Analysis of 168,907 tweets at Daryl Morey following his Oct. 4 pro-HK tweet: ~50% came from accounts with 0-13 followers, 4,855 users had never tweeted
'This wasn't primarily a bot swarm. It was a troll mob.' After his tweet supporting Hong Kong's protesters, the Houston Rockets' general manager … Tweets: @coinbiased and @jadande See also Mediagazer Tweets: @coinbiased : Consider this next time you scroll the comments section https://twitter.com/... J.A. Adande / @jadande : “There were 4,855 total users in all who had never tweeted until they replied to Morey, and 3,677 accounts didn't exist until his tweet.” Curious about those 1,178 “sleeper” accounts. Were they just waiting on an NBA GM to tweet about Chinese policy? https://www.wsj.com/... See also Mediagazer
Context & Ripple Effects
Daryl Morey's Oct. 4 tweet supporting Hong Kong's protesters landed weeks after coverage of Hong Kong protesters turning to Twitter and the struggles new users faced getting onto the platform. The Wall Street Journal's analysis of the 168,907 tweets directed at him now reframes what looked like an organic nationalist backlash as something engineered.
The finding that roughly half the replies came from accounts with 0–13 followers, including 4,855 users who had never tweeted before, echoes earlier coordinated campaigns Twitter has acted on — such as the hundreds of pro-Saudi accounts suspended over Khashoggi tweets — but with a twist: these accounts are largely human-operated, not automated.
First-order effects
- The Houston Rockets and the NBA are dealing with a harassment wave driven by real, low-activity accounts, meaning conventional bot-purging tools catch little of it while the pile-on continues in plain sight.
Second-order effects
- Twitter's own framing of spam — its CEO later claimed human reviews found under 5% of mDAU were spam (a counting method built around automation signals) — is exposed as ill-suited to troll mobs, pressuring the platform to redefine what counts as manipulated activity.
Third-order effects
- If influence operations keep favoring dormant or freshly created human accounts over botnets, detection shifts from technical fingerprints to behavioral patterns — the same distinction that surfaced when Chinese-language bots drowned out zero-COVID protest tweets three years later, suggesting a durable playbook rather than a one-off.
The trend: Coordinated influence campaigns are shifting from automated bot swarms to human-run troll mobs built on empty accounts, slipping past spam filters calibrated to detect machines.