X purges 42,000 chatbots that sold fake engagement

Fake popularity as a product

X removed 42,000 accounts that were automating replies using chatbots on 24 July 2026, with the platform's head of product Nikita Bier announcing the purge in a public thread that tracked the operation in real time.verifiedVerified Source: x.com^[1] These were not random nuisance bots. They were built deliberately to manufacture the appearance of popularity and then monetised.

Bier said the business model was straightforward: grow an account to look credible, then sell or rent it to a company wanting paid promotion. Bier said X found the accounts were turning AI-generated replies into a commercial product, with operators treating manufactured engagement as inventory.verifiedVerified Source: x.com^[1]

One operator, 40 pivots

Not every spam network folded quietly. Bier said some operators had been adapting their methods for six months, cycling through new approaches each time X's systems shut one down.verifiedVerified Source: x.com^[2] One stood out for sheer persistence.

That operator changed tactics 40 times over six months before finally being shut down, with the account pattern centred on promoting trading content.^[2] Each pivot required effort and carried risk, yet the operator kept returning because the revenue from selling fake reach made it worthwhile.

Response times drop from months to hours

X blocked the Grok-powered spam operator on 25 July 2026, the day after the initial announcement.^[3] Bier pointed to that speed as evidence of a structural shift in how the platform handles new threats.

X's turnaround time for identifying and acting on new spam has fallen to 12 to 18 hours, compared with months under the platform's prior leadership.^[3] The public thread itself was part of the response strategy: by documenting each operator's methods openly, X put spammers on notice that detection would be fast and public.

Why engagement numbers are now weaker evidence

The economics of the scheme explain why the problem matters beyond X's own platform health. Operators were not spamming for its own sake. They were building assets: accounts with high reply volumes and apparent follower engagement that looked, on the surface, like genuine influence.

Businesses and marketers who rely on reply counts or follower numbers to judge whether a creator has real pull are now working with a metric that has been actively gamed at commercial scale. The 42,000 accounts removed in this single action represent only the operators X caught, not the full extent of the market for manufactured engagement.^[1]

What Australian marketers should do differently

For Australian brands spending on influencer partnerships or sponsored content on X, the episode sharpens a question that good media buyers have always asked: what is the evidence that this audience is real? Follower counts and reply volumes are the easiest figures to read and the easiest to fake. Deeper signals, such as the consistency of engagement over time, the ratio of replies to original posts, and whether the same accounts appear across multiple sponsored threads, are harder to manufacture at scale.

The operator documented by Bier spent six months adapting to stay inside the platform, suggesting the returns from fake reach were large enough to justify sustained investment in evasion.^[2] X's response time has now fallen to 12 to 18 hours for new threats, but the window between a new tactic appearing and detection remains open long enough for campaigns built on fake engagement to run and bill.^[3]