Punish Your Best Users, Not Your Worst: The Enforcement Rule Most Platforms Get Backwards

Based on the research ofXiaohui Zhang, Zaiyan Wei, Qianzhou Du, and Zhongju Zhang, "Social Media Moderation and Content Generation: Evidence from User Bans," MIS Quarterly 50(1), 2026

Studying user bans on a large question-and-answer community, Xiaohui Zhang, Zaiyan Wei, Qianzhou Du, and Zhongju Zhang found that a temporary suspension does not reliably calm bad behavior. After the ban lifts, sanctioned users post measurably more answers, and those answers are lower in quality. The authors trace this to psychological reactance: strip away someone's freedom and they rush to reclaim it, flooding the platform with quick, emotional, low-effort content. But the effect flips for users who hold earned platform recognition. Because status is an asset they can lose, the same ban threatens something they value, and they respond by posting less and raising quality. Status, not the sanction, decides whether enforcement helps or hurts. That single contingency should reshape how every platform designs penalties.

Ride-Hailing: Uber Learns That a Deactivation Threat Is Only as Strong as What the Driver Has to Lose

Uber runs a four-tier status ladder in Uber Pro, from Blue up through Gold, Platinum, and Diamond, with standing earned over a rolling three-month window from trip points and service quality. Diamond drivers earn 5% more on most trip fares through a Pro Perk, plus priority airport queueing, a Trip Radar advantage, and VIP support. Upper-tier drivers at Gold and above with at least 2,000 lifetime trips also unlock fully funded ASU Online tuition. Separately, Uber can deactivate any driver for low ratings or safety issues, with a documented review-and-reinstatement path.

Put those two systems side by side and you see the paper's mechanism in the field. A deactivation threat against a driver with no earned tier costs them only the marginal trip income, so the low-stake response is to appeal, churn, or return with minimum service effort. A Diamond driver facing the same review is threatened with the loss of a tangible, hard-won bundle of earnings and perks, which gives them a strong reason to fix service quality rather than push volume.

The implication is that a pure deactivation threat only reliably improves the drivers who already hold losable status. For the large pool of low-tenure drivers, suspend-and-reinstate produces churn and grudging compliance, not better service.

The blunt ban is your weakest tool against the very users who generate the most low-quality volume.

Home-Sharing: Airbnb's Real Leverage Is the Badge, Not the Ban

Over roughly two years Airbnb removed more than 400,000 low-quality listings that failed its hosting standards, its largest-ever purge, and afterward reported a 15% drop in quality-related customer-service issues and an average listing rating above 4.75. In its first Global Quality Report the platform added that more than 80% of recent reviews now land at five stars and fewer than 1% at one star. In parallel it runs Superhost status, which demands a 4.8-plus rating, a cancellation rate below 1%, a 90% response rate, and at least 10 stays, reassessed quarterly and lost when the numbers slip. Superhosts get priority search placement and rewards, and the platform now counts more than 1 million of them worldwide.

This is the flip point made concrete. A marginal, low-review host who gets delisted has little sunk reputation to protect, so the reactance response is to relist under a fresh account or minimally patch the listing. A Superhost facing a quality warning is threatened with a badge that took a year of five-star stays to build, where offsetting a single one-star review requires roughly 19 five-star reviews. That host has far more to lose and invests in genuine repair.

So delisting cleans the catalog, but durable behavior change comes from hosts with reputation at stake. The leverage is the badge.

Creator Platforms: YouTube's Demonetization Only Bites the Creators Worth Keeping

In July 2025 YouTube tightened its Partner Program, renaming "repetitious" content to "inauthentic" and later expanding guidance to demonetize "mass-produced," "unsatisfying or off-putting," and AI-persona content, while still paying content that adds original commentary, editing, storytelling, or educational value. The behavioral split matches the paper exactly. Low-stake clip farms and compilation channels respond to demonetization the reactance way, spinning up new channels and flooding the platform with high-volume, low-effort uploads. A high-stake creator responds constructively. The owner of HPad, an 18-year appliance-testing channel with more than 225,000 subscribers, publicly reported losing at least $15,000 over three months when his classification shifted to "unsatisfying or off-putting." For a creator with that much invested brand equity, the rational move is to raise production quality, not to churn.

So demonetization suppresses payouts but not necessarily the low-quality flood, because uninvested accounts simply respawn and re-upload (see the exhibit). It becomes a quality-raising incentive only where channel identity and audience are the asset at risk.

App Stores: Google Play Pays for the Arms Race a Blunt Ban Sets Off

Google Play terminates developer accounts for accumulated policy strikes. On termination, all of a developer's apps are removed, they cannot publish new apps, and any linked accounts are permanently banned too, a rule built specifically to stop banned developers from opening fresh accounts. The scale of that fight is now public: in 2024 Google banned 158,000 developer accounts and blocked 2.36 million policy-violating apps from publishing, up from 2.28 million the year before, while also stopping 1.3 million apps from gaining unnecessary access to sensitive user data, with more than 92% of its harmful-app reviews now AI-assisted. The machine kept running in 2025, with more than 80,000 developer accounts banned and roughly 1.75 million apps blocked (see the exhibit). Those are the numbers of an expensive, permanent detection war, and the linked-account clause is the tell.

A low-stake spam or clone developer does exactly what reactance predicts, opening new accounts and re-publishing high-volume, low-quality apps to reclaim distribution. A developer with a genuine hit app and years of ratings holds a far larger sunk asset, so the constructive path in Google's own flow, fix the violations and appeal, is the rational one. Nate Shoffner's widely cited account of a Google Play termination shows the pattern from the ground: the ban cited linkage to a previously terminated account, confirming that Google spends real effort chasing respawns rather than reforming invested developers.

Account bans alone therefore invite whack-a-mole and force the platform to fund that arms race indefinitely. Enforcement turns constructive only where the developer holds a losable, hard-to-recreate asset.

The Rule: Enforcement Design Is Downstream of Status Design

Across all four platforms the pattern holds. A sanction only moves behavior in the direction you want when the target has accumulated visible, losable stake. Absent that stake, enforcement triggers reactance and hands you more volume, lower quality, and account churn. So build the thing users are afraid to lose first, whether tiered reputation, badges, monetization eligibility, or earnings perks, then aim enforcement at that asset rather than at raw access. Two moves recur everywhere: gate low-stake accounts at entry, because they will churn-and-burn no matter what you do afterward, and correct high-stake accounts by putting their status at risk rather than issuing a blunt ban. The blunt ban is your weakest tool against exactly the users who produce the most low-quality volume, and the status threat is your strongest tool for exactly the users worth keeping. Design the stake before you need the stick.