI content while you play.
Generative AI tools were used to assist with certain in-game text localizations and visual assets, including illustrations, promotional artwork, and community materials. All final content is created by our team.
Coven of Thorns uses generative AI tools during development to assist artists in the pre-visualization: character and creature portraits, and some marketing images and trailer animation. All generated assets are selected, edited, composited and integrated by the Revenant Machine team before they ship. Game design, code, writing and card rules are authored by hand. The game does not generate any content with AI while you play.
Visual assets: Generative AI output was used in the creation of creature and environment art assets. That output was reduced to a 1-bit palette, hand-corrected at the pixel level, and assembled into final sprites and tilesets in-house.
Audio: Ambient audio is synthesized procedurally at runtime by a custom DSP synthesizer — no AI model runs in the shipped game. Generative AI was used during development to explore the synthesizer's parameter space; the resulting presets were selected and tuned by ear.
Writing: All narrative, item text, and dialogue is written and edited in-house; where AI tools assisted with drafting, the result was rewritten and approved by hand.
Store page and marketing assets: Marketing imagery uses art assets from the game, produced as described above. AI tools assisted with drafting and formatting store page copy.
No AI model runs while the game is playing. All AI use occurred during development, and every result was human reviewed and edited before shipping.
AI tools have been used to create music and the game logo. We also plan to use a self-hosted AI model on our own GPU hardware for text-only translation of in-game chat.
Name of AI model included with the game: Self-hosted AI translator
Generative AI was used during the development of Territory Idle Defense to create one in-game map, assist with programming and debugging, and help write and refine in-game attack descriptions, upgrade descriptions, and instructional text. AI was also used to proofread and polish the developer's original Steam store page description.
All other in-game artwork, including characters, enemies, and animations, was created by human artists.
The game does not use live-generated AI. No content is generated by generative AI during gameplay.
Ashwend started as an experiment in how far AI can take game development. The code, interface, art, models and animation were all made with AI.
I direct the project and test the game in play, but I don't write or read the code myself. That was the point of the experiment. The AI works within written architecture rules, and the code base is checked by automated tests, linting and static analysis to catch problems early.
Some of the game's 2D artwork (item icons and sprites) was created with the assistance of generative AI tools and then manually reviewed, edited and integrated by the developer. Parts of the text localization were also produced with AI assistance and reviewed by the developer. No content is generated by AI at runtime — all AI-assisted assets were created during development and curated by a human before being included in the game.
Some of the game’s 2D environment art was made with image-generation tools, then cut up, edited and assembled by hand. AI tools were also used during development to write automated tests. We plan to replace all AI-generated art before launch.
Some artwork and promotional materials in this game were created with the assistance of AI tools. All such materials were selected, edited, composited, and reviewed by the development team. The game’s story, gameplay design, programming, level logic, and final content were created and edited by the development team. The game does not provide generative AI features for players and does not generate content from player input in real time.
Name of AI model included with the game: GPT-6.1 Astra
Some visual assets in Hellraiders, including creature illustrations, character artwork, and promotional images, were created with generative AI tools during development. The development team reviews and selects these assets before they are included. Hellraiders does not generate AI content during gameplay.
Some artwork and text in the game and its marketing materials were created or revised with AI assistance. Bundled local language models generate NPC conversations, organize memories and propose actions based on character state. Included Chinese and English speech models can read replies that pass content checks. Characters have histories and relationships but are fictional, not real people. Item transfers, attitudes, relayed messages and quest changes still require game-rule and, where needed, character-consent checks; generated dialogue alone does not complete an action. The main story also contains authored scenes and choices. The Steam version uses only included local models: no third-party AI account, API key, subscription or per-message payment is required. AI and automatic speech can be disabled in settings.
Name of AI model included with the game: Bundled local models: Chinese text: ArityFlow-Qwen3-4B-Instruct-2507-RolePlay-Lantern.Q4_K_M.gguf; English text: Qwen3.5-4B-MiniFantasy-Q4_K_M.gguf. Retrieval embeddings: BGE-small-zh-v1.5 Q8_0. Chinese speech: Qwen3-TTS 12Hz 0.6B Base; English speech: Microsoft VibeVoice-Realtime-0.5B. Text models run through NobodyWho/llama.cpp. No player-configured external AI service is supported in the revised Steam build.
We're a small team, and AI tools helped us make the game's music, sound effects, pixel art and in-game text. Every piece was directed, chosen and refined by us, and nothing is generated by AI while you play.
I’m making Age of Harvest on my own. I use AI tools for coding only, throughout development. Everything else is my own work, made without generative AI. I design each gameplay system and the interface, draw the illustrations and pixel art, animate the sprites, compose the music, and build the world. I’ve also done all the writing.
We are a small indie team, and generative AI is one of the tools that helps us make this game. Some of the visual assets, texts and marketing materials were created with its help.
The vision behind the game is ours. We decide what the game should be and give detailed directions on how things should look, sound and feel. AI doesn't always get it right, so we review what it produces, rework or replace what doesn't fit, and only keep what matches the game we want to make.
I build NOSTRA: Crystalbound on my own. Generative AI tools helped create the art, the music and parts of the sound; all of it is selected, reworked and put together by hand before it ships.
Wakelocked was made with the help of generative AI tools during development. Nothing is generated while you play.
AI tools helped write what's said on the radio, the lyrics of its songs, the achievements and the store page text. The radio's voices were made with text-to-speech models, its songs with AI music generators, many of the sound effects with an AI audio model, and the logo with an AI image generator. We chose, reviewed and tested all of it.
All the 3D models, the boat and Cal included, are licensed.
Generative AI tools were used to assist with the creation and refinement of some visual content during development. All content is prepared before release. No generative AI is used during gameplay.
All 3D art and crew voices were made by hand. A few background music tracks and the logo were generated with AI tools, and AI helped draft the translations.
AI tools were used during development as a coding assistant: editing, reviewing, testing and debugging source code. The English in-game text and store text were translated from the developer's Japanese originals with the help of AI tools and then reviewed by the developer. All in-game art, music and sound effects are created by human artists without generative AI.
Game assets were created with the help of generative AI tools, then selected and edited by the development team. No content is generated live during gameplay.
During development, we used generative AI tools to assist with programming and to create or refine a small portion of the game’s art assets. The development team reviewed, edited, tested, and integrated all code and art assets produced with AI assistance. We also used AI tools to help create some of our Steam marketing materials.
The game does not use AI to generate content during gameplay and does not connect to any external AI services.
Localisation was created with assistance from generative AI.
Generative AI was used to prototype early musical concepts.
No content is generated during gameplay.
Generative AI tools assisted with some 2D visuals, promotional images, icons, title lettering and text localization. This content was created in advance and integrated into the game. The game does not generate AI content during gameplay.
Some of the visual assets, animations, sound effects and music in Tome of Ruin were created with the help of generative AI tools. Every asset was reviewed and edited by hand to meet the game’s quality standards, and many were remade by hand from scratch. Some store page translations were also produced with AI assistance. The game does not generate any content with AI during gameplay.
Many of the game's 3D models, textures and illustrations, and the translations of its text, were made with generative AI tools and then reviewed and edited by the developer. An optional bug-report helper on our servers uses a small text model to ask players up to two follow-up questions about a report they send.
Connects to 3rd-Party Service for AI Content Generation: Google Gemini 2.5 Flash-Lite via OpenRouter (allowed alternatives: OpenAI GPT-4o mini, Meta Llama 3.1 8B Instruct, Mistral Nemo)
Generative AI was used to create some images and audio,which were then manually edited and refined by our team.
No AI content is generated while the game is running.
Pawlosseum uses generative AI as a support tool for creating visual assets and some sound effects. The assets were developed following our art direction, with later editing and adjustments.
The game design, story, character design, mechanics and balance are original. No content is generated by AI while you play.
AI is used only in (as human retouched starting point):
- Some background/location images
- 2D animation (of human made characters)
- Character voices, and narration
- Some unique player sprites (by supporter request)
(No AI content is generated during gameplay. All normal character sprites, UI icons, and item sprites are human made. Willing to replace all AI assets with human made assets with enough support of the game, voice acting and 2D animation are expensive and I'm solo dev.)
Symloquy uses its official AI service to generate character dialogue and voice in real time. Some background images were created using generative AI.
Connects to 3rd-Party Service for AI Content Generation: Symloquy AI
We use generative AI tools to assist in the creation of 2D visual assets, 3D models, texts, and code. This includes certain in-game graphics, 3D models, UI elements, promotional artworks, and community items such as achievement icons. All AI-generated content serves merely as a foundation. It is heavily edited, adjusted, and manually finalized.
AI has mostly been used for programming & scripting. The places where it has been used for pre-rendered assets is the sky boats which was taken from the art I had drawn for Emplacement, the other was the under tiles so it could be generated into 3D models.
Some of the game's art was made with the help of generative AI (Google Gemini): character portraits, menu and screen backgrounds, some objects (chests, portals, crypt entrances) and icons. The pixel art drawn in code, the texts and the translations were made with the help of an AI assistant (Claude). All of this content was prepared in advance and selected and refined by the developer. The game does not generate anything with AI during gameplay.
Generative AI was only used for the game's title logo and app icon. The store screenshots were captured directly in the game. No AI content is generated during gameplay.
We used generative AI for game and promotional art, audio (including voices), and parts of the story text, then edited the results ourselves. The game doesn’t generate AI content while you play.
Generative AI tools were used to assist with certain in-game text localizations and visual assets, including illustrations, promotional artwork, and community materials. All final content is created by our team.
Coven of Thorns uses generative AI tools during development to assist artists in the pre-visualization: character and creature portraits, and some marketing images and trailer animation. All generated assets are selected, edited, composited and integrated by the Revenant Machine team before they ship. Game design, code, writing and card rules are authored by hand. The game does not generate any content with AI while you play.
Visual assets: Generative AI output was used in the creation of creature and environment art assets. That output was reduced to a 1-bit palette, hand-corrected at the pixel level, and assembled into final sprites and tilesets in-house.
Audio: Ambient audio is synthesized procedurally at runtime by a custom DSP synthesizer — no AI model runs in the shipped game. Generative AI was used during development to explore the synthesizer's parameter space; the resulting presets were selected and tuned by ear.
Writing: All narrative, item text, and dialogue is written and edited in-house; where AI tools assisted with drafting, the result was rewritten and approved by hand.
Store page and marketing assets: Marketing imagery uses art assets from the game, produced as described above. AI tools assisted with drafting and formatting store page copy.
No AI model runs while the game is playing. All AI use occurred during development, and every result was human reviewed and edited before shipping.
AI tools have been used to create music and the game logo. We also plan to use a self-hosted AI model on our own GPU hardware for text-only translation of in-game chat.
Name of AI model included with the game: Self-hosted AI translator
Generative AI was used during the development of Territory Idle Defense to create one in-game map, assist with programming and debugging, and help write and refine in-game attack descriptions, upgrade descriptions, and instructional text. AI was also used to proofread and polish the developer's original Steam store page description.
All other in-game artwork, including characters, enemies, and animations, was created by human artists.
The game does not use live-generated AI. No content is generated by generative AI during gameplay.
Ashwend started as an experiment in how far AI can take game development. The code, interface, art, models and animation were all made with AI.
I direct the project and test the game in play, but I don't write or read the code myself. That was the point of the experiment. The AI works within written architecture rules, and the code base is checked by automated tests, linting and static analysis to catch problems early.
Some of the game's 2D artwork (item icons and sprites) was created with the assistance of generative AI tools and then manually reviewed, edited and integrated by the developer. Parts of the text localization were also produced with AI assistance and reviewed by the developer. No content is generated by AI at runtime — all AI-assisted assets were created during development and curated by a human before being included in the game.
Some of the game’s 2D environment art was made with image-generation tools, then cut up, edited and assembled by hand. AI tools were also used during development to write automated tests. We plan to replace all AI-generated art before launch.
Some artwork and promotional materials in this game were created with the assistance of AI tools. All such materials were selected, edited, composited, and reviewed by the development team. The game’s story, gameplay design, programming, level logic, and final content were created and edited by the development team. The game does not provide generative AI features for players and does not generate content from player input in real time.
Name of AI model included with the game: GPT-6.1 Astra
Some visual assets in Hellraiders, including creature illustrations, character artwork, and promotional images, were created with generative AI tools during development. The development team reviews and selects these assets before they are included. Hellraiders does not generate AI content during gameplay.
Some artwork and text in the game and its marketing materials were created or revised with AI assistance. Bundled local language models generate NPC conversations, organize memories and propose actions based on character state. Included Chinese and English speech models can read replies that pass content checks. Characters have histories and relationships but are fictional, not real people. Item transfers, attitudes, relayed messages and quest changes still require game-rule and, where needed, character-consent checks; generated dialogue alone does not complete an action. The main story also contains authored scenes and choices. The Steam version uses only included local models: no third-party AI account, API key, subscription or per-message payment is required. AI and automatic speech can be disabled in settings.
Name of AI model included with the game: Bundled local models: Chinese text: ArityFlow-Qwen3-4B-Instruct-2507-RolePlay-Lantern.Q4_K_M.gguf; English text: Qwen3.5-4B-MiniFantasy-Q4_K_M.gguf. Retrieval embeddings: BGE-small-zh-v1.5 Q8_0. Chinese speech: Qwen3-TTS 12Hz 0.6B Base; English speech: Microsoft VibeVoice-Realtime-0.5B. Text models run through NobodyWho/llama.cpp. No player-configured external AI service is supported in the revised Steam build.
We're a small team, and AI tools helped us make the game's music, sound effects, pixel art and in-game text. Every piece was directed, chosen and refined by us, and nothing is generated by AI while you play.
I’m making Age of Harvest on my own. I use AI tools for coding only, throughout development. Everything else is my own work, made without generative AI. I design each gameplay system and the interface, draw the illustrations and pixel art, animate the sprites, compose the music, and build the world. I’ve also done all the writing.
We are a small indie team, and generative AI is one of the tools that helps us make this game. Some of the visual assets, texts and marketing materials were created with its help.
The vision behind the game is ours. We decide what the game should be and give detailed directions on how things should look, sound and feel. AI doesn't always get it right, so we review what it produces, rework or replace what doesn't fit, and only keep what matches the game we want to make.
I build NOSTRA: Crystalbound on my own. Generative AI tools helped create the art, the music and parts of the sound; all of it is selected, reworked and put together by hand before it ships.
Wakelocked was made with the help of generative AI tools during development. Nothing is generated while you play.
AI tools helped write what's said on the radio, the lyrics of its songs, the achievements and the store page text. The radio's voices were made with text-to-speech models, its songs with AI music generators, many of the sound effects with an AI audio model, and the logo with an AI image generator. We chose, reviewed and tested all of it.
All the 3D models, the boat and Cal included, are licensed.
Generative AI tools were used to assist with the creation and refinement of some visual content during development. All content is prepared before release. No generative AI is used during gameplay.
All 3D art and crew voices were made by hand. A few background music tracks and the logo were generated with AI tools, and AI helped draft the translations.