The Worlds Report - Issue #6
- Roblox demand fell 7.7% week over week on a like-for-like basis, and Horror was the only genre that grew (+9.6%). The fall landed inside Q3, the quarter that is historically Roblox's seasonal high point, so the calendar does not explain it.
- Roblox said what changed, inside the window. On August 5 the company said discovery now optimizes for measured long-term retention and accepts a lower frequency of impressions for games that emphasize short-term monetization, concentrated in the US under-13 audience, and a creator-facing post followed on August 6. Treat this week's genre shift as event-driven and possibly temporary: the company describes it as a test, it is a ranking gradient rather than a switch, and full rollout is late August or early September.
- Shipping an update stopped predicting direction on Roblox. Seven of the eight biggest decliners shipped a content update inside the window and fell anyway (Grow a Garden 2 -58.0M new visits, Animal Hospital -52.0M, Candy & Chocolate Keyboard Escape -51.9M). The biggest accelerator shipped nothing: Murder Mystery 2 (+44.1M) was last updated July 23 and was first published in 2014.
- Fortnite's GTA-6 theme is finished. Zero GTA-VI islands appear in the top-15, down from five last week, and the Simulator bucket that holds those driving islands fell from 20.4% to 10.2% of CCU. No Fortnite patch and no ecosystem release notes landed in this window, so this is a three-week-old hype wave decaying rather than a fresh event.
- Scorecard: 5 hit, 0 partial, 1 miss. Running 36 hit, 8 partial, 4 miss. The miss is call 3, which said the Roblox Horror share was borrowed and would give it back. Horror rose to 15.4% instead.
Running record: 36 hits · 8 partials · 4 misses
Last week's six calls, graded against this week's numbers. Five hit, one miss. The miss is the more useful of the two results, because it broke for a reason we were not modelling at all. Call 1 hit and carries an asterisk we are adding ourselves: its metric passed while the reasoning under it broke, because content updates showed up on both sides of the board this week.
- ✓ HIT
On Roblox, the growth event is an update to a base you already have. Build for a few weeks, then land a named content update rather than betting everything on a cold launch. MED-HIGH
Met, and the metric passed while the reasoning under it broke. Three of the top-5 accelerators qualify: Driving Empire (first published 2019, updated August 8, +16.6M new visits), Flee the Facility (2017, August 7, +15.7M) and Dandy's World (2024, August 9, +12.0M). The thesis did not survive the week. The biggest accelerator on the board, Murder Mystery 2 (+44.1M), has shipped nothing since July 23, and seven of the eight biggest decliners did ship a content update inside the window and fell anyway. Shipping an update no longer predicts which way a title moves, so the corrected version is re-registered as call 2 this week.
- ✓ HIT
Do not launch a keyboard-escape or brainrot meme-sim. The format is spent, and updates have stopped reviving it. MED-HIGH
Both legs met. Roblox Simulator/Tycoon came in at 23.9% of new visits, below the 25% leg and a seventh straight weekly decline (42.6, 38.6, 34.4, 32.4, 29.7, 25.0, 23.9), and Candy & Chocolate Keyboard Escape lost 51.9M new visits despite republishing on August 10. One qualifier stated plainly: Steal a Brainrot turned positive this week at +9.4M, so the meme cohort is thinning rather than finished.
- ✗ MISS
Treat the Roblox Horror share as borrowed. It rests on one title's July 23 update with nothing fresh behind it this week, so expect give-back. MED
Both legs failed. Roblox Horror rose to 15.4% of new visits from 13.7% instead of falling below 13%, and Murder Mystery 2 was the biggest accelerator on the board at +44.1M rather than a decelerator. Its last content update is still dated July 23, a third consecutive week unchanged, so nothing new rescued it. This call was published without an escape hatch on purpose and the miss stands in the denominator. The mechanism is what we got wrong: we modelled the Horror share as a decaying content tail, and what moved it was a change to how Roblox routes attention, announced on August 5 and August 6.
- ✓ HIT
On Fortnite, run the red-vs-blue PvP factory. It is the platform's resting state, and the GTA-6 theme has been giving share back for two weeks. HIGH
Both legs met, the second by a wide margin. Fortnite PvP at 53.7% of CCU plus Obby at 18.7% came to 72.4%, past the 60% leg. No GTA-VI island was the top island by peak CCU, and none appears anywhere in the top-15, down from five last week.
- ✓ HIT
Take the Survival-to-Fortnite lane and build it short. Six straight weeks vacant, and the session-length gap is the reason it stays that way. MED
Zero Survival/Co-op islands in the Fortnite top-15, a seventh straight week open. The genre rounds to 0.0% of Fortnite CCU, against 11.9% of Roblox new visits on the other side of the gap.
- ✓ HIT
On VRChat, ship the update and expect it to decay. Favorites are the stored demand and an update is how you withdraw against it. MED
FISH ǃ [PYRAMID], the fishing world that got there on a content update rather than a fresh launch, was still drawing a crowd this week, which is what the call asked for. The decay half arrived on schedule: its crowd has thinned every week since that update landed. Going forward these calls are phrased in plain descriptive terms rather than in tiers, because our VRChat reporting stays qualitative and a tier is a figure written in words.
Three things to know about this chart. It is Roblox only this week, because Fortnite has no usable cross-week genre comparison: island codes rotate through discovery and the turnover between the two weeks was effectively complete, so any Fortnite week-over-week would be comparing two different sets of islands. Genres under 1% share are left off so a small base cannot own the axis, which removes Roblox Party (0.7%) and Other (0.9%). And the two Roblox measures sit on different bases by design: share covers this week's full set of games, while the week-over-week compares only games present in both weeks. They agree everywhere except Obby, whose 2.0% share is carried almost entirely by one large returning obby course with no prior-week figure behind it; compared like-for-like that genre is -20.3% and holds 0.2%.
Survival / Co-op PvE
11.9% of new visits
Roblox
0.0% of Fortnite CCU
Fortnite Creative
One gap closed this week and it is worth recording as a closure rather than quietly dropping it. Horror was flagged here last week at 13.7% on Roblox against 4.0% on Fortnite. This week Fortnite Horror tripled to 12.0% against Roblox's 15.4%, and the genre now travels. Escape rooms work on both platforms and the spread is no longer exploitable.
That leaves Survival/Co-op as the one flagged opening still worth building into, in its seventh straight week of near-total vacancy: 11.9% of Roblox new visits against 0.0% of Fortnite CCU, with no island in the Fortnite top-15. Seven weeks is long enough that the vacancy is a property of the platform rather than an oversight, and the property is session length. Fortnite's median session ran 13.2 minutes this week, and a survival loop that needs half an hour to pay off has nowhere to land inside that.
What is new is that the shape is finally visible. ZOMBIE BOSS FIGHT (CO-OP) cleared the Fortnite top-15 at 281 peak CCU while holding 25.5 minutes per player, 1.9 times the platform median, and the classifier files it under Horror rather than Survival. It got there by being a bounded boss encounter with an end state rather than an open survival grind. That is the version of the lane the platform will actually pay for.
How a creator exploits it: build the deep version on Roblox, where a returning base compounds, and ship a short co-op boss encounter into the Fortnite lane, one that resolves inside roughly ten minutes and has a win condition. The shelf over there is still empty.
Three shocks, three shapes
This week each platform absorbed a different shock, and the shape of the platform decided who absorbed it. On Roblox (winner-take-all, where the top 10 hold 55.7% of new visits and the top title alone holds 11.7%) the shock was a change in distribution, and in a winner-take-all curve distribution is the product. If the announced ranking change is what moved the week, and it is the only dated candidate available, then the response is the informative part: demand did not redistribute across the tail, it contracted 7.7% outright and pooled into the one genre whose leader best matches the new criterion. A 2014 title gained 44.1M new visits without shipping anything, which is only possible where a handful of surfaces route most of the demand. On Fortnite (long tail, where the top island holds 3.9% of CCU, the top 10 hold 26.0%, and the median island peaked at 6 on roughly 13.2 minute sessions) the shock was the absence of one. No patch, no event. A long tail with no compounding base reverted to its default, and the red-vs-blue factory refilled the top-15 with islands that did not exist a week earlier. On VRChat there was no platform event at all this week, and the resting shape showed itself cleanly: the world that had shipped a content update kept giving back the crowd that update bought it, while the hangout that shipped nothing carried the same regulars it had a fortnight earlier. The creator-facing version is that the three platforms punish different mistakes. On Roblox the expensive mistake is building for the ranking function that existed when you started, since a long build cycle can outlive the criterion it was optimized for. On Fortnite the expensive mistake is building anything you intend to keep, because discovery guarantees your island is disposable, so amortize across many cheap ones. On VRChat the expensive mistake is shipping an event where a place was needed.
Roblox
top-10 share 55.7% top-1 11.7%
Winner-take-all. The top 10 hold 55.7% of new visits and the top title alone holds 11.7%. The median world took 46,832 new visits. Read these as levels rather than as a week-over-week move: the mix of games behind the comparison changed enough this week that the change in concentration is not a sound read.
Fortnite Creative
top-10 share 26% top-1 3.9%
Long tail. The median island peaked at 6 CCU on roughly 13.2 minute sessions. A fresh niche island is viable and disposable: the top island by peak CCU this week first showed up in our figures on August 6.
VRChat, in brief
No VRChat platform event of any kind landed this week, so everything here is organic behaviour rather than a response to something the platform shipped. FISH ǃ [PYRAMID] is a fishing game that grows by shipping content drops, and this week it completed the arc of one. The crowd its [PYRAMID] update pulled in has thinned every week since it landed. Popcorn Palace, a chill and sleep hangout built around private rooms, shipped nothing at all over the same stretch and is carrying the same regulars it was carrying a fortnight ago, with nothing having happened to it either way. The places that hold on VRChat are mostly not games. Prismic's Avatar Search is a utility, somewhere you go to find an avatar rather than to do anything, and people keep saving it to their favorites long past the point where a tool should have saturated. The same shape repeats across YTS 2.1 for watching YouTube with other people, 1's Optimized Box and Avatar Testing Chamber for mirrors and rigging checks, VRChat Home, and a row of places whose whole proposition is somewhere comfortable to be: everyone is asleep except you, Just A Cabin, place to eat cereal, and the Japanese-language Himari Ryokan. Even the games that hold (Murder 4, Among Us, Terrors of Nowhere) are short social rounds you play with whoever is already standing there, closer to a hangout with a rule set than to something you queue for.
A VRChat call we should not have made
One thing we said last week did not happen, and it is worth saying so. We flagged a race world that was being saved heavily by people who were not actually in it, and read that as stored demand banking ahead of a spike. This week the saves stopped and the world stayed empty. Favorites accruing ahead of attendance is sometimes a pre-load and sometimes just a world people bookmark and never open, and one week does not tell the two apart. We should not have called it. The Spider-Lair: Evolved and Ghost Hotel are sitting in that same ambiguous spot now, and they are worth watching rather than calling. What a builder should take from the week: build the place. If you are shipping updates to hold a crowd, price in that the crowd leaves between them. FISH ǃ [PYRAMID] is a good world and it still emptied out steadily for a fortnight. A comfortable room with private instances, or a tool people need often enough to keep in their favorites, earns attendance without a content treadmill.
Analyst notes
Roblox demand fell 7.7% this week on a like-for-like basis, and it fell during the quarter that is normally the strongest one on the platform. Exactly one genre grew. That combination is the whole issue, and the reason it is worth your time is that Roblox told us what changed while the week was still running: on August 5 the company said discovery now optimizes for measured long-term retention and accepts fewer impressions for games that lean on short-term monetization, and on August 6 it said the same thing to creators directly.
We are calling that event-driven rather than structural, which changes what you should do about it: wait for the rollout to finish in late August or early September before committing anything expensive to a rebuild. The company calls it a test, describes a gradient rather than a switch, and our own numbers only cover one week of it. The accelerator side of the board does not fit the story cleanly, which is exactly what a gradient looks like in data.
What does not need waiting on is the update finding. Seven of the eight biggest decliners shipped a content update inside the window and fell anyway, while the biggest gainer on the board has shipped nothing since July 23. If your plan for the next quarter is a content update on a base you already have, that plan needs a second leg now.
Each call below gets graded in next week's issue against the metric named on it.
-
Design for retention before monetization. Roblox changed what discovery rewards inside this window and said so publicly, so build the first session and the return visit before the earner. MED-HIGH
Graded next week on: At least two of the Roblox top-5 accelerators are titles first published before 2020, and Horror/Escape holds at least 15% of new visits on a like-for-like basis.
-
Stop treating a content update as the growth lever on Roblox. This corrects our own call from last week. MED
Graded next week on: At least three of the top-5 Roblox decelerators shipped a content update inside the window. If that happens, updating again failed to prevent decline and the lever is confirmed dead.
-
Do not build a 2025-vintage collect-sim (farm, pet or fishing) as an evergreen hedge. This is the advice we gave in last week's issue, and it did not survive the week. MED
Graded next week on: Grow a Garden 2, Fish It! and Fisch are all net decelerators again.
-
On Fortnite, drop the GTA-6 theme and build an escape room instead. They are cheap to build and they fit the platform's session length. MED
Graded next week on: Zero GTA-VI islands in the Fortnite top-15, and Horror/Escape holds at least 8% of Fortnite CCU. Graded on within-week composition, since Fortnite has no usable week-over-week comparison.
-
Take the Survival-to-Fortnite lane as a short co-op boss fight rather than a survival grind. The shape that cleared the top-15 this week is a bounded encounter. MED
Graded next week on: Still zero Survival/Co-op islands in the Fortnite top-15 by peak CCU.
-
On VRChat, build the permanent destination rather than the update event. The crowd a content drop pulls in leaves again over the following weeks. MED
Graded next week on: The crowd at FISH ǃ [PYRAMID] thins for a third straight week, and Popcorn Palace is still carrying its regulars without having shipped a content update.
- WARN The two Roblox columns sit on different bases, by design. Share covers this week's full set of games; week-over-week compares only games present in both weeks. The mix of games behind the comparison shifted enough this week that the like-for-like direction is the sound read, and it is what every Roblox claim in this issue rests on.
- WARN This week's Roblox genre shift is event-driven and we are labelling it that way. Roblox announced a discovery change on August 5 (earnings call) and August 6 (creator-facing post) that re-aims recommendations at measured long-term retention and accepts fewer impressions for games emphasizing short-term monetization, primarily in the US under-13 audience. The company frames it as a test and a ranking gradient rather than a switch, and full rollout is late August or early September. One week's move is consistent with it, not proof of it, and it may give back. This is not a structural lane opening.
- INFO Roblox Obby's 2.0% share is one returning title, not a surge. A single large obby course re-entered the comparison this week with no prior-week figure to compare against, which is why the raw change for that genre reads as a jump. Compared like-for-like, Obby is -20.3% and holds 0.2% of new visits. Nothing else on the board has that shape.
- WARN No per-genre week-over-week is shown for Fortnite. Island codes rotate through discovery and the turnover between the two weeks was effectively complete, so any cross-week genre change would compare two different sets of islands rather than the same ones. Fortnite claims in this issue rest on within-week share, top-15 composition, and per-island movement inside the week, all of which remain sound. Total Fortnite CCU is also not comparable week over week for the same reason.
- INFO The Fortnite figures themselves check out; what degraded is how many islands report one at all. A spot-check of the week's fastest riser against a public third-party island page came in within 4.4% of our peak CCU figure, on a peak the page dates to August 9.
- INFO Read the concentration figures as levels, not as moves. Roblox top-10 share of 55.7% and Fortnite top-10 share of 26.0% are sound as levels. Both platforms' week-over-week movement in concentration is a comparison artifact this week and is not reported.
- INFO The Roblox spot-check passed and it is the week's load-bearing check. Murder Mystery 2's live player count read within 8.7% of ours in the falling direction, which is normal drift for a world that size between a mid-morning and a midday reading, and its content-update date returned July 23 for a third consecutive week. That single field is what turns call 3 into a clean miss rather than a void. Fifteen content-update and first-published dates used in this issue were fetched individually from Roblox's public listings.
- INFO VRChat was checked on favorites and not on presence. A world's favorites count reconciled to within 0.001% of the platform's own public figure. Live presence is not verifiable without a logged-in session, so the VRChat read in this issue is qualitative, which is also where we keep it by policy.
- INFO Roblox's business context has moved and it belongs next to the genre numbers. The company lost roughly $9B in market value on August 5, has guided Q3 bookings down 14% to 18% year over year, and paid out $363M through the developer exchange against a $477M peak. The pool creators are paid from is shrinking while the ranking function changes underneath them.
- INFO Nothing on the Fortnite side landed in this window. The build held at v41.30 with no patch, no downtime incident and no ecosystem release notes. Chapter 7 Season 4 is dated August 20 and sits outside this window entirely.
- INFO Method caveats. Fortnite island codes rotate through discovery, so per-island week-over-week is not possible. The Fortnite Simulator bucket includes GTA-style driving islands, which is why that share swings on a GTA hype cycle rather than on anything about simulators. VRChat reads stay qualitative and carry no per-world figures. Newly seen means a world first showed up in our figures, not that it launched.
Get the next issue by email
Every Tuesday, about 10:00 ET.
Free and weekly, with double opt-in. Unsubscribe any time.