I recently fell down the rabbit hole of watching live streams on Bilibili—specifically virtual streamers whose main act is singing, known in the community as *geshi* (singing VTubers). When people hear I've been watching live streams, their first assumption is usually that I'm watching attractive women dance on camera. Anyone familiar with Bilibili knows that most of these streamers never show their faces or bodies at all; they perform behind 2D or 3D cartoon avatars driven by real-time facial capture. Others who know a bit about the VTuber industry assume viewers are there to buy "emotional value." That has never quite worked on me either. For one thing, emotional value is a notoriously nebulous concept. For another, my background in data analytics and a stint doing amateur game design make it almost impossible for me to switch off my analytical brain. When I see a mystery loot box on screen, I reflexively calculate its expected value and check how the payout table was balanced. When a streamer coos sweetly into the microphone, I find myself evaluating her script pacing and conversion rate. Call it an occupational hazard, but I simply cannot sink into the illusion the way a typical viewer might.

Even so, I found an enormous amount of joy in these rooms—for a completely different reason. On Bilibili, every room's audio-video stream, live chat (*danmu*), and gift log are publicly visible in real time, making the platform a remarkably transparent, large-scale data source. Over a few weeks, I used AI to build a suite of monitoring and analysis tools, collected a large dataset across rooms, and lurked in streamer and agency operations groups until I had mapped out how money actually moves through this ecosystem, what niches exist, and how the players interact. Building those tools and satisfying my curiosity turned out to be far more fun than buying emotional comfort.

After about three weeks of immersion, I organized what I learned into a four-part series. This first piece starts from the platform's perspective: why pure, non-risqué live streaming is a commercial dead end on its own, and how Bilibili—and live-streaming platforms more broadly—engineered it into a massive market.

To see the problem, start with live streaming in its most stripped-down form. As my friend Lizheng pointed out in [this video](https://youtu.be/gBj0JY7LDpE), imagine walking past a street musician busking on the sidewalk. If they sing well, passersby drop a tip into the guitar case. Content in exchange for tips is the most natural business model imaginable.

The trouble is that no matter how breathtaking the performance, a street tip tops out at a few dozen or maybe a hundred RMB (around $15). Nobody pulls thousands of dollars out of their pocket and dumps it into a guitar case on the sidewalk. Moving that exact setup online changes nothing. If a live room consists of nothing more than a singer, a microphone, and a donate button, putting a ¥1,000 gift on the menu won't make anyone click it. Ninety-nine percent of passersby will listen for free and walk right out without leaving a dime. Maybe the remaining 1% of kind-hearted listeners will toss in ¥10 or ¥20 to request a song, and a die-hard fan might spend ¥100 to show support—and that is the hard ceiling.

Yet anyone who has glanced at the live-streaming industry knows reality looks nothing like that. Lavish spending is routine; seeing viewers burn through tens of thousands or even hundreds of thousands of RMB in a single night is hardly unusual. What props up revenue on that scale, and what commercial mechanisms make it work? That is what this article is about.

## Why Would Anyone Pay for a Few Glowing Icons?

To understand how Bilibili breaks through that tipping ceiling, we first have to ask why casual viewers refuse to pay in a natural, unengineered stream. Two barriers stand in the way. First, paying online requires scanning a QR code and topping up a wallet—a huge psychological hurdle when going from zero to the very first cent. Second, as long as the product in the viewer's mind is "listening to one song," people will anchor its worth at two or three dollars. Jacking up song prices or having the streamer aggressively beg for gifts only drives newcomers away.

How do you break that stalemate? Bilibili takes a route that squeezes the streamer. It introduced a virtually zero-cost item called the Fan Club Light-Up Badge, priced at ¥0.1 (about 1.5 cents). Sending a single ¥0.1 token grants the viewer a dedicated fan badge for that streamer; if the viewer goes several days without sending a gift, the badge grays out and goes dark until they pay again to relight it.

On its own, a ¥0.1 badge wouldn't do much. The catch is that Bilibili ties the number of daily lit fan badges directly into the streamer's algorithmic traffic KPIs. The more viewers light up badges in a room, the more recommendation traffic the platform feeds the streamer. That pushes newer singing streamers into offering loss-leader promos: spend ¥0.1 to light up a fan badge, and a newcomer gets a free song request plus a personal thank-you by name after the song ends. Meanwhile, badge levels run on "intimacy points," and the very first ¥0.1 badge token sent each day awards 6x points—nudging regulars to swipe one token every day just to keep their streak alive. Without spending a single cent on subsidies, the platform uses one KPI rule and one point multiplier to borrow the streamer's own voice and get new viewers past the friction of their first payment.

Getting a customer to spend their first ten fen solves wallet friction, though—it still doesn't lift total spending past the street-tip ceiling. To push spending into the hundreds and thousands, platforms introduce two level badges that appear right next to a viewer's username whenever they enter a room or send a chat message. On the surface, this looks borderline absurd: digital badges create zero intrinsic value, nobody gets genuine joy just from having two glowing pixel icons beside their handle, and if the streamer ignored them, they would be completely worthless. Just like the ¥0.1 fan token, however, they matter for one simple reason: because the streamer cares, they work.

What actually pushes spending past the ¥100 mark is that these badges quietly swap out the product being sold. Instead of selling a song, the room is now selling its only genuinely scarce resource: the live attention of the human streamer on the other side of the screen. When dozens of chat messages scroll past at once, whose comment gets read, whose request gets answered, and whose name gets called out warmly is a far more valuable commodity than hearing a couple of tunes. A quick glance at the number and color of a badge tells the streamer who the real big spenders are and where to direct her limited energy.

There is a subtle design problem here, too: why maintain two separate level systems instead of one? If badges were tied strictly to a single streamer, then the moment that streamer retired, went on hiatus, or lost a whale's interest, tens of thousands of RMB worth of status would vanish overnight—and a disillusioned whale might quit the platform altogether. Conversely, if there were only a single site-wide level that upgraded no matter whose room you spent in, individual streamers would have far less incentive to cultivate loyal regulars of their own. So Bilibili split status into two parallel ledgers: a room-specific Fan Badge for each streamer, and a site-wide Wealth/Honor Level that accumulates across every room on the platform, unlocking global entry animations and chat styling that signal spending power to any streamer the moment you walk in.

| Level Tier | Single-Streamer Fan Badge Cost (with 1.5x Captain boost) | Site-Wide Honor Level Cumulative Cost |
| --- | --- | --- |
| Level 10 | ¥10.3 | ¥200 |
| Level 20 | ¥108 | ¥1,100 |
| Level 30 | ¥1,600 | ¥8,300 |
| Level 40 | ¥20,000 | ¥64,000 |
| Level 50 | ¥253,600 | ¥330,000 |
| Level 60 | ~¥1.414M | ¥1.50M |

The table above shows how much it costs to reach each tier across both ledgers (with the Fan Badge column calculated assuming the 1.5x speed boost from the monthly Captain subscription covered in the next section). Up through Level 20, the single-streamer badge is just a hundred-yuan warm-up: ¥10.3 for Level 10, ¥108 for Level 20. Past Level 30, both curves go exponential. Hitting Level 60 on a single streamer's badge takes roughly ¥1.414 million (~$195,000), while reaching Honor Level 60 site-wide costs ¥1.50 million—and Level 70 climbs to a staggering ¥4.70 million.

That creates one more edge case: what if a whale spends hundreds of thousands to reach a high Honor Level and then stops paying entirely, coasting on flashy entry effects in every room for free? Bilibili plugged that hole in its [official Honor Level rules](https://www.bilibili.com/read/cv28896490): go 7 consecutive days without spending money in live rooms, and your Honor badge and privileges gray out and lock until you send another paid gift to relight them.

## Turning Casual Tips into Monthly Rent

Even with dual badges and a 7-day decay timer, a loophole remains. A whale only needs to toss in a few cents once a week to keep the badge lit, while ordinary viewers might drop ¥50 when they're in a good mood today and disappear next month when work gets busy. That revenue volatility is a headache for any platform. The natural question is: why not turn live-stream patronage into rain-or-shine recurring revenue, like a Netflix subscription? That is exactly why Bilibili launched its [Grand Voyage membership system](https://link.bilibili.com/p/live-h5-recharge/guard.html), a monthly subscription with three tiers: Captain at ¥198/month (~$27), Admiral at ¥1,998/month, and Governor at ¥19,998/month.

Much like the badges above, the official perks Bilibili hands out to Captains are remarkably underwhelming: a special icon, an entry banner, colored chat messages, and a 1.5x boost to Fan Badge progression—every single one of them a zero-marginal-cost digital symbol. Once again, all the perks that actually convince viewers to open their wallets are paid for out of the streamer's own time and pocket.

Why do streamers willingly stuff real perks into these memberships? Because Bilibili mounts the room's active subscriber count right at the top of the screen as a public scoreboard, tying it both to social prestige and to recommendation traffic. If that number looks anemic, the streamer loses face and loses algorithmic reach. To defend the number floating above her head, she has little choice but to invest her own time, energy, and cash to make the membership worthwhile. Common perks include free priority song requests for Captains and access to a private WeChat or QQ group for everyday chat. Many streamers also spend ¥20 to ¥30 or more out of pocket per subscriber to ship custom physical merch and handwritten letters with free postage.

Once again, without spending a dime on fulfillment, the platform uses a single public counter to force streamers to pack real human companionship and physical gifts into a monthly subscription. Once a streamer turns her Captains into a tight-knit circle of a dozen or twenty regulars, canceling stops being a private click. Nobody notices when you cancel a video streaming plan, but in a small or mid-sized singing room where the streamer knows every Captain by name, letting your subscription lapse makes the public counter above her head drop by 1 in front of everyone, and your name vanishes from the roster. Within a small social circle, that awkwardness is palpable.

On top of social pressure, the platform layers in auto-renewal price locks. Buying a single month of Captain costs ¥198, while enabling auto-renewal drops the price to ¥168/month. Long-time subscribers who locked in an early promotional rate and never let it lapse still pay a grandfathered ¥138/month. Miss a single billing cycle, and not only does your badge go dark—giving the streamer every incentive to check in on you personally—your permanent ¥138 grandfathered rate is gone for good.

## The Probability Trap Inside a ¥15 Blind Box

Beyond levels and subscriptions, Bilibili has another clever piece of design up its sleeve. Suppose daily song requests had a flat, transparent price: send a fixed ¥15 gift to hear one song. A clear price tag gives the viewer's rational brain a clean stopping point—you pay ¥15, hear your song, and instinctively hit the brakes on your budget. To push volume higher and nudge viewers toward irrational spending, the platform introduced the ¥15 Heartbeat Blind Box, and singing streamers almost universally adopted it as the required currency for requesting a song. Whatever gift pops out of the ¥15 box goes straight to the streamer, and everything on the viewer's side—Fan Badge intimacy, Honor Level progress, room leaderboard rank, and visual effects—is credited according to the sticker value of the unboxed item. Pull a Romantic Castle worth ¥2,233 from a ¥15 box, and the system treats you as having dropped a ¥2,233 gift; pull a ¥2 Movie Ticket, and you get ¥2 worth of credit.

Most people's gut reaction when they hear this is to suspect Bilibili rigs the RNG behind the scenes. In practice, it doesn't have to—and legally it can't. Under China's [blind-box regulatory guidelines](https://policy.mofcom.gov.cn/claw/policyInfo.shtml?id=5424) issued by the State Administration for Market Regulation in June 2023, item values, rules, and exact probabilities must be prominently disclosed, with algorithmic manipulation strictly prohibited. As we will explore in more detail later in this series, we sampled nearly 4,000 real Heartbeat Blind Box pulls across 20 rooms and observed the following distribution:

| Unboxed Gift | Sticker Value | Observed Count | Observed Share |
| --- | --- | --- | --- |
| Heart Pillow | ¥16 | 1,714 | 45.51% |
| Cotton Candy | ¥9 | 1,665 | 44.21% |
| Movie Ticket | ¥2 | 361 | 9.59% |
| Jeweled Scepter | ¥40 | 20 | 0.53% |
| Jade Seal | ¥200 | 1 | 0.03% |
| Romantic Castle | ¥2,233 | 5 | 0.133% |

Across all 3,766 observed pulls, the average unboxed sticker value came out to ¥14.68—remarkably close to both the ¥15 ticket price and Bilibili's published expected value. At first glance, that looks almost identical to sending a fixed ¥15 gift. The catch sits in the bottom row. Although the ¥16 Heart Pillow shows up over 45% of the time, it nets just ¥1 per pull—nowhere near enough to dig out of the hole left by ¥9 Cotton Candy and ¥2 Movie Tickets. Once you pull more than a few boxes, breaking even or coming out ahead depends entirely on hitting the ¥2,233 Romantic Castle. Yet the Castle is extraordinarily rare, showing up roughly once every 750 pulls—just 5 times in all 3,766 draws. Strip out those 5 Castles, and the average value of the remaining pulls plunges to ¥11.73.

Even during Bilibili's Friday 5x jackpot boost events—when the platform raises Castle odds enough to push the single-pull expected value up to ¥18.69—the lived outcome barely budges. With an infinite bankroll, someone who kept pulling forever until Castles appeared really would come out ahead in the long run. Real viewers, however, operate under finite budgets and rarely pull anywhere near enough times for the law of large numbers to bail them out. Even under that positive-EV ¥18.69 Friday boost, 76.9% of players who spend ¥1,500 on 100 consecutive pulls still walk away net losers. In plain terms, the vast majority of budget-constrained viewers steadily lose money, subsidizing with their own wallets and luck the tiny handful of jackpot winners who get the rush of blasting a ¥2,233 Castle across the screen. Every Castle pull also triggers a site-wide broadcast banner, drawing curious onlookers into the room and tempting even more people to try their luck.

Compounding the math is the sunk-cost trap. Even viewers who know enough probability to realize that each pull is independent—that losing 100 times in a row does nothing to improve the odds on pull 101—still feel a visceral sting after unboxing a pile of Cotton Candy and Movie Tickets. Stopping feels like admitting all the money they already put in went down the drain. Trying to claw their way back to even, they grit their teeth and keep clicking, one box after another.

From the streamer's perspective, encouraging viewers to pull blind boxes is a clear win as well. Unlike any individual viewer, a streamer faces dozens of people pulling across sessions—100 pulls from one viewer, 50 from another—so the law of large numbers actually works in her favor, letting her reliably capture the mathematical expectation. Meanwhile, the thrill of gambling, the lure of the Castle jackpot, and irrational sunk-cost chasing all push viewers to buy far more boxes than they would fixed-price songs, lifting overall room revenue—plus a bonus burst of traffic whenever a Romantic Castle lights up the site-wide banner.

## Giving Lurkers an Excuse to Spend

Stepping back to look at the first three mechanisms together reveals a clear progression. The first two designs—the fan badge and the Grand Voyage subscription—are fundamentally ways for the platform to extract free labor from streamers, using traffic KPIs and a public scoreboard to make streamers pour their own attention, time, and physical merch into otherwise worthless digital icons. Starting with the third design, the Heartbeat Blind Box, the platform pivots toward manipulating viewer psychology directly, using rare jackpots and sunk costs to tap into individual gambling impulses. Yet individual gambling impulses only kick in once a viewer already feels like requesting a song or opening a box. In an actual singing stream, a session often runs for several hours of alternating songs and relaxed chat. For monetization, that calm atmosphere is a serious obstacle: even when viewers genuinely like the streamer and have balance sitting in their wallets, without an external spark they tend to leave the stream running in the background as silent lurkers, telling themselves they'll tip next time. Singing streamers, meanwhile, are often too bashful to hold out their hand for money after every single song. What the platform needs is a way to stir up collective emotion and manufacture frequent, socially natural occasions to spend. The textbook implementation is [PK Chaos Battles](https://live.bilibili.com/activity/live-activity-full/chaos/pc.html).

In a PK battle, two singing streamers connect on split-screen, trade half-songs to showcase their voices, and cross-pollinate audiences. Above the split-screen feed, the platform draws a zero-cost red-versus-blue live score bar, pitting the two audiences against each other through gift-giving. A standard round lasts 5 minutes, and between minutes 3 and 4 the system activates a 1-minute double-points window where every ¥1 spent counts for 2x PK score on the bar.

Winning a PK round carries virtually no cash prize—it is little more than a number ticking on a screen. What makes that red-and-blue bar so potent is how instantly it turns passive listening into a team sport. Seeing their streamer trail behind triggers viewers' protective rivalry—nobody wants to watch their favorite lose in front of another room—and the 1-minute double-points window adds the irresistible bargain logic that spending one yuan right now hits like two. Just as importantly, it hands shy streamers a graceful excuse to ask for support. Instead of awkwardly begging for tips, all a streamer has to say into the mic is, "Double points are up—help me put some points on the board and hold the lead!"

Out of the 20 singing rooms we monitored, 19 ran PK battles. Across those rooms, active PK voting accounted for just 8.9% of total broadcast hours, yet generated 67.1% of all gift revenue—an hourly monetization rate 20.9 times higher than normal streaming. Better still, over 60% of all PK spending erupted inside that single 60-second double-points window between minutes 3 and 4. With nothing more than a zero-cost red-and-blue bar at the top of the screen, the platform effortlessly shakes loose money that would otherwise stay parked in viewers' wallets.

## Closing Thoughts

Coming back to where we started, it becomes clear why transplanting a street busker straight onto the internet is a commercial dead end—and how platforms like Bilibili turned it into an enormous industry anyway. Online, singing itself isn't scarce. Hundreds of people can listen in the same room at zero marginal cost to either the platform or the performer; when everyone can listen for free, nobody is willing to pay high prices for the music alone, and pure content hits a hard ceiling around a hundred yuan. What the platform actually monetizes is the web of code-driven rules wrapped around the stream. Bilibili never sings a note and never ships a physical product. It writes a few lines of code to mint badges, mount a public counter, publish a loot table, and draw a progress bar—squeezing streamers into supplying real time, energy, and merch on one end, while prodding viewers' vanity, loss aversion, and team rivalry on the other until spending feels almost automatic.

Once you see that, Bilibili's 50% revenue cut on standard gifts makes a lot more commercial sense. To an outsider, taking half the pot while producing none of the content looks predatory. Flip the perspective around, though: without the artificial table the platform built—without Bilibili playing bad cop behind the scenes for streamers and manufacturing spending triggers for viewers—almost none of that massive revenue stream would exist in a natural state.

Up to this point, however, we have looked at the system strictly through the platform's eyes, treating everyone inside as rational—or predictably irrational—pieces moving according to the rules. Code may be cold, but the people sitting at both ends of the table are very much human. The whales dropping tens or hundreds of thousands of RMB a night are usually sharp, street-smart people in real life; can they really not see through a few glowing pixels, blind-box odds, and red-and-blue bars? If they see right through the game, what are they actually buying, and why do they stay at the table? And on the other side of the screen, for the streamers pouring their time and emotions into defending the public counter above their heads every day, does the math actually pencil out—and what toll does it take? We will explore those questions in the next piece.
