Step 1: inventory#
| Asset or system | AI used? | Pre-generated or live? | Human review | Disclosed where |
|---|---|---|---|---|
| Character dialogue | ||||
| Quests / levels / cases | ||||
| 2D art, textures, UI | ||||
| 3D models, animation | ||||
| Voice, music, SFX | ||||
| Marketing art and trailers | ||||
| Code | ||||
| Localisation |
Fill in every row, including "no". The inventory is what you'll answer platform questionnaires from, and it keeps answers consistent between stores and over updates.
Step 2: live-generated content guardrails#
Stores that ask about live-generated content want to know what stops it producing illegal or harmful material. Describe your layers concretely:
GUARDRAILS FOR LIVE-GENERATED CONTENT
Systems that generate content at runtime: [e.g. character dialogue]
Model(s) and hosting: [family, where it runs]
Input handling: [filtering of player input; prompt-injection handling]
Character constraints: [tone rules, banned topics, rating envelope: e.g. Teen]
Output checks: [classifier categories checked before display]
On failure: [authored in-world fallback line; no raw output shown]
Player reporting: [in-game key/menu; response time target]
Updates: [server-side guardrail updates without client patch]
Testing: [eval suite size, red-team scenarios, pass thresholds]Our disclosure and ratings guide explains rating envelopes and why fallbacks should be authored lines rather than blocks.
Step 3: store-page wording#
Specific beats vague. Pick the case closest to yours and adapt it:
Step 4: pre-submission checks#
- Inventory complete; every "yes" has a reviewer named
- Store questionnaire answers match the inventory exactly
- Store-page disclosure written in plain language and reviewed by the narrative lead
- Guardrails description matches what the build actually does
- In-game reporting works and routes to a monitored queue
- Privacy policy covers retention and any training on player input
- Disclosure update process defined for future patches
Frequently asked questions#
Do I need to disclose AI use on Steam?
Steam asks developers to disclose AI-generated content and to describe guardrails for live-generated content. Check current Steam documentation; policies change.
How should I word an AI disclosure?
Say what is generated, what is human-made, why AI is used, what guardrails exist and how players can report problems. Specific disclosures land better than vague ones.