The headline numbers are striking. More than 1.77 million game results without filtering, fewer than 425,000 when No AI is selected, according to Engadget's November 3 report.

It would be tempting to interpret the difference as a measurement of how many games on itch.io contain generative AI. That interpretation would be wrong.

The storefront does not independently scan every project's files to determine whether AI tools were involved. Its filtering system uses declarations supplied by creators. A missing negative declaration and an affirmative AI disclosure are not equivalent, even though both may prevent a project from appearing in No AI results.

This is a metadata classification problem as much as an AI moderation problem.

The difference between disclosure and detection

itch.io supports categories including AI Assisted, AI Graphics, AI Audio, AI Text and AI Code. These labels describe different forms of generative AI usage and allow visitors to narrow search results based on project metadata.

The No AI option focuses on creators who explicitly indicate that their work does not contain generative AI contributions.

This design avoids the technical and practical challenges involved in attempting to identify AI-generated material automatically. Such detection would require analyzing different types of assets, code and creative content, each with its own limitations.

The trade-off is reliance on the publisher's description of the project.

An inaccurate disclosure may result in incorrect classification. An incomplete disclosure may keep a legitimate handmade project out of the filtered results. The system can improve browsing without establishing an independently verified production history.

The reported reduction in visible games therefore cannot be used to calculate the proportion of generative AI projects on the storefront.

Automated publishing creates a separate scaling problem

itch.io has reported that the number of game pages created each day doubled over a two-month period. Downloads and purchases did not experience equivalent growth.

That mismatch suggests increasing pressure on the storefront's discovery systems, although it does not prove that every additional submission was created using generative AI.

The platform's quality guidelines already address excessive automatically generated submissions. Examples include algorithmically produced projects with little human involvement, randomly generated content lacking meaningful curation and multiple near-identical versions of the same work.

These behaviors can undermine discovery regardless of the specific technology used to produce the content.

itch.io has reported deleting several terabytes of unwanted generated material. The platform has not provided a reliable count of affected projects or detailed the distribution of file types.

A reference to ten terabytes of poorly optimized generated models appeared in itch.io's public explanation, but that example should not be treated as a verified figure for the actual cleanup.

Data volume also says little about the number of unique projects removed. Large model files and small game packages can create substantial differences between storage consumption and page counts.

Discovery restrictions are not always deletions

itch.io's official creator guidelines distinguish between removing a project from discovery and deleting access to it entirely.

A page that violates quality requirements may be excluded from browsing, search and recommendations while remaining reachable through a direct URL or the creator's profile.

This gives the platform a less drastic moderation option than permanent removal. It also allows the discovery system to enforce quality standards without making every problematic listing inaccessible.

The distinction matters when discussing large-scale moderation actions. Deleted storage, delisted pages and accounts subject to enforcement are different metrics.

The November announcement does not provide enough information to determine how these measures were distributed across the reported cleanup.

Older projects face a metadata disadvantage

A growing dependence on disclosure fields changes how creators need to maintain their listings.

Projects published before the current classification requirements may not have complete information about generative AI. Their creators may need to revisit project settings to ensure the correct discovery categories apply.

This does not mean those projects used AI. It means the discovery system may not have enough information to classify them as explicitly AI-free.

The expanded requirements now cover additional categories, including tabletop games, books, comics and soundtracks. Game assets were already subject to stricter disclosure rules.

That broader scope makes metadata consistency important across different forms of creative work, not just downloadable video games.

Generative AI is not the same as conventional game AI

The official guidelines draw a useful technical distinction between generative AI and established algorithmic systems used in game development.

NPC pathfinding, traditional enemy behavior, procedural level generation and dynamic difficulty adjustment do not automatically require generative AI disclosure.

Instead, the policy targets systems that generate new content using models trained on large datasets, such as text generators, image models and comparable audio-generation tools.

The distinction prevents ordinary game programming techniques from being incorrectly grouped with more recent generative systems.

It also introduces practical complexity. A development pipeline may combine conventional algorithms, manually authored content and generative AI tools. The resulting disclosure has to describe relevant outputs accurately.

A project might use a generated illustration for one element, while another could automate substantial portions of its production. Both situations require clear metadata if users are expected to make meaningful decisions.

Filtering cannot solve the entire moderation problem

itch.io's open publishing model lowers barriers for independent creators. It also makes the platform vulnerable to high-volume automated submissions when the cost of producing new pages approaches zero.

Discovery and moderation resources do not scale automatically with submission volume. Additional projects compete for the same limited user attention, while the platform must identify low-value or abusive uploads without unnecessarily restricting legitimate creators.

The No AI filter improves user control over declared AI usage. It does not independently verify each project, measure creative quality or eliminate automated spam.

Similarly, deleting abusive submissions can reduce the immediate burden without preventing future waves of automatically produced content.

itch.io has chosen to preserve access for legitimate AI-assisted projects while enforcing existing rules against excessive automated publishing.

Its effectiveness will depend on reliable metadata, appropriate moderation decisions and the ability to keep the discovery index useful as new content continues to arrive.