What the June 2026 X bot purge flagged

On April 9, 2026, X Head of Product Nikita Bier put a number on the bot purge: roughly 208 accounts suspended per minute at peak. Most of what got swept was not covert. It failed three checks X runs on every request: IP origin, device fingerprint, and genuine human touch input.

The X bot purge June 2026 ran at 208 suspensions per minute

On April 9, 2026, X Head of Product Nikita Bier put a number on the bot purge: roughly 208 accounts suspended per minute during the peak enforcement phase. That rate is the headline most coverage led with. It is also the least useful fact for anyone trying to figure out whether their own account is at risk.

The April wave was not a one-off. It was the latest move in a campaign that started in October 2025, when X removed 1.7 million reply-spam bots in a single action. February 2026 brought a dedicated initiative against AI-powered bots, which introduced 'human-only interaction' rules that flagged accounts with no detectable manual activity. By March 2026, Bier said roughly half the product team was working on spam mitigation, with the purge features themselves launching in April.

We find it clearer to read the whole thing as distinct phases. October 2025 hit reply spam. February 2026 targeted AI-powered bots. March 2026 was a faulty classifier that X corrected within 12 hours. April 2026 was peak enforcement. The activity people are seeing in June 2026 is the same campaign still running, not a new event.

Scale matters for context. Estimates put the total bot population on X somewhere between 50 and 84 million, derived from 9 to 15 percent of roughly 560 million monthly active users. Even an aggressive purge removes detectable cohorts rather than draining the pool. That is why the waves keep coming back: the detectable layer regenerates, and X keeps re-running the filter.

What types of accounts did the April 2026 X bot purge flag?

The April 2026 purge flagged a handful of recognizable profiles. Pure automation bots with no human session activity. Accounts with high repost-to-original-content ratios and no organic browsing signals. Accounts operating from datacenter or VPS IP addresses. And coordinated networks showing swarm-style behavior. Most suspended accounts matched more than one of these at once.

Legitimate creators got caught too. NSFW accounts, alt accounts, and accounts that lean heavily on reposts were swept in as false positives. Some lost more than 20,000 followers; at least one account lost 4,000 or more in a matter of days. Appeals, as we cover later, were described by affected creators as largely ineffective.

The false-positive pattern was consistent enough to name. The accounts that got misclassified used automation to repost heavily but rarely logged into X in a real browser. They generated no organic scroll, no tap events, no variable dwell time. At the session layer, an account that only ever reposts on a schedule looks identical to a spam bot, because the thing that separates the two, a human actually using the app, never shows up in the telemetry.

Profile completeness turned out to be a quiet weighting factor. Accounts with a phone number verified before February 2026, a non-default profile photo, a filled-in bio, and at least 90 days of age experienced materially lower suspension rates, even when their behavioral signals sat in the gray zone. The classifier treats those attributes as prior evidence that a real person is behind the account.

The corollary is the part most guides skip. New accounts running any automation before they reached that completeness baseline were the single highest-risk cohort across every wave. If you are going to automate anything, the account needs an established identity first. A blank profile posting on a schedule from day one is the exact shape the filter was built to catch.

X moved bot detection into the API gateway in late 2024

X moved bot detection off client-side JavaScript and into the API gateway in late 2024. The old approach checked for bots with scripts that ran in the browser, which any headless browser able to execute JavaScript could satisfy. Moving the checks to the gateway means every request gets profiled before it reaches the application, no matter what client sent it.

The gateway scores four signals on every request. Cookie entropy, meaning how consistent and realistic the session cookie chain looks. Device fingerprint, covering browser version, screen resolution, font list, and hardware specifics. IP reputation, which asks whether the originating address belongs to a known hosting provider, datacenter, proxy, or a residential block. And the presence or absence of genuine human touch and tap input captured upstream.

Of those four, IP reputation did the most damage in the April wave, and it is the one almost nobody writes about. Home residential IP addresses receive materially lower suspicion scores at the gateway than datacenter, VPS, or proxy addresses. Automation running on a user's own home connection sidesteps the largest single signal that flagged most purged bots: a mismatch between the browser session a request claims to be and a known hosting-provider address block.

This is why cloud-based automation tools got hit so hard. A tool hosted in a datacenter announces its origin on every request, independent of how careful its posting behavior is. We watched the April wave sweep accounts whose content output was unremarkable, purely because the requests came from an address range X associates with bulk automation. The hosting choice, not the content, set the risk profile for a large share of those suspensions.

None of this is in X's public policy pages. That gap explains the most frustrating outcome of 2026: accounts that followed every published automation guideline still got suspended, because the rule that actually mattered was never written down.

Why legitimate creators got caught in the June 2026 X bot purge

Two different things got conflated in coverage of the 2026 purge, and separating them matters. The March 12, 2026 ban wave, first reported in Japan, was a code problem. A faulty spam filter misclassified legitimate accounts for roughly 12 hours. Bier acknowledged the error, and X said 99 percent of mistakenly suspended accounts were restored.

The April wave was not a bug. Its false positives came from the classifier doing exactly what it was designed to do. Accounts with a high repost-to-original ratio and no detectable manual activity were scored the same as pure spam bots, because at the session layer they produced the same evidence: none. The signal that would have cleared them, real human use, simply was not there to read.

Concretely, an account that used a scheduling tool to repost curated content all day but never opened X organically lacked the scroll depth, the variable dwell time, and the tap event signatures that detection now treats as proof of a human. The automation was not the disqualifier. The absence of any countervailing human signal was.

Appeals during the April wave were largely ineffective, by the accounts of the people filing them. X's automated review does not tell you which signal triggered the suspension, so you are arguing against a decision without knowing its basis. Absent direct evidence of human session activity, there is little to push back with.

Do not assume the March timeline applies to April. The 12-hour, 99-percent recovery happened because X publicly admitted a classifier error and reversed it in bulk. April was described as intentional enforcement, and it carried no equivalent commitment to restore anyone. The fast, clean reversal was the exception, not the template.

Swarm detection, IP reputation, and session fingerprinting: the three detection layers

X's detection works in three layers, and knowing which one flagged you determines what actually fixes the problem. The gateway layer, covered above, handles IP reputation, device fingerprint, cookie entropy, and touch input. The second and third layers catch things the gateway cannot see on a single request.

The second layer is swarm and coordinated behavior detection. X clusters accounts by conversation thread, then checks for text similarity across replies, how many usernames in the cluster match bot-naming patterns, how many accounts in the group are brand new, and whether the same suspicious URLs appear across multiple accounts. You do not have to be a bot to get flagged here. Being part of a coordinated group is enough to land you in a suspension review.

The third layer is temporal pattern analysis. X flags accounts whose posting cadence is unnaturally regular. An account that posts at exactly :00 past the hour, five times a day, produces a statistical signature no human generates. The counterintuitive part, and we have seen this hold up repeatedly, is that volume matters less than regularity. Posting a lot is not the tell. Posting like a metronome is.

In practice, an account posting 10 times a day at randomized intervals survived the purge more reliably than one posting 5 times a day at fixed intervals. The fix is mechanical: put a jitter window of at least 8 to 22 minutes around any scheduled post time, and vary the daily post count by plus or minus 2. That is usually enough to break the regularity the timing detector is built to find.

The reason to identify the layer is that the responses do not transfer. An IP reputation flag is a hosting problem. A temporal flag is a scheduling problem. A swarm association is a who-you-reply-to problem. Changing your posting times does nothing for an account that got caught because of its datacenter IP, and the reverse holds just as firmly.

How the X bot purge June 2026 changed creator monetization

If you lost followers to the purge, here is the part that should lower your blood pressure: X's payout model now counts impressions only from verified and Premium users. Bot-driven follower counts and bot views never contributed to monetization in the first place. Removing them improves payout quality for accounts with real audiences rather than hurting it.

The numbers underneath this are worth sitting with. X paid $415 million to creators in 2025, up from $260 million in 2024. The distribution is steeply skewed: the top 1 percent of monetized accounts earn over $52,000 a year, while the median monetized account earns under $400 a year. A large follower count has never mapped cleanly onto that revenue.

So bot-inflated follower counts were already a bad proxy for earning potential. The purge does not widen the gap between followers and revenue; it just makes the gap visible. An account that looked big because of bots and earned little was always in that position. The follower drop only removed the cosmetic part.

The practical consequence splits two ways. Accounts that kept their verified and Premium followers after the drop will see little or no change in payouts. Accounts whose engagement was bot-driven will watch payouts settle to their real audience size, which is roughly where they already were.

The single check worth running is impressions per post. If your average impressions held steady after the follower count fell, the accounts you lost were not contributing to reach or revenue, and there is nothing to recover. If impressions fell in proportion, the lost accounts were real, and that is a different conversation.

Schedule posts, not engagement: the automation rule that keeps X accounts safe in 2026

X's Platform Manipulation and Spam Policy is explicit about what is banned: automated follows and unfollows, mass identical replies, bulk direct messages, and engagement automation of any kind. Violations move you up a tiered enforcement ladder that starts with feature restriction and ends in permanent suspension. The ladder is real, and the purge accelerated how fast accounts climb it.

The rule that separated survivors from suspensions is short. Automate content creation and scheduling. Never automate engagement. Mass likes, mass follows and unfollows, auto-retweets, and keyword-triggered replies are the highest-risk automation category in 2026, full stop. If an action reaches out and touches another user's account, do not let a machine do it.

There is a subtler version of this that caught a lot of accounts. Engagement actions performed inside the same automated session as a posting action are a compound signal. A session that posts content and then immediately likes or follows other accounts does not look like how humans behave. The safe practice we have seen hold up is to separate posting sessions from any engagement by at least 15 to 30 minutes of idle time, or to keep engagement entirely in manual sessions.

X's automation policy also requires any API-based bot account to disclose that it is automated in its profile bio, and X is rolling out automated profile labels to surface that at the platform level rather than trusting accounts to self-report. If you run a disclosed, posting-only bot, this is fine. The labels are aimed at accounts pretending to be human, not at transparent automation.

The Free-tier API limits that correlate with suspension risk sit around 2,400 posts per day, 400 follows per day, 1,000 likes per day, and 500 direct messages per day. The trap is treating those as targets. Running near any of them on a consistent daily pattern is a primary suspension trigger, more than the raw volume is. Combine the volume discipline with the jitter window of 8 to 22 minutes and the plus-or-minus 2 daily variation from earlier, and you avoid both the rate ceiling and the regularity signal at once.

If your account is suspended: the X ban appeal process in 2026

If your account is suspended, the appeal runs through X's in-app support form at help.x.com. During the April 2026 wave, response times ranged from hours to weeks, with no published SLA. Set expectations accordingly and file once, cleanly, rather than spamming the form.

The form does not tell you which policy X believes you broke, which makes a generic appeal weak. The appeals that worked best post-purge did three things: stated plainly that the account is human-operated, included evidence of original content creation, and described any automation tools in use. If you use a scheduling tool, name it and specify that it handles posting only, not engagement. That specificity reads very differently from a bare 'I am not a bot.'

Calibrate your expectations against the right wave. The March 2026 false-positive episode resolved in about 12 hours with 99 percent of accounts restored, but that happened because X publicly owned a classifier error and reversed it in bulk. The April enforcement wave came with no such public commitment. If you were caught in intentional enforcement, individual appeal quality matters more than waiting for a mass reversal that may never come.

One concrete lever helps at the review stage. Adding a verified phone number before you submit strengthens the identity signal X weighs. Accounts without a linked phone had lower appeal success rates during the 2026 waves, and the same profile completeness that lowered suspension odds, a verified phone, an original avatar, a real bio, and account age, keeps working as a weighting factor even once a human is reviewing the case.

If you run automation and want to avoid this entire process, the throughline of the whole purge is the cheapest insurance: an established, complete profile, posting on a residential connection with randomized timing, that never automates engagement. Almost everything that got swept in 2026 violated at least one part of that sentence.

Frequently asked questions

Why am I losing followers on X in 2026 and is it the bot purge?

If you lost followers in late March, April, or June 2026, the most likely cause is X's ongoing bot purge campaign, which has run in distinct waves since October 2025. Follower drops from a purge are permanent; those accounts are removed from the platform. If your impressions per post did not drop alongside your follower count, the removed accounts were not contributing to your reach.

What types of accounts did the X April 2026 bot purge actually flag and suspend?

The April 2026 purge targeted pure automation bots with no human session activity, accounts with high repost-to-original-content ratios and no organic browsing signals, accounts operating from datacenter or VPS IP addresses, and coordinated account networks showing swarm behavior. Some legitimate creator accounts with heavy repost usage and no manual login activity were swept in as false positives.

How does X detect automated accounts and what behavioral signals trigger a suspension in 2026?

X's detection runs at the API gateway and checks four signals on every request: IP address reputation (residential vs. datacenter), device fingerprint, cookie entropy, and the presence of genuine human touch input. On top of that, X applies swarm detection (coordinated network clustering by thread) and temporal pattern analysis (unnaturally regular posting schedules). Any single layer can trigger a review; multiple signals together typically result in suspension.

Can a legitimate account get suspended in the X bot purge and how do you appeal?

Yes. The April 2026 purge suspended legitimate creator accounts as false positives, particularly those with high repost ratios and no organic login activity. To appeal, submit through X's support form at help.x.com, state clearly that the account is human-operated, describe any automation tools used for posting only rather than engagement, and include evidence of original content creation. Adding a phone number before or during the appeal improves your outcome.

Does losing bot followers hurt my X creator monetization payouts?

No. X's current revenue payout model counts only impressions from verified and Premium users. Bot followers do not contribute to monetization, so losing them does not reduce payouts. The practical check is whether your average impressions per post held steady after the follower drop. If impressions were unchanged, your real audience was unaffected and your payout should be too.

What is the difference between allowed automation and banned automation on X in 2026?

Allowed automation covers content creation and scheduling: drafting posts with AI tools and publishing them on a schedule with randomized timing. Banned automation covers all engagement actions: automated follows and unfollows, mass likes, auto-retweets, bulk direct messages, and keyword-triggered replies. The consistent rule from the 2026 purge is to never automate any action that directly interacts with another user's account.

How long does the X bot purge last and when will follower counts stabilize?

The 2026 purge campaign has run in distinct waves across October 2025, February 2026, March 2026, April 2026, and into June. There is no published timeline for conclusion. Follower counts typically stabilize within days of each individual wave ending, but new waves restart the drop. X has not announced a completion date for the overall campaign.

Did the March 2026 X ban wave affect real accounts and what caused it?

Yes. The March 12, 2026 ban wave was caused by a faulty spam filter that misclassified legitimate accounts for approximately 12 hours. X Head of Product Nikita Bier acknowledged the error publicly and stated that 99 percent of mistakenly suspended accounts were restored. This wave was a classifier failure, distinct from the intentional April enforcement campaign, and should not be used as a benchmark for how April-wave appeals would be handled.

Is using a scheduling tool safe on X or will it trigger bot detection in 2026?

Scheduling tools are generally safe if they handle posting only and do not automate engagement. The key risk factors are IP origin (cloud-based tools routing through datacenters carry higher detection risk than tools running on residential connections), cadence regularity (fixed-interval scheduling is a suspension trigger), and whether the tool performs any engagement actions. Posting-only schedulers with randomized timing and residential IP routing survived the 2026 purges at materially higher rates.

What posting frequency limits should I stay within to avoid X spam flags in 2026?

Third-party analysis of X's API policy cites approximately 2,400 posts per day, 400 follows per day, 1,000 likes per day, and 500 direct messages per day as Free-tier thresholds. Approaching these limits on a consistent daily pattern is more dangerous than the volume alone. Vary daily post counts by plus or minus 2 and use a jitter window of 8 to 22 minutes around any scheduled post time to avoid the regularity signal that X's temporal pattern detection targets.

Sources and further reading

• X Platform Manipulation and Spam Policy
• X Automation Rules
• Social Media Today coverage of the X bot purge with Nikita Bier's figures