Always-On Residents and Evening Explorers: How Simplified Chinese and Japanese Players Use VRChat

Almost all play in VRChat is player-generated: players upload worlds and avatars, then meet, chat and join events in instances opened from those worlds. Social VR has no clear goal to beat, so "gameplay" is mostly a choice of who to be with, where, and doing what, and those choices are hard to see in the game's public data.

Companion tools offer a side window. VRCX and its fork VRCX-0 run on PC alongside VRChat and read the friend Feed, game logs and favorites to help players answer questions like "which instance are my friends in right now", "who was in that instance just now" and "which worlds did I visit last week". Which pages players open, and how often, reflects what they care about in VRChat. This report treats these pages and tools as measuring instruments. The subject is the players who use them, not the tools themselves.

This report covers only the two groups with large enough samples: Simplified Chinese users (CN) and Japanese players (JP). The research questions are:

1. RQ1 Rhythm: At what local times are the two groups online? Do they show all-night sessions or a weekend concentration?
2. RQ2 Frequency and duration: How long does each session last, and how often do players come back? What share of players "play all the time" and what share "drop by now and then"?
3. RQ3 Focus: Do players care more about finding friends, looking back on where they have been and managing relationships, or about changing avatars, exploring worlds and organizing communities? What kind of player does each type of feature represent?
4. RQ4 Cultural reading: Do these differences match known community cultures such as sleeping in VR, meetups and avatar editing?

VRChat is developed by VRChat Inc. It launched free on Steam in Early Access in February 2017^[1] and now supports PC VR, PC desktop mode, Quest and Pico, with iOS and Android added in October 2025^[2]. Desktop mode needs no headset and still gives access to worlds, avatars, voice and every social feature.

The basic unit of player activity is the instance. One world can host many instances, each holding up to 80 people, with visibility types Public, Friends+, Friends, Invite+, Invite and group instances^[3]. Players see on their friends list which instance a friend is in and can go straight there, which is called joining a friend's instance (Chinese players call it 跟车). They can also send an invite or request an invite. Four statuses, Join Me, Online, Ask Me and Do Not Disturb, decide whether friends can see and join their instance^[4]. Groups (launched November 2022) offer member and role management, announcements, events and group instances, much like clubs inside VRChat^[5]. Players can also favorite friends, worlds and avatars, and mute other players, hide their avatars or block them. There are only four server regions: US West, US East, Europe and Japan^[6].

The platform is still growing. Peak concurrent users reached 148,886 on New Year's Day 2026^[7], and a Japanese-language concert in February 2026 pushed the record to 158,192^[8].

Japan. VRChat does not publish data by country, but third-party traffic data suggests Japan is one of its largest single sources: Sensor Tower estimates that Japan accounted for 37% of about 650 million visits to the VRChat website in 2025^[9]. The ソーシャルVRライフスタイル調査 (Social VR Lifestyle Survey) by バーチャル美少女ねむ and Mila, which covers VR users only, found that the share of Japanese respondents using social VR "almost every day" was 51%, 48% and 51.7% in 2021, 2023 and 2025. Their sessions were shorter than in North America and Europe, though (sessions of 3 hours or more in 2023: Japan 44%, North America 75%, Europe 76%)^[10]. A Japanese industry survey found that logins cluster between 20:00 and 04:00 local time with a peak at 22:00, and that about three in ten respondents had fallen asleep wearing a headset, which is sleeping in VR^[11]. Academic studies have also documented what "falling asleep together" means in social VR^[12] and why people sleep in VR^[13]. The Japanese community has two other notable cultures. One is a large number of regular meetups and events: Virtual Market (Vket) 2025 Summer drew 1.35 million online visitors^[14]. The other is aesthetic and identity practice built around avatars^[15]. The author's own first-hand view as a VRCX-0 developer matches this: sleeping in VR is common, and the meetup and event culture is strong.

Simplified Chinese users. VRChat does not operate in China, and its website is blocked on Chinese networks^[16]. A Chinese community networking guide lists "third-party apps can't reach VRChat" as a common problem and recommends game accelerators^[17]. No reliable estimate of the player count exists. Qualitative research has described how Chinese players negotiate identity in VRChat^[18]. The author's observation is that whether tools can reach the VRChat API directly depends on the ISP and the time of day, and that Simplified Chinese players mostly focus on switching avatars and browsing worlds.

VRCX is a VRChat companion app open-sourced in 2019. It works through VRChat's Web API and local game logs and offers Friend Locations, Feed, Game Log, Instance History, favorites, Notifications and more^[19]. VRCX-0 is a fork of it. Table 1 maps the pages and tools used in this report to play motivations, and the later analysis of what players focus on follows this mapping. The mapping is based on what each page actually contains in the client source code.

User research on social VR relies mainly on interviews and surveys. Sykownik et al. surveyed the activities and motives of social VR users^[20], Chen et al. studied whether new VRChat players stay^[21], and Kondo and Hirose ran surveys inside three Japanese VRChat worlds^[22]. Studies based on behavior logs are fewer; Tsutsui et al. analyzed large-scale logs from the Japanese platform cluster^[23]. Research on traditional online games has built up more work on behavioral clustering^[24] and on comparing play styles across countries^[25], and some studies warn that "country" is only a rough proxy for culture^[26]. This report belongs to the behavior-log category and compares its results with the surveys above wherever possible.

The data comes from VRCX-0's anonymous telemetry. The observation period runs from 2026-07-01 to 2026-09-30 (92 days), and tool data covers August 14 to September 30. The telemetry contains no account, friend, world or instance information. This report uses only online records at "player × hour/day" granularity and open counts for pages and tools (Table 2).

All durations in this report are companion-tool online time, which is an upper bound on time spent in VRChat. Players usually keep the tool open while they play VRChat, but having it open does not mean they are playing. Section 5.4 and Section 6 discuss what this means.

CN group: the system or UI language is Simplified Chinese (zh-CN, zh-Hans), or the IP location is CN. IP alone would miss many players: only 63.0% of this group have a CN IP location, and the rest appear as Japan (14.2%), Hong Kong (9.7%), Singapore, the US and other exit nodes, yet 91.6% are in the UTC+8 time zone. Traditional Chinese users are not counted as CN.

JP group: the language is Japanese. In this group, 97.7% of IPs and 98.0% of time zones are in Japan. Widening the definition to "Tokyo time zone + Japanese IP + non-Chinese language" adds 42 players with an English UI to the JP group, and the main conclusions do not change (Appendix B).

Players with other languages and from other regions are more scattered, and English users come from many countries, so this report does not cover them.

Active days and active-day share. An active day is a UTC day with online records. The UTC day boundary falls at 08:00 local time for CN and 09:00 for JP, both in the morning trough, so it has little effect on the comparison. Active-day share = active days / observed days (counted from first appearance or from July 1). Frequency and segment analyses include only players observed for 28 days or more, so that newcomers are not compared with veteran players on active days.

Online time. Time is counted from the interval between two reports, with each interval capped at 30 minutes. Days with ≥22 hours online are always-on days.

Online sessions. When the gap between a player's adjacent client sessions is under 60 minutes (client updates, restarts), they are merged into one online session and treated as "one time online". Session length runs from first appearance to the last heartbeat, and the gap runs from the end of one session to the start of the next.

Hourly presence. The set of hours each player was online on each local date. A heartbeat counts the previous 30 minutes toward the hour in which it was sent, so the curves shift right by at most half an hour. Days with ≥20 hours online count as always-on days. The "non-always-on days" curve excludes them to highlight daily routines when players switch the client on and off by hand.

Page focus. Three measures complement each other: (a) coverage, the share of players who used the page during the observation period; (b) composition, the page's share of all of a player's page visits, independent of total activity; (c) re-check intensity, the number of times the page was opened on a day when it was opened at all.

Frequency segments. Trial (≤2 active days), sporadic (active-day share <15%), occasional (15%–40%), regular (40%–80%) and daily (≥80%). Players are also grouped by average daily online time into light (<3 hours), medium (3–12 hours) and heavy (≥12 hours), matching the tiers on the "Usage" page of the VRCX-0 telemetry dashboard.

Distributions are compared with two-sided Mann–Whitney U tests, with Cliff's δ^[27] as the effect size (|δ|<0.147 negligible, <0.33 small, <0.474 medium^[28]). 95% confidence intervals for median differences come from 4,000 bootstrap resamples^[36], and confidence intervals for proportions use the Wilson interval^[29].

Coverage depends on observed days: JP players were observed for more days and so had more chances to come across any page. Coverage is therefore stratified by page-visit days (3–9, 10–29, ≥30 days) and summarized with the Mantel–Haenszel pooled odds ratio^[30]. Re-check intensity uses Poisson regression controlling for the log of same-day online hours, with standard errors clustered by player. Each family of tests (pages, tools) uses Benjamini–Hochberg control of the false discovery rate^[31] and reports q values. q<0.05 counts as confirmed, and 0.05–0.20 as an exploratory signal.

The group difference in daily players is also tested with logistic regression controlling for friend-count tier and month of first appearance. The feature profiles in Section 4.6 use φ coefficients for co-occurrence of play motivations; group differences stratified by page-visit days and weighted by number of users (300-resample bootstrap intervals); partial Spearman correlations controlling for page-visit days; linear regression with friend tier and activity (HC3 robust standard errors); and Wilcoxon signed-rank tests on two periods of the same player. Play archetypes come from k-means clustering on Hellinger-transformed page composition^[32], evaluated with the silhouette coefficient^[33] and cluster Jaccard stability over 60 bootstrap resamples^[34].

Both groups peak at 22:00–23:00 local time. Across all active days (Figure 1, top), the JP curve is higher and flatter overall: 53.0% of JP's active days are always-on days with ≥20 hours online, versus 28.3% for CN. The median hours online per day are 21 and 13 respectively.

Once always-on days are removed (Figure 1, bottom), differences in daily routine show up. CN's trough is at 07:00–08:00 (16%); the curve starts rising at 09:00 and climbs steadily through the afternoon and evening. JP's trough comes later, at 10:00–11:00 (24%), and 31%–38% of JP players are still online at 03:00–05:00, compared with 20%–29% of CN. Computed over each player's non-always-on days, the median share of online hours falling late at night (hours 0–5) is 0.31 for JP and 0.25 for CN (δ = −0.20, p = 1.6×10⁻⁴). In the daytime (hours 9–17) it is 0.31 for CN and 0.26 for JP (δ = 0.20), and in the evening (hours 18–23) 0.36 for CN and 0.32 for JP (δ = 0.21).

Session start and end times (Figure 2) are more direct traces of what players do by hand. Both groups' session starts peak at 20:00–21:00. CN has a secondary peak at 12:00–13:00 (6.2% and 5.8%, versus 3.6% and 4.5% for JP), while JP is higher at 08:00 (4.1% vs 2.4%). Session ends differ more: 34.0% of CN session ends fall between 22:00 and 01:00, versus 21.1% for JP. JP has secondary peaks at 03:00 (7.1%) and 07:00 (5.7%), where CN has only 4.6% and 2.1%.

The 07:00 peak in session ends, together with JP staying online all night on more than half of active days, fits sleeping in VR, which is common in the Japanese community: players fall asleep wearing a headset or with the game running, then shut down after getting up and heading out. Telemetry cannot tell "sleeping in VR" apart from "a PC left on all night", so this is consistent evidence, not a direct measurement. Days that are "online late at night, then off for at least 3 hours in the daytime" make up about 7%–11% in both groups (p = 0.42). This shows that JP's all-nighters mostly happen on always-on days when the client is never closed.

No weekend effect. For players who are not daily players, the ratio of weekend to weekday logon probability has a median of 1.00 for CN and 0.94 for JP (p = 0.30). In both groups each day of the week accounts for 13.7%–15.3% of active players, which is nearly flat. For these players, VRChat is an everyday evening routine, not weekend entertainment.

CN logs on often and briefly; JP logs on less often but stays a long time (Table 3). CN has a median of 3.84 online sessions per week and JP 1.97 (δ = 0.22). The per-player median session length is 5.4 hours for CN and 12.8 hours for JP (δ = −0.21). 26.8% of JP online sessions run for more than 3 days straight, versus 12.6% for CN, while CN sessions cluster at 1–12 hours (37.5%, JP 21.3%) (Figure 3). In both groups about a quarter of sessions last under 15 minutes, mostly a quick look before closing.

The return gap (from the end of one session to the next logon) is almost the same in both groups: a per-player median of 12.7 hours versus 11.3 hours (p = 0.79). The only difference is that CN more often comes back after one night (12–24 hours, 25.4% vs 19.1%), while JP more often comes back after 2–7 days (16.4% vs 10.7%), which reflects rest after long always-on stretches (Figure 4). Counted in days, 90.3% (CN) and 91.4% (JP) of returns happen the next day. The long tail of gaps comes mainly from daily and weekly rhythms and does not need to be explained as "bursty" use^[35].

Daily players make up 46.0% of JP and 31.8% of CN. Trial players who came on only 1–2 days make up 24.5% of CN and 14.3% of JP (χ² test p = 7.8×10⁻⁵, Figure 5). By average daily online time, 49.2% of JP are heavy (≥12 hours), versus 25.6% of CN. Crossing the two segmentations, 84.5% of JP's daily players are also heavy, compared with only 52.8% of CN's, and another 45.5% of CN's daily players are medium. For CN, "playing every day" mostly means a few hours each evening. For JP, it mostly means never closing the client.

Do the differences come from social circle size or from when players joined? JP players have more friends (Section 4.6) and joined earlier (89% of CN are players who first appeared this quarter, versus 79% of JP). After controlling for friend-count tier and month of first appearance, the odds ratio for a JP player being a daily player is 2.06 (95% CI 1.29–3.29, p = 0.002), and the mean difference in active-day share shrinks to 8.5 percentage points (4.7–12.3). Part of the difference comes from how far along players are, but a gap in long-term stability remains. Among players with fewer than 100 friends, the daily share is 55% for CN and 83% for JP; among those with 100–500 friends it is 64% vs 71%.

The two groups use almost the same features. After stratifying by observed days, none of the 17 pages differs significantly in coverage (Mantel–Haenszel q ≥0.42 for all). Over 90% of players used Friend Locations, Game Log and Friend History, and about 80% used Current Players, Instance History, Search, My Avatars and Mutual Friend Graph. On each page-visit day players open 3.1 distinct pages on average, the same in both groups.

The difference is in how they split their attention. Figure 7 shows each page's average share of a player's total visits. JP spends more of its visits on Game Log (16.7% vs 13.2%, q = 0.004), Favorite Friends (3.7% vs 2.7%, q = 0.016), Notifications (q = 0.032) and Dashboard (q = 0.034). CN spends more on My Avatars (6.6% vs 3.9%, q = 0.050), Search (q = 0.068), Mutual Friend Graph (q = 0.096), and Favorite Worlds and Favorite Avatars (q = 0.16–0.24).

JP keeps checking on "people" through the day. How many times a page is opened in a day depends on that day's online time, and JP players are online longer, so they naturally look more often. After controlling for same-day online time with Poisson regression (Figure 8), JP still opens Friend Locations 28% more often than CN (IRR 1.28, 95% CI 1.10–1.50, q = 0.017), Instance History 25% more, Favorite Friends 27% more, the Friends page 25% more, Game Log 22% more and Friend History 20% more (all q <0.05). My Avatars (1.08), Search (1.12), Favorite Worlds (0.93) and Current Players (1.12) show no difference.

These are two different ways of paying attention. Over long hours online, JP keeps track of where friends are, who they spent the day with, and who added or removed them. CN checks the "people" pages once or twice a day and puts more attention on their own look and the worlds they can visit.

Play archetypes. Clustering by page composition gives 5 archetypes (Figure 9): Friend tracker (Friend Locations 45%), Log reviewer (Game Log 34%), Friend watcher (Friend History 38%), Curator (Favorite Worlds 14%, Favorite Friends 9%) and Generalist (no dominant page). Log reviewers make up 28.3% of JP (CN 17.9%) and Friend trackers 13.7% (CN 9.7%). Curators make up 18.5% of CN (JP 14.1%) and Friend watchers 16.9% (JP 11.7%). The difference in distribution has p = 0.015. The silhouette coefficient is only 0.11, and cluster stability (Jaccard) is 0.35–0.59. Player preferences are continuous, so the archetypes are only representative points on a spectrum and should not be read as distinct types.

Tool use is low. In both groups only about half of players opened any tool in 7 weeks, and only 21.8% of "player × tool" pairs appear on two or more days. Samples for individual tools are small, so everything below is an exploratory signal (Figure 10).

JP leans toward running communities and creation. JP uses Group Moderation (12.2% vs 4.3%, OR 0.39, q = 0.06), Export My Avatars (8.2% vs 2.0%, q = 0.06), App Launcher (18.9% vs 9.6%, q = 0.06), Discord Names (10.2% vs 4.3%), Export Friends List, the Calendar (groups) tool (5.6% vs 2.0%) and Screenshot (17.3% vs 10.3%) more than CN. Group Moderation lists only groups where the player has moderation rights, so the people using it are essentially group organizers. This fits Japan's meetup culture, where regular events require managing members, publishing schedules and coordinating on Discord. Export My Avatars points to players who have uploaded several avatars of their own. App Launcher is mostly used to start and stop face tracking, OSC tools and desktop overlays together with VRChat. JP's heavier use of it also fits their long online hours and extra gear, but no outside data backs this up.

CN leans toward local maintenance. CN uses VRChat Registry Backup (5.3% vs 1.0%, OR 3.79, q = 0.09), VRChat Data (12.9% vs 6.1%, q = 0.07), AI Connections (14.2% vs 9.7%) and VRChat Config (21.2% vs 16.8%) more. One possibility is that CN players more often deal with local problems around cache, graphics quality and networking. AI Connections is mainly used for social AI and translation, which may relate to CN players more often being in foreign-language instances. Both are guesses.

The sections above compared the two groups feature by feature. This section takes another angle and asks what kind of player each feature represents. Pages and tools are grouped into 13 play motivations (Table 1) and examined from five directions: which motivations appear together, what their users have in common, how friend count changes what players focus on, how newcomers change as they become veterans, and what gets used at which time of day. This section uses data from August 14 to September 30, covering 302 CN and 196 JP players. Everything below describes correlations. It is meant to suggest profiles, not causes.

Coding whether each player used a motivation as 0/1 and computing pairwise φ correlations (Figure 11) reveals two clusters. The first is in-game social: Finding friends, In-instance social and Looking back correlate most strongly (φ 0.39–0.50), and Exploring worlds and Avatars & creation also attach to this cluster (φ 0.25–0.43). The second is out-of-game roles: Running communities, Status automation, Tinkering, Photography and AI assistance correlate with each other (φ 0.25–0.47) and only weakly with the first cluster (φ mostly below 0.2).

The players who organize events, adjust settings and sort photos tend to be the same deeply engaged players. This kind of engagement and "finding and watching people in the game" are two independent axes. Some players chase their friends every day but never touch the tools, while others rarely look at Friend Locations but run groups. The tools in the second cluster are all reached from the same Tools page, so whether a player opens the Tools page at all also makes them correlate, and this part should be discounted.

In usage rates across the 13 motivations (Figure 12), the groups differ significantly only on Running communities: JP 24.5%, CN 11.9% (stratified OR 0.49, q = 0.04). Exploring worlds (OR 1.49) and AI assistance (OR 1.64) lean toward CN, but not significantly.

Active players use everything more, so comparing users with non-users directly mixes in activity level. Table 4 first stratifies by page-visit days and compares only players with similar activity. The differences whose intervals exclude 0 are:

• Users of Running communities are 17 percentage points more likely to be JP, spend 2.2 more hours online per day and have used 7 more features. They are heavy JP players who stay online for long hours and use everything.
• Users of Status automation have a 9.5 percentage point higher share of always-on days and slightly more late-night time online. Only players whose client stays open all the time need to switch status automatically by schedule or by instance.
• Users of Exploring worlds are 14.5 percentage points less likely to be daily players and spend 3.9 percentage points more of their online time late at night. They do not come every day, and when they do, they often browse worlds at night.
• Users of Avatars & creation spend 4.3 percentage points more of their online time in the daytime and are 16 percentage points more likely to have fewer than 100 friends (interval includes 0). They are players whose social circle is still small and who also work on their look during the day.
• Users of AI assistance are 12 percentage points less likely to be JP, meaning they lean CN.

Pages show matching patterns. Partial rank correlations between visit share and player traits (controlling for page-visit days, BH-corrected) with q<0.05 are as follows. Game Log correlates positively with the share of always-on days (ρ = 0.18), late-night online time (0.15) and being JP (0.13). Friend History (0.17) and Favorite Friends (0.14) correlate positively with friend count. Instance History (−0.19), Mutual Friend Graph (−0.14), Moderation (−0.14) and My Avatars (−0.13) correlate negatively with friend count. Search correlates negatively with being JP (−0.13).

The clearest picture comes from the Game Log: the more often players stay logged in all day and online late at night, the more they read the Game Log. It records the players who came and went in an instance and the worlds visited. For someone who idles or sleeps in VR, reading the log after waking up shows "who came by while I was away, and where we went".

Among players who reported a friend count (CN 386, JP 197), 15.7% of JP have 500 or more friends, versus only 5.4% of CN (p = 3.7×10⁻⁵). 44.3% of CN have fewer than 100 friends, versus 32.0% of JP (Figure 13). 12.9% of JP players and 2.6% of CN players have more than 1 million Feed entries stored locally.

Players with different friend counts look at different things (Table 5). With fewer than 100 friends, Friend Locations accounts for 16.0% of visits, My Avatars 8.6% and Instance History 7.9%. With more than 500 friends, Friend Locations drops to 10.7%, while Friend History rises to 20.1%, Game Log to 21.0% and Favorite Friends to 7.1%. Players with small circles are "finding people": they check who is online and where, recall where they met whom, and work on their own look. Players with large circles are "managing people": a location list of several hundred friends is too long to follow, so they use Favorite Friends to pin down close friends and Friend History to track adds, removals and name changes.

Does this explain the group differences? Partly. After controlling for friend count and activity, JP's lead on Game Log (+1.7 percentage points) and Favorite Friends (+1.1) shrinks and is no longer significant, which suggests JP's "social bookkeeping" is mainly due to larger friend circles. CN's lean toward exploration and avatars remains, however: Search −1.9, Favorite Worlds −2.1, My Avatars −1.8 and Favorite Avatars −0.6 percentage points (all q <0.05). This looks more like a difference in taste between the two groups, consistent with the author's observation that Simplified Chinese players mostly focus on switching avatars and browsing worlds.

This quarter, 193 new players (CN 106, JP 87) had enough page visits both in week 1 after joining and from week 4 on. Comparing the same players' visit composition across the two periods (Figure 14, Wilcoxon signed-rank test), the largest drops are in Mutual Friend Graph (6.0% → 3.1%), Dashboard (4.7% → 2.6%), Current Players (9.3% → 7.2%), My Avatars (6.4% → 4.9%) and Instance History (7.2% → 5.8%). The rises are in Friend History (12.6% → 18.5%), Game Log (15.4% → 18.9%) and Notifications (1.1% → 2.5%). Both groups change in the same direction.

Newcomers are getting to know their surroundings: who is in the instance, who they share friends with, how to set up their own avatars. Once settled, their attention turns to everyday relationship upkeep: who added me, whom I met yesterday, who sent me an invite. Part of the early period is also spent learning the tool itself, such as trying out a Dashboard, so not all of the change is a change in how they play.

Page records carry the time of the last visit, which gives a rough view of when each feature is used. The differences are small and mostly follow the online rhythm. The highest late-night (hours 0–5) shares belong to Friend Locations (29.4%) and Friend History (28.7%). The lowest belong to Dashboard (24.4%), Moderation (24.9%) and Favorite Friends (25.1%), which are more concentrated in the evening, hours 18–23. Late at night, players check who is still awake. "Chores" such as organizing favorites and handling blocks tend to happen at a set time in the evening.

The JP data shows a way of life that amounts to "living in VRChat". On more than half of active days the client stays open all day, single online sessions often span several days, and session ends show secondary peaks at 03:00 and 07:00. This matches the Japanese survey finding that about three in ten respondents have slept in VR^[11], and the author's observation that sleeping in VR is common.

During these long stretches of presence, JP keeps paying attention to social ties. Through the day they check more often where friends are, look back at whom they met today and organize Favorite Friends. The longer they stay logged in, the more they read the Game Log, and players whose client is always on use Status automation more. These habits go hand in hand with sleeping in VR: status switches to Do Not Disturb automatically before sleep, and after waking they read the log to see who came by. JP players have larger friend circles and more Feed activity, the result of tending their social circles over a long time, which also explains why they need to check more often. Group Moderation, Calendar (groups) and Discord Names are more common, which points to the role of meetup organizer^[14]. Export My Avatars is more common too, which points to creators who upload avatars they made or modified.

CN's rhythm is closer to ordinary evening entertainment. They are online in the daytime and early evening, shut down to sleep between 22:00 and 01:00, and log on several times a week for a few hours each time. Their attention goes more to their own look and to places they can go: My Avatars, Search, Favorite Worlds, Favorite Avatars and Mutual Friend Graph. This fits the author's observation that they mostly focus on switching avatars and browsing worlds. A high share have fewer than 100 friends. Most CN players are still building their social circles, so they have less need to track "people".

Friend count explains part of the difference. Players with few friends look more at Friend Locations, Instance History and their own avatars anyway, and CN has more such players, so CN as a whole looks more like the "newcomer pattern". But after controlling for friend count, CN's preference for Search, Favorite Worlds and avatars remains, which looks more like a difference in taste. At similar friend counts, CN is also less stable than JP (among players with fewer than 100 friends, the daily share is 55% vs 83%). This may relate to schedules set by school or work, or to network availability that varies by time of day. The current data cannot tell these apart.

Both groups play at night, peaking at 22:00–23:00. Both return on a next-day rhythm, neither shows a weekend effect, and their page coverage is almost identical. For both, VRChat is an everyday social space. They differ in how long and how steadily they engage, and in how they split their attention between "people" and "things".

Of three external data points, two agree with this report and one marks the limits of how it can be read. First, the Japanese survey found that 48%–52% use social VR "almost every day"^[10], and 46.0% of JP players in this report are daily players (95% CI 40.0%–52.2%), so the two agree. Second, the Japanese survey reported a login peak at 22:00 local time^[11], and both groups here peak at 22:00–23:00. Third, the same survey found that only 44% of Japanese respondents used social VR for 3 hours or more per session, less than in North America and Europe^[10], while the median JP online session in this report is 12.8 hours. So JP's long durations are mostly "presence" rather than continuous "play": the client and PC stay on while players go in and out of the game, sleep and go to work. "Living in VRChat" means VRChat is an always-on social backdrop, not a dozen-plus hours of continuous gaming.

1. Online time is not play time. The data records only whether the client is online, not whether VRChat is running. JP's always-on presence may be sleeping in VR, or simply a PC that is never turned off. Launching at startup and staying in the system tray also push up the always-on share. The inference about sleeping in VR rests only on consistent evidence.
2. Sample representativeness. VRCX-0 users are a specific subgroup of VRChat players: PC players (over 90% on Windows), more socially engaged, and willing to install third-party tools. The conclusions do not extend to Quest, mobile or VRChat players as a whole.
3. Grouping by language. CN is based mainly on Simplified Chinese and includes Simplified Chinese players overseas. JP is based on Japanese and misses Japanese players who use an English UI (the wider definition in Appendix B does not change the conclusions).
4. Page visits cover only some interactions. User, world, avatar and group details open as pop-ups and do not count as page visits. The default home page, Feed, is not counted. Notifications show as a drawer by default, so visits to the table page understate attention to notifications. Tools are recorded as opens, not actual use, and automation tools keep running in the background once configured.
5. Short observation period with seasonal effects. The data covers only one quarter, which includes summer vacation and Japan's Obon and Respect for the Aged Day holidays. A surge of new CN players before China's National Day holiday at the end of September (158 new players in the week of September 28) falls outside this report's window.
6. Feature profiles are correlational. The results in Section 4.6 are all observational. They are stratified by or controlled for activity but may still be affected by other factors. Tool motivations are all reached from the same Tools page, so the correlations among them are overstated.
7. Multiple comparisons and effect sizes. Most significant differences have "small" effect sizes (|δ| 0.20–0.27). Tool-level q values are 0.06–0.20 and should be treated as hypotheses to be tested.

Using anonymous VRCX-0 telemetry, this report compared how Simplified Chinese users and Japanese players play VRChat. The two groups use roughly the same features, both play in the evening, and both come back the next day. The difference is that JP treats VRChat as a community they live in for the long term: always-on, up all night, checking on friends again and again, organizing meetups and making their own avatars. CN treats VRChat as a new world to explore in the evening: they log on and off, focus on their look and on worlds, and their social circles are still forming. Viewed by feature, player engagement splits into two axes, "in-game social" and "out-of-game roles". Focus shifts from "finding people" to "managing people" as friend counts grow, and new players mature along the same path.

For companion tools and community organizers, the value for JP lies in social tracking and community management during long hours online. For CN, it lies in better exploration and avatar management, and in helping new players build their social circles. Future work could, with consent, add a signal for whether VRChat is running in order to separate play from idling, and include more language groups for a wider comparison.