The AI Disclosure Report Steam Next Fest, June 2026
AI is not a binary. The studios that disclosed AI and still performed used it least visibly, on surfaces players never inspect. This report maps the disclosure numbers, the performance gap, and the practical line between AI as an asset and AI as a liability.
Published June 2026
Edition June 15–22, 2026
Basis GDCo · SteamDB · developer statements
4,382
Registered demos (GDCo pool)
26.5%
Of demos disclose AI (GDCo pool)
1 of 10
AI-disclosed in top-played
~8,682
Total event entries
The frame. Two populations, two rates, often conflated. Across the 4,382 registered demos GameDiscoverCo tracked, 1,163 carried the AI disclosure, about 26.5 percent. Across the full event of roughly 8,682 entries, near 1,700 disclosed, about 20 percent. Demos disclose at a higher rate than the event as a whole, because shipping a playable build is more likely to put visible content on screen. Both numbers are real. Most coverage cites whichever is larger without saying which population it describes.
The Scale
The field outgrew the festival.
Volume became the problem.
June 2026 was the largest Next Fest on record by a wide margin. The growth is not incidental to the AI story. It is the AI story, because the same tooling that lets a small team ship a demo at all is what filled the field past the point where discovery still functions.
Edition
Total demos
AI disclosed
% disclosed
Basis
October 2024
~2,900
~400 (est.)
~14%
Estimated
February 2025
~2,300
Not tracked
~15% (est.)
Estimated
June 2025
~2,645
Not tracked
~16% (est.)
Estimated
October 2025
~2,960
504–507
~17%
Counted
February 2026
~3,455
~750 (demo subset)
~17–21%
Counted
June 2026
4,382 demos / ~8,682 entries
1,163 demos / ~1,700 entries
26.5% demos / ~20% entries
GDCo + SteamDB
The apples-to-apples number. The only clean consecutive-edition comparison uses GameDiscoverCo's demo pool with one methodology: 21.2 percent disclosure in February 2026, rising to 26.5 percent in June 2026. Same measurement, same Valve-defined demo population, back-to-back editions. That five-point jump is the real trend, not the wider event-pool figures that shift with how each outlet draws the boundary.
For platform context beyond Next Fest, roughly 7,818 Steam titles disclosed generative AI use in the first half of 2025 alone, against about 1,000 in all of 2024. By 2025, close to one in five new monthly Steam releases carried a disclosure. In the week ending June 14, 2026, immediately before the festival, more than 300 games shipped on Steam and 120 of them disclosed AI, a 40 percent rate for new weekly releases, per GameDiscoverCo tracking. The festival rate trails the new-release rate, which is the leading indicator worth watching.
What Is Actually Disclosed
The disclosure tells you almost nothing.
One checkbox, a dozen meanings.
A single disclosure covers a placeholder texture replaced before launch and a game built entirely from generated assets. That is the system's central flaw and the reason raw disclosure counts mislead. Here is what sits underneath the checkbox, ordered by how often it appears.
Visual assets (art, textures)
~60%
Marketing / store page art
High
Audio (music, SFX, voice)
Smaller
Translation / localization
Common
Writing / dialogue
Lower
Code / game logic
Exempt*
Live-generated content
Rare
The split that matters. Visual content drives roughly 60 percent of disclosures. Audio is a smaller share but generates the loudest reaction, because a synthetic voice is unmistakable in a way a generated texture is not. The category a studio picks determines its risk far more than whether it discloses at all. *Code assistance stopped requiring disclosure under the January 2026 policy rewrite.
The Performance Gap
AI-disclosed games miss the top tier.
Every edition, the same shape.
The clearest performance signal is the official most-played list. AI-disclosed games appear there at a fraction of their share of the field, and the gap holds across every edition with data. This is the one performance claim the public numbers actually support.
Edition
AI in top 10
AI in top 100
Pool rate
Gap
October 2025
0–1
~3 in top 50
17%
~3–4×
February 2026
Not specified
10 of 100 (10%)
17–21%
~2×
June 2026
1 of 10 (10%)
Not yet counted
26.5%
2.5×+
In February 2026, a manual review of the top 100 demos found exactly 10 carrying AI disclosures, a 10 percent rate against a 17 to 21 percent pool. Four of those ten disclosed AI only for placeholder assets already slated for replacement. In June 2026, the top 10 most-played carried a single AI-disclosed title against a 26.5 percent demo pool, a gap of more than two and a half times.
The one that made it: Embers of the Uncrowned, rank #10. A dark-fantasy MMORPG from Nexon, a publicly traded publisher with roughly 9,800 employees. Its disclosure is broad and forward-looking, covering visual content, marketing, live chat translation, and partial localization, with explicit deference to human authorship. It performed because the demo shipped a polished, multi-hour build and because Nexon's name carries genre trust that offsets disclosure hesitancy. That is brand immunity, not a model an indie can copy.
Data note: The 26.5% disclosure rate applies to GameDiscoverCo's tracked pool of 4,382 eligible demos. For the full event of roughly 8,682 entries, the disclosure rate is nearer to 20 percent — both figures are valid but refer to different populations.
Wishlist performance: the honest gap in the data
No public dataset compares AI-disclosed and non-disclosed games on wishlists across a Next Fest. Valve does not publish AI-stratified figures, and no independent survey has done the stratification. What is observable: none of the top-tier wishlist earners in February or June 2026 carry visible AI disclosures, and the two most-played demos of June 2026 both carry none. That is a pattern, not a controlled comparison, and this report will not present it as one.
The counterexample is not one game. It is the top two. The two most-played demos of the largest Next Fest on record were both handcrafted and both carried no AI disclosure. BOMBANANA, from indie team Lefto Studio, finished #1 with more than 675,000 players, over a million demo library adds, and a 40,000-plus peak concurrent on SteamDB. IRON NEST, a two-person team, finished #2 at 99 percent Overwhelmingly Positive with 124,622 wishlists in five days. Against a field disclosing AI at 20 to 26.5 percent, the chart topped out on craft. That is a pattern, not a cherry-picked exception. It still does not prove AI lost, because no one measured the AI cohort against a control, but it is a far harder data point than any single breakout.
The Avoidance Effect
The penalty players impose themselves.
The policy never had to.
Valve's disclosure carries no formal penalty. Players built one anyway. The mechanism is consistent across independent reports and it compounds through the festival's own ranking algorithm.
Step 1
Check
Informed players scan for a disclosure before engaging, often via a browser extension.
Step 2
Skip
On finding one, most skip the demo regardless of the stated scope of AI use.
Step 3
Signal
Lower playtime and conversion feed the algorithm a weak-engagement signal.
Step 4
Bury
After Day 3, ranking shifts from randomized to engagement-weighted. The signal compounds.
A Kotaku test of 16 games from the June 2026 main hub found 10 tripped an AI-disclosure warning from a browser extension, a 62.5 percent hit rate among featured games. Treat that as one spot check, not a field rate, the sample is sixteen games. The behavioral conclusion holds regardless of the exact percentage: disclosed games face real, measurable avoidance from exactly the engaged genre fans a serious studio is trying to win.
The avoidance reaches the buy side. The behavior is not confined to players. Publishers have stated publicly that AI capsule art is an immediate reason to skip a demo while browsing. When the people writing checks pass on the capsule image, the penalty is no longer just a player preference. It is a market signal that forms before the gameplay is ever seen.
The discovery cost lands on everyone. At roughly one in four demos disclosing AI, a quarter of the randomized early-festival exposures risk generating a negative signal from informed players. That degrades the algorithm's ability to identify which games deserve amplification, which means the handcrafted unknown loses its best window, the randomized early-fest slot, to the noise. The damage is not the AI games. It is what they do to discovery for the games below the median.
The Playbook
The studios that used AI and still won.
They used it where no one looks.
Here the report turns from the crisis to the pattern underneath it. Every team that cracked the top 100 carrying an AI disclosure shared one trait: the AI footprint was narrow, confined to non-hero surfaces, and almost always paired with a replacement plan or a human-finalization statement. The disclosure was something a player met after forming a positive impression of the gameplay, never before. What follows is drawn from developer statements and industry reporting, not independent audit, and is framed as such.
The Warhounds lesson, in one line. Everplay disclosed AI portraits as temporary placeholders, named the exact surface, and committed to replacing them with 100-plus hand-made portraits. It still drew an AI complaint in nearly every second negative review. Then they eliminated the AI entirely, rendering portraits directly from each character's in-game 3D model, a system that solves character customization better than static AI art ever could. Community tolerance runs lower than formal review scores suggest, and the studios that performed treated AI as something to engineer out, not defend.
The one consistent failure mode
Across every profile, the punished surface is the same: AI art on the storefront, capsule images and screenshots, where it is most visible and triggers avoidance before any gameplay impression forms. None of the performers led with AI on a hero surface. The ones that did are not in the top 100.
Use Tiers
Map the AI to the player's eye.
Surface area is the whole risk.
Synthesizing across the performers, AI use sorts into three tiers by how directly a player encounters it. The teams that performed lived almost entirely in Tier 1, with limited and clearly communicated Tier 2. None operated primarily in Tier 3.
Tier 1 · Lowest risk
Invisible or justified
Live-service localization, procedural text for simulation scale, AI-assisted painting with human finalization, code assistance (now exempt). The player never meets it as art.
Tier 2 · Moderate risk
Visible but defensible
Demo placeholder portraits with an executed replacement plan, ambient music in a gameplay-first genre, translation with human review. Survivable with transparent disclosure.
Tier 3 · High risk
Visible and controversial
AI capsule and marketing art not clearly replaced, AI voice for player-facing characters, identifiable AI dialogue, and any undisclosed use. This is where demos die before momentum.
The test, stated plainly. If the AI is in the backend generating content players never directly encounter as art, it is almost certainly an asset. If it is in the first image a player sees on the store page, it is almost certainly a liability. The disclosure is identical in both cases. The outcome is not.
Team Size
Scale changes what AI is for.
Gaps, systems, or shortcuts.
How a team uses AI tracks its size with surprising consistency. Small teams cover discipline gaps they cannot staff. Large teams scale systems they already run. The teams that get punished use AI as a visible shortcut on inspected surfaces, and that failure is independent of size.
Any AI on the capsule or hero character is seen instantly
Best practice
Human art for the first five seconds; AI for everything secondary
Small · 5–20
Neurona, Everplay, Triple Espresso
AI covers
Specific scale problems: procedural faces, personality text
The edge
Veteran teams know which disciplines need human craft
The warning
Even disclosed, replaced portraits drew heavy review backlash
Large · 50+
Nexon
AI covers
Live-service pipelines: localization, scaled systems
The buffer
Established brand trust offsets disclosure hesitancy
The catch
Depends on credibility built over years; indies cannot borrow it
The review-volume signal. One analysis of nearly 10,000 paid Steam titles released across 2025 reported that an AI disclosure tracked with materially fewer reviews on average, and that the effect ran stronger for higher-ambition games, not weaker. This report cites that as reported, not independently verified. If it holds, the mechanism is intuitive: the players most likely to punish AI are the engaged genre fans who write reviews and build communities, exactly the audience a serious studio needs. Casual players never check. The core audience does.
Valve Policy
A self-report system at the wrong scale.
Visible enough to hurt, too weak to inform.
Every structural flaw in the disclosure system traces to one design choice: it is voluntary, unenforced, and buried at the bottom of the store page. The January 2026 rewrite clarified scope but left the enforcement gap untouched.
2024
The policy is born
Mandatory disclosure for AI-generated content in the game, store page, or marketing. Two categories established: pre-generated and live-generated, the latter requiring guardrail disclosures.
January 2026
The critical rewrite
First major revision since 2024. Clarified that AI-powered dev tools used for efficiency, code helpers, ideation, do not require disclosure. The line: disclose what ships or appears in marketing; exempt internal tooling that produces no player-facing output. Added a Steam Overlay button to report illegal live-generated content.
Ongoing
The enforcement gap
No automated detection. No audit. No penalty for non-disclosure. A developer can ship AI art throughout a game, decline to disclose, and face zero consequence unless community members catch it.
The undercount is structural. Because compliance is voluntary and unpenalized, the 20 to 26.5 percent figure is almost certainly a floor. The true share of demos using AI in some form, including the undisclosed, is unknown, and observers consistently describe it as higher. The disclosure system produces the worst outcome available: visible enough to drive player avoidance and artist resentment, too weak to actually inform a purchase or filter the content it exists to label.
What Valve has not done
No AI filter or hide toggle, the single most-requested feature for two years running. No enforcement mechanism or penalty. No quality review for AI-heavy demos. No badge near the top of the store page. No public response to the filter requests. Valve has historically resisted curation, preferring market dynamics to handle quality. A field of 8,600 demos is the stress test of whether that philosophy survives.
Council View
Decisions for October 2026.
The right question to ask.
The lesson is not avoid AI, and it is not use AI freely. It is map your AI use to the player's path through your store page and demo, and ask at each touchpoint whether the AI is visible. That single question separates the performers from the casualties.
01
Audit by surface
Walk your store page as a player would. Capsule, screenshots, trailer, first demo minute. Any AI on those surfaces is a Tier 3 liability. Move it or replace it.
02
Keep AI in the backend
Localization, procedural text, pipeline acceleration with human finalization. If the player never meets it as art, it reads as an asset, not a tell.
03
If you disclose, name and replace
Specific surface, explicit replacement plan, executed. The performers treated disclosed AI as something to engineer out, and said so.
What is changing by October. Per GameDiscoverCo analysis, Godot's share of the field rose from 9.2 to 12.6 percent, a tilt toward smaller handcrafted projects. Browser extensions that flag disclosures are spreading, which will deepen early-engagement drops for disclosed games. New-release AI rates near 40 percent point toward a festival rate approaching 30 percent. And curated alternatives to the mega-festival are being tested, a segmentation that could route around the saturation entirely. The studios that win the next edition will be the ones that understood the surface-area rule before the field did.
The reframe. AI raised the noise floor. Execution decided who got noticed. The disclosure debate treats AI as a moral binary, used or not. The performance data treats it as a question of placement, seen or not. For a studio planning October, only the second framing is actionable, and it is the one the winners already operate by.
About the Author
Abbas Saleem
Abbas Saleem is a Principal Consultant at Llama & Griffin Consultants, advising game studios, streaming platforms, and investment funds across six continents. He writes The Pattern Recognition: gaming industry intelligence 12 to 24 months before it becomes consensus. LinkedIn | Schedule a call