36.4% of October Steam Next Fest Games Disclose AI Use: A Record Data Point and a Crisis of Meaning for the “Disclosure System”#

By Wubian (NTGame Editorial · AI + Gaming Watch Column)

Hello everyone, I’m Wubian.

On October 19, the day the October Steam Next Fest opened, I copied the tracker site’s numbers into my topic notebook as usual. Halfway through, I stopped: 7145 participating games, the most of any edition; about 2596 of them disclosed the use of generative AI, or 36.4%. In June’s edition that share was just over 19% — in four months it rose about 16 percentage points, nearly doubling.

Having watched AI for years, I’m used to indicators setting new highs, but “the disclosure share nearly doubled in four months” still made me look twice. Not because it proves “more people are making games with AI” — that direction is unsurprising — but because it shakes something more fundamental: the AI content disclosure system was designed precisely so players could “filter by it,” and when disclosed games approach forty percent of the shelf, what can that tag still filter out? Xiaoxiao, producer of Flower Crown Strategy, said something during this Next Fest that I recorded verbatim in my notebook: “The AI trend is unavoidable. In the future, once the share rises further, the AI content disclosure option may lose its meaning, and users will ultimately judge by quality alone.”

First, the sourcing. The previous article, Behind Steam Next Fest’s Record 8,202 Demos: AI Supply Glut, Player Backlash, and Demo Self-Preservation Under Clone Panic, covered the phenomenon side of this Next Fest — player backlash, complaints about sameness, developers cancelling public demos in self-preservation. This piece divides labor with it: it focuses on the data itself and the disclosure system — where 36.4% comes from, why it rose so fast, what is happening to the tag’s filtering value, and the two forms of “failure” the disclosure system is beginning to exhibit. All specific numbers, dates, and quotes come from public reporting by multiple outlets retrieved on October 25, 2026 (the primary statistical source is the TouchGame weekly citing SteamDB, the Steam data tracker); discrepancies are flagged at each point; anything officially unstated or unmeasurable is written as such, without speculation.

1. The Data First: 7145 Games, 36.4%, and the Four Sets of Figures Beside Them#

1.1 The Core Numbers#

First, the core statistics for this Next Fest:

On its own, 36.4% is just a percentage; it gains texture in the context of the shelf: of every 100 games in this Next Fest, more than 36 disclosed on their store pages that generative AI was used in development; nearly 4 in every 10. In June, that figure was “just over 19%” — back then disclosure was still a minority signal; four months later, it has become a majority-in-the-making that cannot be ignored on the shelf.

1.2 Five Sets of Figures Side by Side#

For this kind of data, methodology matters more than the numbers themselves. Different trackers give slightly different counts for the same Fest; here are all the figures I could find:

A few clarifications. First, the three October figures dovetail: the spread among 2593, 2595, and about 2596 is in single digits, and the difference between 36% and 36.4% is rounding — the conclusion is robust. Second, the Russian outlet’s October 21 snapshot differs from the primary source on both absolute numbers (7217 and 2628 vs. 7145 and about 2596) while the percentage is the same 36.4% — the snapshot timing differs and the participant list changes dynamically during the event; such discrepancies do not affect the trend conclusion. Third, the June baseline has two metrics: TheGamer’s “just over 19%” and GameDiscoverCo’s 26.5%; the samples differ, and this article’s body uses the former against 36.4%, keeping the latter as a methodology note only. Fourth, for the remaining games after excluding the AI tag, English-language outlets’ counts range from 4549 to 4552.

Whichever metric you adopt, the conclusion is the same: over four months, the disclosure share rose from just over 19% to 36.4% — from under a fifth to over a third. This is not noise; it is a trend.

1.3 Why the Absolute Numbers Cannot Be Compared Directly#

Let me answer a question any data worker would ask: with both months’ shares in hand, can’t we compute “absolute growth”? No — and it’s worth explaining why.

• TheGamer’s June metric gives only the share (just over 19%), not the absolute totals of participants and disclosures for that edition, so there is no way to know how many games 19% corresponds to;
• GameDiscoverCo’s June metric does have complete absolutes (4382 demos, 1163 disclosed), but its sample differs from TheGamer’s — dividing them yields 26.5%, which never matched the former’s 19% in the first place; forcing a comparison between 1163 and this edition’s roughly 2596 is like measuring distance with two rulers of different scales;
• This edition’s 7145 and about 2596 come from SteamDB’s participant-list statistics; without a same-method June snapshot, the period-over-period comparison can only be made between “each one’s own share.”

So this article makes only the “share-versus-share” comparison (19% vs. 36.4%) and leaves the absolute comparison blank — not laziness, but the only honest approach given two incompatible methodologies. Incidentally, the “dynamic participant list” point was already raised above (the Russian snapshot’s 7217 vs. the primary source’s 7145); it reminds us that such statistics inherently carry a point-in-time quality, and any cross-period comparison precise to the digit deserves extra caution.

2. Why the Share Rose: The Supply-Side Capacity Logic#

2.1 How the Disclosure System Works, and What 36.4% Means#

To understand this number, you must first see the disclosure system’s boundaries. Valve’s requirements for store pages are as follows:

The third row is the system’s most fragile link: with no automated detection, disclosure rests entirely on honesty. Two direct consequences follow — first, 36.4% is a floor, not a ceiling; the true share may be higher, but by how much is unmeasurable; second, games that “used AI only in internal workflows” look identical on the store page to games that “used no AI at all,” and players cannot tell them apart.

2.2 Falling Capacity Barriers: Participation and Disclosure Set New Highs Together#

The share rose so fast against an unavoidable supply-side backdrop: the barrier to making demos is falling. Industry interviews during this Next Fest offered these estimates — Yulia Tishina, CEO of IGROPROM (the Russian game industry organization), estimated that a vertical slice for an indie game used to take 4–6 people several months, and can now be done by 1–2 people in a month and a half to two months, with concept art, textures, UI, code scaffolding, rough voiceover, and translation all accelerated by AI; Anastasia Dzhioeva, game development director at Astrum Entertainment, said generative AI lets small teams build projects at lower cost, but “it’s easier to start and harder to succeed commercially.”

The sourcing must be pinned down: these are two practitioners’ estimates and opinions, not systematic statistics, and this article cites them as opinions. But pieced together with two hard data points, the logic is self-consistent —

• Participation sets a record: 7145, the most of any edition. More teams can produce demos, so the shelf naturally grows;
• Disclosure sets a record: 36.4%. Concept art, textures, UI, voiceover, and translation are precisely high-frequency categories of “player-facing content”; when AI accelerates these steps, the probability of AI-generated content remaining in the product rises accordingly.

The two curves are two faces of the same coin: falling capacity barriers pushed up total participation and also pushed up the absolute number and share of works “that used AI.” Record participation and record disclosure are less like two events than two readings of the same one.

2.3 Disclosure Categories and Accelerated Steps: A Mapping Table#

More interesting is setting two lists side by side: one is the categories covered by the disclosure system (Valve’s required “player-facing AI-generated content”), the other is the capacity steps Tishina said AI accelerated. The mapping is my analytical synthesis, not a statistical finding, but the neatness of the correspondence justifies its own table:

This table shows at least two things. First, nearly every step estimated to be AI-accelerated falls inside the disclosure system’s coverage — there is a coherent transmission chain between capacity acceleration and rising disclosure: whatever the accelerated steps produce is what gets disclosed on the store page. Second, the sole exception is “code scaffolding”: under the disclosure system’s boundaries, AI in internal workflows need not be reported, meaning the production steps actually transformed by AI are wider than the store-page disclosure list. This again echoes Section 2.1’s conclusion — 36.4% is a floor, and the hidden mass above that floor lies precisely on the side the system does not require reporting.

3. The Dilution of “Disclosure”: A Filter That Is Breaking Down#

3.1 The Tag’s Filtering Value Decays as the Share Rises#

The entire value of the disclosure system rests on one premise: disclosers are the minority. Only a minority’s tag carries discriminating power. Applied to the data, the decay path is clear — the second half of the table below is projection, not statistics, clearly marked:

What this table is saying is not “disclosure is useless,” but that its usefulness declines as coverage rises: when disclosers are the minority, “has disclosure” is an effective signal; when disclosers approach or pass half, the effective signal becomes “no disclosure” — and because the system relies on voluntary reporting, “no disclosure” mixes “truly didn’t use AI” with “didn’t report it,” making that signal equally suspect. In other words, this filter leaks at both ends.

Walking the decay through with this edition’s numbers: in June, just over 19% — fewer than one in five participating games disclosed AI, so when players saw the disclosure tag they were most likely looking at a shelf minority, and the tag’s “identification” role far outweighed its “exclusion” role; this edition, 36.4% — over a third, so the frequency of running into disclosure tags while browsing nearly doubled (share terms, not absolutes), and the tag begins sliding from “marking a minority” toward “describing the norm”; once past half, “used AI” is no longer distinguishing information, and “didn’t use AI” becomes the minority needing special marking — yet the system provides no verification means for “non-disclosers,” so players wouldn’t even have a reliable minority tag. The filter does not fail suddenly; it dulls cell by cell like this.

3.2 The Xiaoxiao Proposition: When Disclosure Becomes Default, Filtering Reverts to Quality Judgment#

This is precisely the institutional logic behind that line from Xiaoxiao, producer of Flower Crown Strategy. Her exact words: “The AI trend is unavoidable. In the future, once the share rises further, the AI content disclosure option may lose its meaning, and users will ultimately judge by quality alone.”

Taken apart, the line stacks three judgments: a trend judgment — the share’s rise is unavoidable; an institutional judgment — the disclosure option’s information content will be diluted to the point of meaninglessness; a substitution judgment — what ultimately carries the filtering function can only be quality itself. Against this edition’s 36.4%, the first judgment is being validated, the second’s projection path is the table above, and the third has already shown its outline in this edition’s charts (expanded in the next section).

As someone who has long followed AI systems, I would add: this is not a dilemma unique to gaming. Any “voluntary-disclosure” information system — origin labeling, ingredient lists, content ratings — depends on the premise that “the act of disclosure itself carries information”; when disclosure coverage approaches ubiquity, the system degrades from “filter” to “registry.” What is special about the game industry is the speed of arrival: four months, from 19% to 36.4%.

3.3 The Floor Problem: Unmeasurable Dark Numbers#

One harder problem remains: we cannot even confirm “the true share today.” The system has no automated detection and relies on voluntary reporting, which means two things — the true share may be higher than 36.4%, but by how much is unmeasurable; and among games that did not report, some used no AI at all and some used it without saying, and the two cannot be distinguished. Whether Valve will tighten the disclosure system, add filters, or penalize non-reporters: no official word as of writing; a fully open question, and this article makes no prediction. The player-side response (petitioning an “exclude AI” filter) and the developer-side response (cancelling public demos to prevent plagiarism) were detailed in the previous piece; the next section looks at them once more from the institutional angle — they happen to constitute the two forms of disclosure-system “failure.”

Taking the “dark number dilemma” apart into three layers shows exactly which cell the problem sits in:

Statistics, reporting, consumption — every layer rests on the assumption “developers tell the truth,” and the system provides no verification tool for that assumption. When a filtering system’s input end is unverifiable and its output end is inoperable, its middle number (36.4%), however precise, can only serve as a trend reading, not a filtering basis — that is the weight of the words “dark number” in this section’s title: what we don’t know may be larger than what we do.

4. Two Forms of Institutional Dissolution: One Turned Selling Point, One Turned Evasion#

4.1 Form One: AI-Free Becomes a Selling Point, Yet Disclosure Carries No Commercial Penalty#

This Next Fest had a telling tension: per analysis by TheDailyGame (citing a TheGamer survey), the event’s top-selling games were mostly AI-free works — players seemed to be rewarding “works without AI” with their purchases; yet at the same time, the overall disclosure share kept rising, and games that disclosed AI did not suffer a universal commercial penalty.

Setting these two facts side by side, the message is: players’ backlash is real (some players said the event was being “utterly ruined” by AI), but it has not yet converted into a purchase-decision standard covering the whole shelf. AI-free is turning from a niche stance into a recognizable selling point, while the disclosure tag itself remains in the position of “notifying” rather than “judging.” This is another manifestation of information dilution: when players cannot filter by the tag, they must rebuild judgment themselves from charts, word-of-mouth, and off-store-page information — the filtering cost is shifted back onto players.

4.2 The Real Cost of Backlash: The Shogun 2 Remaster Lost Its Store Listing#

Does player backlash carry a hard cost? There is one recent sample. Days before this Next Fest opened, the remaster of Total War: SHOGUN 2 drew “Overwhelmingly Negative” reviews over AI-upscaled assets and was ultimately pulled from Steam, replaced by the original — the AI controversy was severe enough that the publisher directly lost a store listing.

The significance of this episode is that it marks where the ceiling of backlash lies: not “an AI tag means it won’t sell,” but “when AI use makes veteran players feel cheated,” the backlash can overturn the entire product. The disclosure system is a separate matter — the remaster’s controversy centered on asset quality and nostalgia expectations, not on disclosure or its absence. Chasun has already covered the full arc of the episode site-side in “SEGA Emergency-Delists Total War: SHOGUN 2 Complete Edition Within Two Days: The First Major Wreck of the AI Retro-Remaster Route”; rather than repeat it, I’ll take only its implication for this piece: player acceptance of AI content is not an abstract “yes” or “no” but tightly bound to product expectations — this “quality judgment first” case is precisely a footnote to the Xiaoxiao proposition.

4.3 Form Two: Public Demos Cancelled, and the Disclosure Data Is Losing Its Observation Window#

If 4.1 is the disclosure system turning into a “selling point,” the other form of dissolution is more hidden: the data source is being actively narrowed. Two moves at this Next Fest — Digital Sun, developer of Moonlighter, deciding its new game ReVamp would have no public playtest, first running a limited community test instead, fearing a public demo would be quickly dissected and copied with AI; and Alliance Arts cancelling the public demo of Drunken Goddess Reflux scheduled for October 19 and switching to private play sessions, explicitly citing fear of AI imitation — look like anti-plagiarism self-preservation (the previous piece detailed the motives and mechanics), but framed within the disclosure system, they also cut away something else: the public playable build.

Much of players’ observation of a game’s AI use comes from direct experience during play: the texture of textures, the traces in voiceover, the style of the text. The fewer public demos, the less experience-based judgment there is, and the weaker players’ ability to cross-check disclosure information. Anti-plagiarism and anti-observation are two faces of the same action in these two cases.

4.4 The Player Side: A Shelf That Cannot Be Filtered#

Finally, let’s lay out the player-side situation plainly. Reddit community complaints during this Next Fest concentrated on two points: too many participating games, and heavy sameness; the community suggested adding an “exclude AI” filter, but Next Fest currently has no AI tag filter. In other words, even setting aside information dilution, players right now cannot perform even the most basic operation of “excluding by tag” — they want to filter but have no tool; and if the share keeps rising, even with a tool, the tag’s own value will decay. Tool absence stacked on signal decay is the player’s complete predicament: an unfilterable shelf, and a pile of increasingly indistinguishable goods.

Walking through the player’s actual decision process at the shelf shows exactly where the missing tool bites:

• Step one, look at the scale: 7145 participating games, the most of any edition — impossible to browse fully, so a filtering tool is the first requirement;
• Step two, want to exclude: the community suggests an “exclude AI” filter — the system provides no AI tag filter, so this path is closed;
• Step three, settle for manually reading disclosure tags — at 36.4%, that means manually combing nearly 2600 store-page statements, a cost too high to be realistic;
• Step four, retreat further to charts and word-of-mouth — the fact that top sellers are mostly AI-free works offers a rough proxy signal, but charts cannot cover the long tail.

After four steps, players are squeezed into the last one: coarse-filter with charts and word-of-mouth, fine-filter with playtime. This neatly explains the phenomenon observed in 4.1 — top sellers skew AI-free while disclosure sets a record, without contradiction: filtering hasn’t disappeared; it has merely retreated from “the tag the system provides” to “costlier, player-built signals.”

4.5 Three Forms of Failure, Compared#

To close the section: the disclosure system’s dissolution showed three forms at this Next Fest, each with a clear factual anchor:

The three forms point to the same endpoint by different routes: signal dilution is the data’s own decay, the selling-point turn is consumer-side repricing, and source avoidance is a supply-side contraction of the observation window. Their combined effect is the disclosure system sliding from “filtering infrastructure” toward “a mere registration formality” — and nothing institutional is currently pulling it back.

5. Conclusion: When Most Games Use AI, What Can Players Still Filter By?#

Having worked through five sections, back to the question in the title: when most games use AI, how much information does disclosure still carry?

My answer: the disclosure system will not disappear, but it is turning from “filter” into “registry” — this edition’s 36.4% is the midpoint of that handover. In the 19% era, “discloses AI” was a minority’s self-declaration; in the 36.4% era, it is standard equipment on nearly half the shelf; and per Xiaoxiao’s judgment, as the share climbs further, “users will ultimately judge by quality alone.” The delisting of the Shogun 2 remaster and the clustering of AI-free works atop the best-seller charts show that quality judgment is already taking over the filtering role — only its cost has changed from “glance at a tag” to “go try it yourself.”

As the editor watching the AI line, I’ll be watching two specific things next: one, the disclosure share at the next Next Fest — if it approaches half, the first half of the Xiaoxiao proposition is confirmed; two, whether Valve adds an AI tag filter to the shelf — no official word yet, but Reddit’s petition has put the demand squarely on the table. How these two play out will decide whether “AI content disclosure” ultimately upgrades into finer-grained tiered disclosure, or quietly degrades into a registry nobody reads. Until then, the only filtering tools that don’t depreciate, for players, remain the old three: playing it, word-of-mouth, and honesty about your own tastes.

— Wubian, NTGame Editorial

Further Reading#

• Behind Steam Next Fest’s Record 8,202 Demos: AI Supply Glut, Player Backlash, and Demo Self-Preservation Under Clone Panic — the phenomenon side of the same event: player backlash and developer anti-plagiarism
• SEGA Emergency-Delists Total War: SHOGUN 2 Complete Edition Within Two Days: The First Major Wreck of the AI Retro-Remaster Route — the full account of the Shogun 2 remaster delisting
• Epic x Disney Delay: Going All-In on “AI-First” Development — “No More Hand-Written Code” and the Developer Growing Pains Behind Big-Publisher AI Workflow Transformation — vendor transformation with AI as a production tool
• Google Playground and Unity: One-Sentence Playable Games and the Ecosystem Shock of AI Game-Creation Platforms — an ecosystem projection after the AI creation barrier collapsed
• AI “Reviving” Dead Games Goes Mainstream: Microsoft Builds a Decompilation Team and the Legal Limits of Preserving Old Games — AI as a preservation means and official co-optation