Yes—Steam requires developers to disclose generative AI when it creates content players consume. That includes shipped artwork, sound, music, voices, narrative, translations, store-page copy, capsule art, trailers, Steam community assets, and content generated during play. Valve’s current test is not whether any employee used an AI tool; it is whether AI-produced material reaches players. Code autocomplete, debugging assistants, automated tests, and internal concept sketches that never ship are not the focus of the current survey wording. A disclosure is not a ban, but the Content Survey must be completed before review, and live generation requires additional safety controls.
What does Steam’s AI rule actually require?
Steamworks’ Content Survey asks whether generative artificial intelligence created content for the game, either pre-rendered or live-generated. The question explicitly covers the game, its store page, Steam community assets, and marketing materials. If the answer is yes, developers must describe the implementation in detail before submitting the product for review.
Valve separates AI use into two categories:
- Pre-Generated: Content made with AI during development and shipped in a form players can see, hear, or read.
- Live-Generated: Content produced with AI while the game is running, such as an NPC response, image, or audio clip.
Valve says efficiency gains from AI-powered development tools are “not the focus” of this section. The concern is player-consumed content, including artwork, sound, narrative, and localization. This is a clarification of the survey’s scope, not a statement that Valve has certified particular products or legalized every use of coding assistants.
Read the live source before every submission because Valve describes the page as evolving: Steamworks Content Survey.
Do I have to disclose AI on Steam?
Disclose it when a generative model helped create material that ships with the game or appears in the associated Steam presence. The survey’s three sections must be completed before a product enters the Review Process, and Valve compares the answers with the submitted build and store page.
| Use | Current Steam treatment | What to explain |
|---|---|---|
| In-game art, sprites, textures, models | Disclose as Pre-Generated | Asset type, location, tool involvement, and human review or editing |
| Music, sound effects, voices | Disclose as Pre-Generated | Which audio is AI-assisted and how it was selected or edited |
| Dialogue, lore, item descriptions | Disclose as Pre-Generated | Where generated text appears and the human approval process |
| AI translations shipped in the game | Disclose as Pre-Generated | Languages, scope, and fluent-human review |
| Capsule art, trailer footage, store copy, community assets | Disclose | The relevant marketing or community material, not only the executable |
| Code assistants, autocomplete, debugging, automated tests | Not the focus of the current wording | No player-facing generative output is identified by the survey examples |
| Internal concepts or placeholders replaced before shipping | Not player-consumed | Keep internal records, but they do not ordinarily trigger this disclosure |
| NPC chat, images, or audio generated during play | Disclose as Live-Generated | Output type plus the safeguards that prevent illegal content |
The practical rule is simple: inventory what a player can encounter, not who clicked the Generate button. An AI-made asset bought from a contractor or marketplace still concerns the shipped result, so obtain the creator’s disclosure and licensing information in writing.
Does using Copilot, ChatGPT, or Cursor for code require Steam disclosure?
Usually not under the current Steam wording when those tools only improve development efficiency and no generated material is shipped as player-facing content. Valve’s January 2026 rewrite specifically says that this efficiency use is not the focus and names creative content instead.
That does not remove other obligations. A coding assistant can still introduce infringing code, an unsafe dependency, or a license conflict, and Steam’s general content and distribution rules still apply. If the same project uses a model to generate dialogue, textures, voice lines, localization, or store copy, disclose those outputs even if the code itself came from ordinary development tools.
Other storefronts may ask a broader question. itch.io’s disclosure form has separate Graphics, Sound, Text & Dialog, and Code categories, so a code-only use that is outside Steam’s current focus may still need a code disclosure there.
What should I write in the Steam AI disclosure?
Write for players and name the actual content, where it appears, and the human review that made it suitable for release. Valve has not published official sample wording; vague statements such as “we used AI tools in our creative process” have been reported by developers as too imprecise.
A useful structure is:
- Identify the content: for example, background textures, optional NPC voice lines, item descriptions, or Spanish localization.
- Identify the location: specify the game mode, menu, chapter, language, store asset, or community material in which players encounter it.
- Describe human control: explain editing, selection, fact checking, performance direction, moderation, or rejection of unsuitable output.
- State boundaries: say which major areas were not AI-generated when that distinction helps players understand the scope.
For example: “AI-assisted tools generated first-pass Spanish translations for optional dialogue. A fluent human editor rewrote and approved every line before inclusion; the English script, character performances, and in-game art were created without generative AI.” This is an example of specificity, not Valve-approved language.
The public store block is headed “AI Generated Content Disclosure” and introduces the developer’s explanation. Treat it as player communication, not an internal compliance note.
Do store-page art, capsule images, and trailers count?
Yes. The threshold question explicitly includes the store page, Steam community assets, and marketing materials, even when the files are not inside the game build. AI-generated capsule art, trailer B-roll, promotional illustrations, or store-page text therefore belong in the inventory and disclosure.
Keep the wording consistent across the build and marketing. Valve’s Pre-Generated definition ties disclosure to promises in the Steam Distribution Agreement: the game must not contain illegal or infringing content and must match its marketing materials. A disclosure does not cure an infringing likeness, unlicensed voice, misleading trailer, or other rule violation.
Does AI localization, voice, or music need disclosing?
Yes, when the output ships and players consume it. Valve names sound, narrative, and localization as examples, so AI music, sound effects, synthetic voices, dialogue, and translated text should be listed as Pre-Generated content.
The commercial evidence suggests that the way these uses are handled matters. In Sulka Haro’s 2026 census of 53,597 Steam releases from July 2023 through July 2026, successful AI-flagged games were more likely than flops to disclose voice (24% versus 8%) and localization (18% versus 6%). That is an association in estimated revenue, not proof that a particular disclosure causes success. The same census found visual AI was mentioned by 72% of sampled flops and 57% of sampled successes.
For localization, distinguish a machine first pass that a fluent editor substantially rewrites from text shipped with only light proofreading. Valve has not published a threshold for how much rewriting removes a disclosure duty; the conservative operational choice is to disclose whenever a generative model materially produced or transformed the shipped words.
What are the rules for live-generated AI?
Live generation is allowed only with an explanation of safeguards that limit illegal output. Because Valve cannot inspect every runtime response before release, the Content Survey asks what guardrails operate around the model.
Safeguards to document
- Input and output moderation, including image, text, and audio filtering where applicable.
- System prompts and topic limits that keep responses inside the game’s intended setting.
- Blocked terms, rate limits, and controls against prompt injection or abuse.
- Logging and a process for investigating harmful generations.
- A player reporting route. Valve’s 2024 announcement described a Steam-overlay path for reporting illegal content in live-generated games.
These are practical controls described by developers, not a published Valve checklist. Explain what your system actually does, how a tester can reach it, and what happens when a filter blocks or escalates output.
Does Steam allow AI-generated adult content?
Valve currently refuses to ship Live-Generated AI Adult Only Sexual Content. Its FAQ says the legal and customer risks are substantial enough that it does not want to ship that category at this time.
This restriction is specifically about live-generated adult-only sexual content. It does not turn the general AI disclosure into an adult-content approval, and all other Steam content rules still apply. If a project includes adult material, submit the required store page and build information and expect longer review timelines.
How do I handle a paid live-AI feature?
If each model call costs the developer money, Steamworks requires the developer to control access and collect payment through a Steam-supported method. Valve lists four approaches:
- Include the expected generation cost in the base game price.
- Sell Steam microtransactions.
- Offer a subscription with unlimited access.
- Sell a one-time DLC unlock.
Valve does not designate one approach as preferred. Decide the entitlement and failure behavior before review so testers can access the feature without an undocumented external payment flow.
Can I remove the AI disclosure after replacing the assets?
Do not assume the public label can be edited after approval. Once the build and store page are approved, some survey answers are locked; Steamworks instructs partners to contact Steam Support with a change summary, the affected answer, and tester instructions.
Maintain an asset register with the tool, plan, generation date, source file, human edits, and replacement status. If AI content is added or removed, contact Support before publishing the update and explain exactly how a tester can reproduce the changed content.
Will Steam reject games that use AI?
AI use itself is not a rejection category. Valve said in October 2026 that it does not want to gatekeep which tools developers use and announced no new AI-tool ban. Reviewers still evaluate the resulting content under the same rules as non-AI content, including legality, infringement, and consistency with marketing.
Completing the survey is mandatory, but Valve has not published an automated AI-detector, an official detection rate, or an official count of AI-disclosure rejections. Incomplete or inaccurate answers can nevertheless surface through review, player reports, or later compliance work. Haro found that 22% of significant-revenue AI games in its census added the flag after release, with additions clustering around the January 2026 wording change.
How hostile are players to disclosed AI?
The disclosure is public and commercially meaningful. Haro’s full census found roughly one in three new Steam games carried a genuine at-submission AI disclosure by mid-2026, while AI-flagged games accounted for an estimated 10–27% of sales in late 2025 and 2026—below their roughly 33% share of launches. Those estimates use the Boxleiter method (reviews multiplied by 30 for units and units multiplied by list price for revenue), are order-of-magnitude figures, and exclude Steam’s cut and discounts.
Quantic Foundry’s optional survey of 1,799 respondents collected from October through December 2025 found 85% had a below-neutral attitude toward generative AI in games and 63% chose its most negative option. Respondents were most negative about artwork, music or sound, and narrative or dialogue; non-creative uses such as dynamic difficulty drew less opposition. The sample leaned male (65%), ages 25–34 (47%), core players (74%), and PC players (90%), so it is not a census of all gamers.
Haro’s commercial analysis also found that the top 1% of Steam titles capture about 94% of estimated revenue, while the median game with any reviews has roughly $300 in estimated revenue and about 28% of games receive zero reviews. These figures describe an extremely concentrated market, not a guaranteed penalty for an individual disclosure. They do suggest that visible, low-effort art generation is a reputational risk, while carefully integrated voice, localization, and writing can be part of a more deliberate production strategy.
How do Steam’s rules compare with itch.io and the Epic Games Store?
The storefronts ask different questions, so one disclosure cannot simply be copied everywhere.
| Storefront | AI disclosure approach | Important difference |
|---|---|---|
| Steam | Mandatory Content Survey before review; player-facing Pre-Generated and Live-Generated categories | Current wording de-emphasizes efficiency tools such as coding assistants and focuses on shipped content |
| itch.io | Generative AI field with Graphics, Sound, Text & Dialog, and Code categories | Code counts as a category; disclosure is required for asset pages and encouraged for games |
| Epic Games Store | No equivalent Steam-style AI label for games | Epic leadership has publicly opposed storefront AI labels; its separate 3D marketplace policies should not be assumed to apply to the game store |
Traditional game AI such as pathfinding, procedural generation, and dynamic difficulty is excluded from itch.io’s generative-AI field. Keep separate records for each storefront rather than treating “AI use” as one universal checkbox.
What changed between 2024 and 2026?
Valve added a generative-AI section to the Content Survey in January 2024. It said the change would allow the vast majority of AI-using games to release, described Pre-Generated and Live-Generated content, and introduced a player-report path for illegal live output.
On or about 16 January 2026, Valve rewrote the developer-facing text. The new wording explicitly says efficiency gains are not the focus and names artwork, sound, narrative, and localization. Commentary from Rock Paper Shotgun characterized this as a clarification rather than a new permission. At Steam Indie Connect in Busan on 27–28 October 2026, Valve representatives acknowledged AI slop, cloning, and extortion threats but announced no new tool ban.
What should developers do before submitting?
Use this sequence to avoid a last-minute mismatch between the survey, build, and store page.
- Inventory player-facing material. Include the executable, downloadable content, menus, screenshots, capsule art, trailers, store copy, Steam community uploads, and runtime outputs.
- Record provenance. For each item, note whether a generative model was involved, the tool and plan, generation date, license terms, and human edits.
- Answer the threshold question honestly. Select yes if any covered content was generated for the game or its Steam materials.
- Write a specific Pre-Generated explanation. Name the content, location, scope, and review process; identify material that was not AI-made when useful.
- Describe Live-Generated controls. Document moderation, prompts, blocked terms, rate limits, logs, reports, and tester access.
- Check commercial rights. Free tiers of generative services may restrict commercial use, and upgrading later may not license earlier outputs. Verify the plan in force when each asset was generated.
- Check Steam’s general rules. Confirm that content is legal, non-infringing, and consistent with the marketing promise.
- Plan payment for metered runtime AI. Use a Steam-supported base-price, microtransaction, subscription, or DLC model.
- Submit with time to spare. Store presence and build review typically take three to five business days; Valve asks partners to plan at least seven, with longer timing possible for Adult Only Sexual Content.
- Re-contact Support after material changes. If AI assets are replaced or added after approval, provide the affected survey result and precise tester instructions before assuming the answer is editable.
What remains unsettled?
Valve has not published a decision tree for AI upscaling of a human-painted texture, motion cleanup on human motion capture, a machine-translated first pass that a fluent editor rewrites, or the amount of overpainting needed to remove a disclosure duty. Practitioners generally recommend the conservative rule: disclose when a generative model produced or materially transformed shipped pixels, audio, or words.
That approach is operational advice, not a Valve guarantee. Over-disclosure has no stated Steam penalty; an under-disclosure that players later disprove can create a larger trust and compliance problem. Copyright is a separate question: in the United States, Thaler v. Perlmutter and the Copyright Office’s January 2025 report emphasize human authorship, with protection assessed case by case when a person determines expressive elements. Steam disclosure does not grant copyright, so retain human-edit records for publisher warranties, indemnities, and disputes.
Frequently Asked Questions
Where does the Steam AI disclosure appear to players?
It appears on the game’s store page in a public block headed “AI Generated Content Disclosure.” The block introduces the developer’s own explanation, so write it in clear player-facing language rather than internal legal shorthand.
Do procedural generation and traditional NPC AI trigger Steam’s survey?
The current Steam examples concern generative content that produces artwork, sound, narrative, localization, or other player-consumed material. Ordinary pathfinding, scripted behavior, and dynamic difficulty are not identified as generative-AI disclosure categories, although a system that creates new player-facing text, images, or audio at runtime is Live-Generated.
What evidence should a small studio keep about AI assets?
Keep an asset register containing the service, commercial plan, generation date, prompts or source inputs where material, human edits, approvals, and the final file. Written confirmations from contractors and marketplace sellers help establish whether purchased content was AI-made and what rights were granted.
Can a player report undisclosed AI content?
Players can report illegal content, and Valve may compare reports with the build, store page, and survey answers. Valve has not published a guaranteed detection rate or a dedicated public process for every undisclosed-AI allegation, so accurate initial disclosure remains the safer process.