Silence the Stars, and the Middle Class Disappears First

Based on the research ofNingzhe Zhou, Yilin Li, and Chong (Alex) Wang, "When Stars Are Silenced: The Rise of the Long Tail and the Crumbling Middle in Creator Ecosystems," International Conference on Information Systems (ICIS), 2025

Every platform trust-and-safety team has, at some point, modeled the same hopeful chart: remove or throttle a handful of dominant creators, and the audience hours those creators used to absorb should redistribute across everyone else, with the biggest lift going to the "almost famous", creators who had the skill and audience base to break out but never got the algorithmic push because a handful of superstars ate the recommendation feed. It is the platform equivalent of breaking up a monopoly and expecting a thriving competitive middle market to emerge. A new paper by Ningzhe Zhou, Yilin Li, and Chong (Alex) Wang, presented at ICIS 2025, tested this expectation against real data from a shock that silenced top creators on a major platform. The distributional shape it found was not a rising tide. It was a bathtub curve with the plug pulled at the top: the long tail surged, the top mostly held its ground, and the middle, precisely the segment platforms most need for succession, crumbled.

What "Crumbling" Actually Looks Like

The mechanism Zhou, Li, and Wang trace is not complicated to state, which is what makes it dangerous to ignore. Star creators get silenced or constrained by some shock, a moderation action, a policy change, an algorithmic demotion, a suspension. The audience attention those stars used to command does not vanish; platforms are attention-conserving systems, and someone inherits the freed-up eyeballs. The paper's contribution is showing where that inheritance actually lands using creator-level data and a distributional difference-in-differences design that separately tracks head, middle, and tail segments through the shock. The tail, the smallest, least-established creators, picks up meaningfully more engagement. The head barely moves. And the middle, the tier just below stardom, loses ground it does not recover.

This matters because "the middle" is not a residual category. It is the developmental pipeline of any creator ecosystem: the accounts with real production quality, a genuine audience base, and enough traction to be plausible successors to the stars above them. When recommendation and discovery algorithms reallocate freed attention toward novelty and freshness rather than toward proven-but-not-yet-huge quality, the tail benefits from the same "anyone could blow up" exposure logic that made platforms exciting in the first place, while the middle, visible enough to no longer be a novelty, small enough to still be starved of consistent algorithmic push, gets squeezed from both directions at once.

Why Attention Skips the Middle Instead of Landing There

The intuitive story of redistribution assumes something like water finding a common level once a wall is removed. Creator platforms do not work that way, because the thing doing the redistributing is a recommendation algorithm optimized for engagement and discovery, not fairness or succession planning. Two forces plausibly explain the skip-the-middle pattern the paper documents.

First, discovery algorithms are tuned to surface what is new and different when top content disappears from a feed slot, and "new and different" describes tail creators far better than it describes mid-tier ones, whose content the algorithm and the audience have already partially seen and priced in. Second, and more structurally, the middle tier's core asset before the shock was often adjacency to the stars, collaborations, shared audiences, being the "similar creators you might also like" recommendation off a star's profile. When the star is silenced, that adjacency disappears along with them; the middle loses a channel the tail never had access to in the first place, so the shock is not neutral for the middle even before any new attention gets reallocated. The middle tier was leaning on the very thing that just got removed.

Removing the biggest creators doesn't level the playing field. It reroutes attention around the level below them.

This is not a phenomenon unique to one platform or one kind of shock. The broader information-systems literature on long-tail and superstar effects keeps finding that changes to platform mechanics reshape distribution in ways that surprise the people who designed the change. Fleder and Hosanagar's 2009 Management Science paper on recommender systems found that some popular recommendation designs concentrate sales toward already-popular items rather than diversify them, a "rich-get-richer" pattern running the opposite direction of the ICIS finding but sharing its core lesson: redistribution mechanisms are never neutral. More directly analogous is Geva, Barzilay, Goldstein, and Oestreicher-Singer's 2024 MIS Quarterly paper on Kickstarter, which used a natural experiment, Kickstarter opening its platform to more participants, to show that broadening access shifted demand toward the head of the distribution rather than flattening it. Different shock, different platform, same warning: opening or closing access at one end of a creator distribution rarely produces the evenly-distributed outcome intuition predicts.

The Governance Trap: Fairness Framing Meets Concentration Reality

The political and regulatory case for demoting or removing dominant creators is almost always framed in the language of fairness and diversity: break the stranglehold of a few mega-accounts, and the ecosystem gets healthier, more competitive, more representative. That framing is not wrong about the tail, the paper's finding that the smallest creators gain ground is real. The problem is that "healthier ecosystem" implicitly means more than tail growth. A creator economy also needs a functioning middle class: the tier that provides realistic aspiration for tail creators, absorbs the risk of any single star's departure, and eventually replenishes the head when a star ages out, burns out, or gets silenced again. Real-world reporting on creator-economy income already shows a parallel concentration story building independent of any single platform's moderation choices, one 2026 analysis of ad-payment data found the top 10% of creators captured 62% of ad payments in 2025, up from 53% in 2023, with the top 1% alone rising from 15% to 21%, even as median creator earnings fell. Digiday's reporting on brand budgets in mid-2025 captured the same squeeze directly from creators and talent managers: "There's almost no middle class anymore," one talent agency owner said, describing brand dollars flowing to micro-influencers doing volume deals or the handful of top-tier names commanding premium rates, with almost nothing in between. A platform-level shock that silences stars and crumbles the middle does not counteract that trend; it accelerates it, on top of a market already hollowing out its center.

Real-world enforcement actions offer a cautionary parallel, even outside the paper's exact mechanism. Twitch's 2026 crackdown on viewbotting, announced by CEO Dan Clancy, imposed concurrent-viewership caps on streamers found to be inflating their numbers, a fraud-prevention measure, not a diversity intervention, but the same structural shape: cap the visible reach of accounts near the top, and the platform must reckon with where the resulting attention actually flows, rather than assuming it lands wherever fairness would prefer. YouTube's 2017 "Adpocalypse," triggered by advertiser boycotts over brand-unsafe content, produced a comparable pattern: mid-tier commentary and drama channels, caught by broadened content-safety guidelines that raised monetization thresholds, saw revenue collapse and pivoted to Patreon and direct sponsorships to survive, while the largest channels weathered the disruption far better. Neither case cleanly replicates the ICIS paper's design, but each shows the same blind spot: interventions aimed at the top of a distribution get evaluated on what they do to the top, when the more consequential effect often lands one or two tiers down.

What "Success" Should Actually Be Measured Against

The paper's real contribution to platform governance is methodological as much as empirical: it insists on measuring distributional effects by tier, not in aggregate. A platform press release celebrating "increased engagement among smaller creators" after a star-silencing event is not lying, the tail genuinely did grow, but it is answering a question nobody asked. The question that matters for ecosystem health is what happened to the creators who were one good quarter away from becoming the platform's next stars, and on the evidence here, they got quieter, not louder.

This has a direct implication for how platforms should design and defend moderation, demonetization, or deranking decisions aimed at top creators. If the stated goal is diversity or fairness, tail growth alone is necessary but not sufficient evidence of success; a platform needs to track the middle tier specifically, because that is where the intervention's costs concentrate even as its benefits get reported elsewhere. And if a platform genuinely wants succession, a pipeline that produces its next generation of stars rather than a permanently bifurcated ecosystem of a few giants and a mass of small accounts, silencing the current stars and hoping the algorithm sorts out the rest is not a strategy. It is closer to demolishing a building's middle floors and calling it renovation because the ground floor got more foot traffic.