On July 25, 2026, X removed roughly 42,000 accounts that used chatbots to automate their replies — the latest wave of a bot purge that has cleared well over 1.7 million spam accounts since 2025. The headline reads like another platform turning against AI, but the target is narrower and more precise: not AI-generated content, but AI-operated accounts. X is purging automated chatbots that engage users with no human in the loop, while the same app pushes a Grok button into the post composer to help people draft posts. This guide separates the two, explains the human-in-the-loop line the enforcement actually applies, details what counts as a chatbot spam account, sizes the purge, resolves the apparent Grok contradiction, and lays out how to run AI content on X at volume without tripping the exact behavior the detectors hunt: automation with no person behind it.
On July 25, 2026, X removed roughly 42,000 accounts that were using chatbots to automate their replies. Head of product Nikita Bier announced the sweep and stated the rationale in one sentence: "X's core value is providing an authentic pulse on humanity — and using AI to programmatically engage with users without a human in the loop runs counter to our mission." The removals are not a one-off. They are the latest visible wave of a bot purge X has run continuously since 2025, and together with earlier rounds it has cleared well over 1.7 million spam accounts.
The thing to read correctly is what was targeted. These were not accounts penalized for posting AI-written text. They were accounts running a chatbot as the operator — a script that watched the timeline and fired off replies automatically, with no person deciding what to say or when to say it. X's enforcement is aimed at that operating model: an automated account that engages on its own. Getting that distinction right is the whole game, because it separates a behavior X is actively suspending from a workflow X still openly encourages.
Every recent platform crackdown has a boundary, and on X the boundary is unusually clear once you see it: the platform is not against AI-generated content, it is against AI-operated accounts. Bier has been consistent about this — he called AI bot profiles "the scourge of X" back in 2025 — and the framing X keeps returning to is "a pulse on humanity." The value of the feed depends on the posts coming from people, or at least from accounts a person is actually driving. AI in the hands of a human is assistance. AI running an account by itself is automation. The first is welcome; the second is what gets purged.
This is why the human-in-the-loop phrase is doing real work rather than serving as PR. It names the exact test the enforcement applies. An account where a person writes, edits, or at minimum approves and publishes the content passes, because there is a human making the decisions the platform wants to protect. An account where a model decides autonomously what to reply and posts it with no human touching the loop fails, because that is precisely the programmatic engagement Bier described as counter to the mission. The detectors are looking for signs that a human is genuinely using the account, not for evidence that a model was involved somewhere in the drafting.
The clearest example is the automated reply bot: an account, usually running through the API or an automation tool, that scans for posts matching a keyword or a target and fires a generated reply at them at machine speed and machine volume. This is the "reply spam" X named in its earlier 1.7-million-account purge, and it is the behavior the 42,000-account sweep hit again. The tell is not the wording of any single reply — it is the pattern: hundreds or thousands of interactions no person could plausibly be typing, engagement that runs around the clock, and no evidence of a human ever actually opening the app to use it.
The category extends past reply bots to the broader family of accounts that engage without a person in the loop: auto-DM spam, which Bier flagged as the next focus after reply spam; scripted mass-following and mass-liking; and content-farm accounts that auto-post scraped or model-generated material on a schedule with nobody home. What ties them together is not that AI produced the words but that no human is operating the account. That is the through-line of the enforcement, and it is why "am I using AI?" is the wrong question to ask about your own account. The right question is "is a person actually running this?"
The numbers give the effort its shape. X has described the campaign as its most systematic bot-detection push to date. Through the spring of 2026, reporting put the suspension rate around 208 accounts per minute at peak — on the order of 300,000 a day — and the cumulative purge has passed 1.7 million accounts, with DM spam named as the next front. This is not a token gesture aimed at a handful of obvious offenders; it is sustained, high-volume removal that has been running for months.
The motive is the same economics driving every platform's move against synthetic spam: generative tools made it trivial to spin up accounts that flood the feed with automated engagement, and a feed drowning in bot activity loses the thing that makes it worth opening. Bier has warned that AI will keep making bots harder to detect and that platforms have to get ahead of a surge that could otherwise overwhelm genuine human activity. X's answer is to defend the "authentic pulse" directly — by removing the accounts that fake it. For the wider pattern across other platforms, see how the platform crackdowns on AI spam raise the quality bar for reach and the parallel move at TikTok against AI-generated spam.
The obvious objection is that X is purging AI bots while aggressively pushing its own AI into the app — a Grok button sits right in the post composer, offering to suggest and draft content. It looks contradictory only until you apply X's own test. Grok in the composer is assistance: it hands a suggestion to a human, who decides whether to use it, edits it, and chooses to publish. There is a person in the loop the entire time. A chatbot wired to auto-reply through the API is automation: no human sees the output before it ships. X is not being inconsistent — it is drawing the same human-in-the-loop line in both directions, encouraging AI that a person directs and removing AI that operates alone.
That resolves the paradox into a usable rule. The presence of AI in your workflow is not what puts an account at risk; the absence of a human is. It is the same principle underneath X's separate crackdown on engagement-bait phrasing and duplicate content, which targets the manufactured-signal behavior that ungoverned automation tends to produce — covered in depth in X's engagement bait detection update. Both enforcement tracks reward the same thing: a real account, run by a real person, posting original content.
If you draft posts with a model, review them, and publish them to your own account yourself, none of this is aimed at you — that is exactly the human-in-the-loop pattern X is protecting. The risk lives entirely in the operating model, not the tooling. Two workflows built from the identical AI outputs can land on opposite sides of the line: one where a person approves and publishes each post to their own accounts is fine, and one where a script auto-engages strangers with generated replies at volume is the thing being suspended.
So the safe posture for running AI content at real scale on X has a shape. Keep a human approving what ships rather than letting a model post autonomously. Publish your own original content to your own accounts, rather than operating reply bots that engage other people's posts on autopilot. Post through official, sanctioned publishing rather than gray-market automation tools that mimic human tapping. And make the content genuinely native and original rather than scraped or reworded, which keeps you clear of the duplicate-content and engagement-bait layers running alongside the bot purge. The goal is volume that still has a person behind it — which is also, not coincidentally, what beats the saturation of AI content across social media in the first place.
The reason Kompozy is worth naming against a chatbot-spam crackdown is that the line X drew — a human directing the account, publishing original content, no autonomous engagement — is the exact architecture Kompozy is built on, not a compliance feature bolted on afterward. Kompozy is a content generation and multi-platform publishing engine, and its default posture is a human in the loop by design: it generates content, but it fans that content to your own accounts through Autopilot behind a per-post review gate, so a person approves what ships. It is the assistance model X encourages, operating at scale — not the automation model X suspends.
Two structural choices keep it on the right side. First, everything Kompozy makes is net-new original content in your own identity: one idea expands across eighteen output formats — Persona Shorts and other avatar video via HeyGen, face-locked Persona Tweets, carousels, photo posts, blogs, newsletters — all governed by a Persona Brief that fixes your voice and point of view. That is first-party publishing of things you created, which is categorically different from an account rebroadcasting scraped or reworded material. Second, Kompozy publishes to your accounts on a schedule you set through sanctioned publishing rails; it does not operate reply bots, auto-DM strangers, mass-follow, or programmatically engage other people's posts — the specific behaviors the purge is removing.
The honest framing is a contrast in kind. If what you want is a bot that watches the timeline and auto-replies to strangers at machine speed, that is precisely what X is now suspending 42,000 accounts at a time for, and Kompozy is not that tool and never will be. What Kompozy gives you instead is the thing the crackdown makes more valuable: a way to produce a real volume of original, on-brand content across eight social platforms plus blog and email, with a human approving each post and your own identity behind all of it. In a feed where X is actively deleting the accounts that fake a human pulse, being an account that genuinely has one — generating and publishing at scale without ever handing the wheel to an autonomous bot — is not just safer, it is the whole advantage. For the reach side of the same shift, see how X now rewards mutual interactions.
On July 25, 2026, X's head of product Nikita Bier announced that the platform removed roughly 42,000 accounts caught using chatbots to automate their replies. He stated the rationale directly: "X's core value is providing an authentic pulse on humanity — and using AI to programmatically engage with users without a human in the loop runs counter to our mission." The sweep is the latest wave of an ongoing bot purge X has run since 2025, which has cleared well over 1.7 million spam accounts in total, with DM spam named as the next focus.
No. X is not penalizing accounts for posting AI-written text — it targets AI-operated accounts, which is a different thing. The boundary the enforcement applies is human-in-the-loop: an account a person actually writes, approves, and publishes from is fine, while an account where a chatbot decides what to reply and posts it autonomously is what gets suspended. The proof that this is not an anti-AI stance is that the same app promotes a Grok button in the post composer to help people draft content. AI as assistance to a human is welcome; AI running an account by itself is the target.
The clearest case is an automated reply bot — an account, usually driven through the API or an automation tool, that scans the timeline and fires generated replies at posts at machine speed and volume with no person deciding what to say. The category extends to auto-DM spam, scripted mass-following and mass-liking, and content-farm accounts that auto-post on a schedule with nobody operating them. The common thread is not that a model produced the words; it is that no human is running the account. That absence of a person in the loop is what the detectors are built to catch.
Not if a person is actually running the account. If you draft posts with a model, review them, and publish them to your own account yourself, you are on the human-in-the-loop side of the line X is protecting. The risk lives in the operating model, not the tooling: two workflows built from identical AI outputs land on opposite sides, depending on whether a human approves and publishes each post or a script auto-engages strangers with generated replies at volume. The safe posture is to keep a person approving what ships and to publish your own original content rather than operating reply bots.
It is sustained and high-volume, not a token action. X has described it as its most systematic bot-detection effort to date, running continuously since 2025. Reporting through the spring of 2026 put the suspension rate around 208 accounts per minute at peak — on the order of 300,000 a day — and the cumulative removals have passed 1.7 million accounts. The July 25, 2026 sweep that removed about 42,000 chatbot accounts is one visible wave of that longer campaign, and Bier has named DM spam as the next front after reply spam.
Keep a human in the loop and publish first-party original content. Concretely: have a person approve what ships rather than letting a model post autonomously; publish your own original posts to your own accounts instead of operating reply bots that auto-engage other people's posts; use official, sanctioned publishing rather than gray-market automation that mimics human tapping; and keep the content genuinely native and original so you also stay clear of the duplicate-content and engagement-bait layers running alongside the bot purge. The goal is volume that still has a person behind it — not a firehose of automated engagement.
On July 25, 2026, X removed roughly 42,000 accounts that used chatbots to automate replies — the latest wave of a bot purge that has cleared over 1.7 million spam accounts since 2025. The target is not AI content but AI automation. Head of product Nikita Bier framed the test as human-in-the-loop: programmatically engaging users with no person driving the account runs counter to X's mission, so automated chatbot accounts get suspended while AI-assisted, human-approved posting stays welcome.
Get started → · ← All guides · Compare Kompozy vs other tools