Last updated: August 31, 2026

Did X suspend my bot in the purge?

Probably, if it auto-replied. On April 9, 2026, X removed 42,000 accounts for chatbot-automated replies, peaking near 208 suspensions per minute. The documented signals were datacenter and VPS IPs, a high repost-to-original ratio, no organic browsing activity, and swarm coordination. Appeal once with a real explanation, then rebuild on scheduled original posts.

This page is for after the suspension. If your account is still live and you want the rules before you build, read the X automation rules for 2026 instead. What follows assumes the damage is already done.

Was it actually the purge?

Worth establishing, because the answer changes what you should do. Three conditions together make it near certain:

The date. On or around April 9, 2026. At 208 suspensions a minute the bulk of the wave landed inside a few hours, so purge suspensions cluster tightly in time rather than trickling in.
The behaviour. The account replied to posts it went looking for, rather than posts that mentioned it. That is the specific pattern the wave targeted.
The notice. It cited platform manipulation or spam in general terms rather than pointing at one offending post. Batch enforcement does not itemise.

If your suspension came weeks later, named a specific post, or hit an account that only ever published its own content, it is ordinary enforcement and not this. Different cause, different appeal.

One thing this was not

The purge did not target automation as such. Scheduled posting through an authorised app was not in scope, before or after. What X removed was automated engagement with other people, which is a narrower and much more specific thing.

The four detection signals

Individually weak. Together they describe a server-run reply bot and essentially nothing else, which is why the false positive rate was low enough to run the wave in batch.

Datacenter and VPS IP ranges

The account transacted with X exclusively from cloud infrastructure. Real people use residential and mobile networks. A handle that has only ever been seen from an AWS or Hetzner range is not being used by a person.

What to do about it: There is no clean fix, and there should not be. Routing a bot through residential proxies to look human is itself a rules violation and a worse one.

High repost-to-original ratio

The account almost never produced anything of its own. When the overwhelming majority of activity is reposting and reacting to other people, the account has no independent reason to exist.

What to do about it: Publish original content. This one is genuinely fixable, and it happens to be the only remaining automatable write path.

No organic browsing signal

A human account generates a long tail of incidental activity: scrolling, opening posts, viewing profiles, dwelling. An API-only account generates writes and nothing else. The absence is as legible as the presence.

What to do about it: Use the account. If nobody ever logs into it, it looks exactly like what it is.

Swarm coordination

Multiple accounts posting in step: same cadence, same phrasings, same targets, same infrastructure. This is the signal that turned individual suspensions into a 42,000-account batch, because catching one account exposed the whole cluster.

What to do about it: One account per pipeline, genuinely separate. If your accounts share a template and a timer, they share a fate.

Why the four together matter more than any one

A legitimate scheduled-posting integration also runs from a datacenter IP. What it does not have is the other three: it produces original content, the human owner actually uses the account, and it is not moving in lockstep with forty other handles. The signal X acted on was the full combination.

What to do now

Appeal once, in plain language

Use the in-app prompt or the account access form in the X Help Center. Describe what the account actually did. One clear appeal beats five.

Turn off the automation first

If your job is still hammering the API against a suspended account, stop it before you appeal. It is still generating the signal you are appealing about, and on pay-per-use it is still billing.

Export what you can from other accounts

If you run several accounts and only some were caught, treat the survivors as at risk and get your content out of a system that can disappear.

Audit the other accounts against the four signals

The wave was cluster-driven. If one account in a group matched, assume the rest are visible too, and change the behaviour rather than waiting.

Create a replacement account

Ban evasion. A separate and more serious violation, and it typically takes the new account down along with anything sharing its signals.

Appeal repeatedly

Repeated appeals on the same case do not move it, and a burst of appeals across accounts is itself a coordination signal.

Move the bot behind residential proxies

This is deliberately evading detection, which is a violation in its own right regardless of what the bot then does.

Rebuild the same bot on the API

Not possible on self-serve access any more. Replies need a summon and the like, follow, and quote-post writes are Enterprise-only.

We cannot tell you the odds on any individual appeal, and anyone quoting you a success rate is guessing. What we can tell you is that the four items on the right make things worse rather than better.

The bot you had cannot be rebuilt

This is worth internalising before you spend a weekend on version two. Even with a restored account and perfect infrastructure hygiene, the API surface an automated reply bot needs no longer exists on self-serve access.

Since February 23, 2026, a reply through POST /2/tweets is only accepted when the original author mentioned or quoted your account first. See the reply restriction guide.
Between April 16 and 20, 2026, like, follow, and quote-post writes moved to Enterprise-only access at roughly $42,000 a month. See what was removed and what survived.

Read the 2026 changes as one policy rather than three incidents. The suspension wave removed the accounts, and the endpoint changes removed the capability, so that the same thing could not simply be rebuilt. Anything you find online describing an auto-liker, a follow-back loop, or a keyword reply bot is now describing an integration that cannot be built.

What to build instead

The line X drew is between automating what you say and automating engagement with other people. The first side is fully open and always was.

Scheduled original posts and threads. Untouched by every 2026 change, and the highest-value thing you can produce anyway.
Retweets. The one engagement write that survived April 2026. Used deliberately, not on a timer. See the auto-retweet guide.
Human-approved DMs. Still supported, and still the highest-intent channel on X when a person signs off each send. See automating X DMs without getting banned.
Replies you were summoned into. Fewer, but they are the conversations where somebody actually asked for you.

OpenTweet posts through the official X OAuth connection with no developer account, and it deliberately does not offer automated liking, following, or mass replying. That is not a feature gap. Those are the write paths that got 42,000 accounts removed, and two of them do not exist on self-serve access any more. If you want to check the health of an account that survived, the shadowban checker is a reasonable first look.

7-day free trial. Cancel anytime.

Frequently asked questions

What was the April 2026 X bot purge?

On April 9, 2026 X removed roughly 42,000 accounts in a single enforcement wave targeting chatbot-automated replies. At its peak the wave ran at close to 208 suspensions per minute. It was a batch action driven by behavioural signals, not individual reports, which is why so many accounts went down within the same few hours.

How do I know if my bot was caught in the purge?

Three things together make it near certain: the suspension landed on or around April 9, 2026, the account sent automated replies to posts it found rather than posts that mentioned it, and the enforcement notice referenced platform manipulation or spam rather than a specific post. Suspensions on other dates for other reasons are ordinary enforcement, not the purge.

What signals did X use to detect automated reply bots?

Four were documented: traffic originating from datacenter and VPS IP ranges, a high repost-to-original content ratio, no organic browsing signal on the account, and swarm coordination across accounts posting in step. Any one is weak evidence. All four together describe a server-run reply bot and nothing else.

Can I appeal an X suspension from the purge?

Yes. Use the in-app appeal prompt or the account access form in the X Help Center. Appeal once, in plain language, describing what the account actually did. Repeated appeals on the same case do not help and filing many of them across accounts looks like exactly the coordinated behaviour that triggered the action.

Should I make a new account to replace the suspended one?

No. Creating a replacement account to get around a suspension is ban evasion, which is a separate and more serious violation than the original one. It also tends to take down the new account and any other accounts sharing the same signals, phone number, or payment method.

Can I just rebuild the same reply bot more carefully?

You cannot rebuild it at all on self-serve API access. Since February 23, 2026 a reply through POST /2/tweets is only accepted when the original author mentioned or quoted your account first, and in April 2026 like, follow, and quote-post writes moved to Enterprise-only pricing. The endpoints an automated reply bot needs are no longer available at any self-serve price.

What is safe to automate on X after the purge?

Publishing your own content. Scheduled original posts, threads, retweets, and human-approved DMs are all supported write paths that survived every 2026 restriction. The line X enforced is between automating what you say and automating engagement with other people, and only the first side is still open.

Automate what you say, not what you like

Scheduled posts, threads, and articles through the official X connection. No developer account, no engagement automation, nothing that got 42,000 accounts removed.