What Should You Record About AI Before Submitting a Visual Novel to a Store?

A visual-novel creator reaches the store submission form with an awkward production history. Some backgrounds began as generated drafts and were repainted. A language model helped reorganize a scene, but the dialogue was rewritten. The final build does not call an AI service while the player is reading. What, exactly, belongs in an AI disclosure?

The safest place to answer is not the percentage of the game that feels human-made. Start with what the player will actually receive, then preserve enough production history to describe how each result got there. Steam and itch.io ask different questions, so one vague sentence cannot reliably cover both.

This is a preparation guide, not legal advice and not a substitute for the current submission form. Platform language can change. Check the official page again when you submit.

Start with the shipped files, not the tool list

Make an inventory of the material that appears in the build, store page, community art, trailer, or downloadable asset package. For a visual novel, that usually means backgrounds, character art, interface art, music, sound effects, voices, narrative text, localization, code, and promotional images.

For every item, record four facts: where the final version appears, whether a generative tool produced any material that survives in it, what a person changed, and whether generation happens before release or while the game is running. Keep model names, dates, licenses, and source files in your private production record when available. The public form may not request every detail, but you cannot write a precise summary after the history has been discarded.

Do not begin with every tool anyone touched. Steam's current Content Survey says the AI section is not focused on efficiency gains from AI-powered development tools. Its focus is AI-created content that ships with the game and is consumed by players, including art, sound, narrative, and localization. That makes the final output the useful first filter.

This does not mean an invisible tool can never matter for another policy, contract, or copyright question. It means the Steam disclosure described here has a stated scope. Keep broader records privately even when a particular form asks for less.

Steam separates prepared content from live generation

Steam's official documentation divides relevant content into pre-generated and live-generated uses. Pre-generated material is made with AI help during development and then ships in a fixed form. A generated background, edited generated portrait, synthetic voice line, or AI-assisted passage that remains in the final visual novel can fall into this working bucket.

Live-generated content is created with AI help while the game is running. Steam asks for the same underlying disclosure and also requires a description of guardrails intended to prevent illegal output. For a conventional visual novel whose story, images, and audio are fixed before release, this live category may not apply. A project that lets the player generate new dialogue or images during play is a different technical and moderation commitment.

Human editing is important, but it is not a reliable reason to erase origin from your notes. The practical question is whether generated material survives in what the player sees or hears, not whether a person later spent many hours improving it.

Steam also states that disclosure does not make otherwise prohibited material acceptable. Its prerelease review still checks the build against the platform agreement and the store materials. Treat the form as a description of the product, not a permission slip.

itch.io classifies the result in another way

itch.io's project editor asks whether a project contains results of generative AI. If the answer is yes, its official announcement says the form asks for output types such as graphics, sound, text and dialogue, and code. The platform then applies generated-content classifications automatically.

Its current quality guidelines add two distinctions that matter. Traditional systems such as pathfinding, procedural level generation, and dynamic difficulty are not treated as generative AI tags merely because they are called game AI. The guidelines also warn against mass-produced pages and ask creators to state AI involvement clearly when a project uses automatic or AI generation.

Be exact about enforcement. The public guideline explicitly says strict disclosure enforcement and possible delisting apply to game asset pages. The broader project guidance asks creators to tag projects accurately, but the cited text does not say every game page receives the same blocking workflow as an asset seller. Do not turn a narrow official sentence into a broader rule.

One further detail is useful for asset creators: itch.io's announcement says generated assets remain within that rule even after modification. That is another reason to record origin and human changes as separate facts.

Build one evidence sheet, then answer each platform

A compact internal sheet can support both submissions. Give each final item a row and use fields such as final location, content type, generation stage, generated contribution, human revision, rights evidence, and player exposure. Add a live-generation field only when the running game calls a model or creates new content.

Do not write disclosure copy until this sheet is complete. Otherwise the loudest tool in your memory becomes the whole story, while a small generated store image or localized line is forgotten.

When you do write the public explanation, prefer concrete nouns and actions. Describe which final materials used generation and what the creator did afterward. Avoid percentages unless you have a defensible counting method. Avoid claims such as fully human-curated when the phrase does not tell a reviewer what was generated.

A useful sentence pattern is: generated material was used for a named content type, then a named human review or revision occurred before release. Use that pattern only when it matches the record. If the script was written without generated text, say so only if you verified it. Specificity is more useful than a broad defense of the technology.

How does this apply to a Novelez project?

Novelez can help create a playable visual novel from a short premise and can keep scenario blocks, characters, backgrounds, and endings in one editable project. It does not decide what another platform's disclosure form means for your release, and exporting a project does not create the missing production history for you.

Record decisions while you work. When you keep a generated scene, replace an image, rewrite dialogue, or export a final package, note which generated results remain. The final project may mix generated, licensed, and manually created assets. The store submission should describe that actual mixture rather than the first draft or the name of the creation tool.

If your visual novel only publishes as a Novelez web link, Steam's submission form is not part of that release. The record is still useful. It prepares you for a later engine export, an itch.io page, a collaborator handoff, or a clear answer when a player asks how the work was made.

Compare the statement with the build before submitting

Open the exact release candidate and move through the inventory one last time. Check the title image, store capsule, screenshots, credits, every route-specific background, voices, localization files, and downloadable extras. A disclosure that matches the main route but misses an alternate ending is incomplete.

Then verify the current official forms. Steam notes that some survey answers become harder to change after review and may require contacting support. That makes a final comparison cheaper than correcting a rushed answer later.

Keep the evidence sheet after launch. If an asset is replaced or a live-generation feature is added, update the record before editing the public description. The aim is not to predict every platform phrase forever. It is to preserve facts well enough that a changed form can be answered honestly.

Which part of your current visual novel would be hardest to document today: the origin of an image, the history of a rewritten scene, or a tool used somewhere in the export process?