# AI reply bots on X: what the July 2026 purge removed and how to spot what's left

By Jonah Ease (https://trustaudience.com/about) · Published 4 Oct 2026 · Last checked 4 Oct 2026

On 24 July 2026, Nikita Bier, then X's head of product, said X had found 42,000 accounts automating replies with chatbots and removed them.[1] Plenty of reply bots remain: they leave generic replies that could sit under any post, their Replies tab is full while their own posts are thin, and a blue check doesn't clear them, because it shows a paid Premium subscription, not an ID check.[2] X's reply controls, hidden replies and the notifications Quality filter can cut much of the noise under your own posts.

## What X said on 24 and 25 July 2026

Bier's post went up on 24 July 2026 (20:12 UTC). It said X had removed 42,000 accounts that automated replies with chatbots, and that AI engaging with users without a human in the loop runs counter to X's mission.[1] Over roughly the next day he kept posting updates. (Bier stepped down as head of product in early August and stayed on as an adviser.[3]) Search Engine Journal's write-up of that thread, published 26 July, reports that he said most of the removed accounts were posting AI "thought-leadership" content to grow and win paid promotion deals, that nearly all spam on X is economically motivated, and that feeds should improve within 6 to 12 hours.[4]

What X didn't publish matters too. No detection method or false-positive rate came with the announcement. Techweez, writing in September, pointed out that X has not published a false-positive rate, so nobody outside X knows how often real users get caught.[5] Search Engine Journal mentioned one account flagged as spam that was reinstated about 12 hours after its owner asked for a review.[4]

The rules behind the purge were already on the books. X's automation rules say automated replies sent on the basis of keyword searches alone are not permitted, and that running any AI reply bot needs X's prior written approval.[6]

## Where the 1.7 million figure comes from

You'll see "X has removed 1.7 million bots" attached to the July story, sometimes as a 2026 running total. The number traces back to a different post. On 12 October 2025 Bier wrote that X had purged 1.7 million bots engaged in reply spam that week, and that direct-message spam was next.[7] So it's a one-week figure from October 2025, not a count for this year.

Between the two, X ran another visible wave: on 9 April 2026 Bier said X was identifying and suspending 208 bots per minute, a number he said was growing.[8] The full dated list, and what each wave did to follower counts, lives in [why your X follower count dropped](https://trustaudience.com/guides/why-x-follower-count-dropped).

## What an AI reply bot looks like in a thread

One reply rarely proves anything. The pattern shows up when you look at the account behind it.

**In the reply itself:**

- It restates the post in smoother words ("This is such an important point about…") and adds no detail, number or experience of its own.
- It ends on a generic question or an emoji, inviting a reply back.
- It lands within a minute or two of a big account posting, often near the top.
- It's about AI or productivity no matter what the post was about. That fits what Bier said most of the July accounts were posting.[4]

**On the profile:**

- The Replies tab shows the same template under unrelated posts: different topics, same three-sentence shape.
- The account's own posts are few, or are promotion for a newsletter, course or "DM for collabs".
- A complete profile: photo, bio, sometimes a checkmark. That's normal for these accounts, which is why profile checks miss them ([how to tell if your X followers are bots](https://trustaudience.com/guides/how-to-tell-if-x-followers-are-bots) covers that gap).

Where you are matters too. In March 2026 Bier said Japan has the most spam replies of any country, targeted by foreign monetization groups looking for verified followers.[9]

Be careful with "it sounds like ChatGPT". Techweez reports that Bier pointed to Paul Graham as a writer who gets caught because AI imitates his style so well.[5] Judge repetition across many replies, not tone in one.

## Why a blue check doesn't settle it

X's help page is explicit: the blue checkmark means an active X Premium subscription plus eligibility checks (a display name, a profile photo, a confirmed phone number, no recent profile changes, no signs of manipulation). It does not mean the account has been ID verified. The same page notes that the European Commission decided on 5 December 2025 that the checkmark misled users under the Digital Services Act, a decision X is appealing.[2]

Premium also buys reach in replies. Every Premium tier includes some level of "reply prioritization", and X says that when it ranks replies it considers, among other things, whether the replier is a Premium subscriber.[2][10] For anyone farming replies for visibility, the subscription is a business cost. Bier's March remark about spam groups hunting for verified followers points the same way.[9]

That's also why X's "Only verified accounts" reply setting is a blunt tool: it filters by who pays, not by how they behave.

## How to cut reply bots under your posts

All of these are built into X.

1. **Choose who can reply.** When you compose a post, tap "Everyone can reply" and pick Everyone, Accounts you follow, Only verified accounts, or Only accounts you mention. You can change it after posting from the post's More menu, under "Change who can reply".[10] "Accounts you follow" is the strongest filter, at the cost of replies from new people.
2. **Hide replies.** On your own post, open a reply's menu and choose Hide reply. Hidden replies stay viewable behind an icon, you can unhide them, and the replier isn't notified.[11]
3. **Let X collapse the obvious ones.** Replies X rates as lower quality sit behind a "Show probable spam" button, based on things like the account's past behaviour and the reply's content.[11]
4. **Filter your notifications.** The Quality filter hides lower-quality items such as duplicate posts and content that appears automated; it's on by default if you're new to X or have reinstalled the app. Advanced filters can mute notifications from new accounts, accounts with a default photo, accounts without a confirmed email or phone, and accounts you don't follow.[12]
5. **Mute, block, report, remove.** Report the clear cases as spam. If a reply account also follows you, you can remove it as a follower on the web without blocking.[13]

## What our "Repeated reply pattern" label means, and what it can't tell you

TrustAudience reads up to 10 recent public posts or replies from each of up to 100 of an account's newest followers ([method](https://trustaudience.com/methodology)). A follower is labelled **Repeated reply pattern** when at least two of the replies we read are substantially duplicated or interchangeable and add little to the topic. Replying often, or asking people to connect, doesn't qualify on its own. The label weighs 0 in the Audience score; Repetitive promotion weighs 25.

Across our 15 current-method public reports (method audience-context-2026-10-03-v3, checked 3 and 4 October 2026), 75 of the 984 newest followers whose posts we could read, about 8%, got the Repeated reply pattern label.[14] That's a small, non-random set of accounts people chose to check, so read it as a sense of scale, not a rate for X.

The limits are real:

- **It describes posts, not authors.** We don't establish whether an account is run by a person or a program, and we don't establish AI authorship. People paste templates too.
- **It covers followers, not repliers.** The accounts replying under your posts may not follow you at all, so a clean report doesn't mean clean replies.
- **It's a snapshot.** Ten posts per follower, from your newest followers, on one day. The Audience score does not verify audience quality.

## Check it by hand

To see how much of your reply section is template traffic:

1. Pick your last five posts that drew at least ten replies.
2. On each, open the top ten replies and tap through to each replier's Replies tab.
3. Read their last ten replies. Mark the account "repeating" if two or more would fit, unchanged, under almost any post.
4. Note whether each repeating account follows you, and whether it has a checkmark.
5. Count. If the repeating share is high, switch your next few posts to "Accounts you follow" and compare the replies you get.

## Sources

1. Nikita Bier on X, post of 24 July 2026 (42,000 accounts automating replies with chatbots removed) - X, https://x.com/nikitabier/status/2080747924380856519, accessed 2026-10-04
2. About X Blue Checkmark - X Help Center, https://help.x.com/en/managing-your-account/about-x-bluecheck, accessed 2026-10-04
3. X's head of product is leaving the company one year after joining (Karissa Bell, 5 August 2026) - Engadget, https://www.engadget.com/2231302/x-head-of-product-nikita-bier-leaving-the-company-one-year-after-joining/, accessed 2026-10-04
4. X Live-Tweets Its Fight Against Chatbot Spam In Real-Time (Roger Montti, 26 July 2026) - Search Engine Journal, https://www.searchenginejournal.com/x-live-tweets-its-fight-against-chatbot-spam-in-real-time/583572/, accessed 2026-10-04
5. X Is Fighting an AI Bot Problem, but Real Users Are Getting Caught Too (3 September 2026) - Techweez, https://techweez.com/2026/09/03/x-ai-generated-bot-replies/, accessed 2026-10-04
6. Automation rules (updated April 2026) - X Help Center, https://help.x.com/en/rules-and-policies/x-automation, accessed 2026-10-04
7. Nikita Bier on X, post of 12 October 2025 (1.7 million reply-spam bots purged that week) - X, https://x.com/nikitabier/status/1977446408136650785, accessed 2026-10-04
8. Nikita Bier on X, post of 9 April 2026 (208 bots suspended per minute) - X, https://x.com/nikitabier/status/2042121229998682517, accessed 2026-10-04
9. Nikita Bier on X, reply of 28 March 2026 (Japan has the highest spam replies) - X, https://x.com/nikitabier/status/2037766646761484362, accessed 2026-10-04
10. About conversations on X - X Help Center, https://help.x.com/en/using-x/x-conversations, accessed 2026-10-04
11. About replies and mentions - X Help Center, https://help.x.com/en/using-x/mentions-and-replies, accessed 2026-10-04
12. About the Notifications timeline - X Help Center, https://help.x.com/en/managing-your-account/understanding-the-notifications-timeline, accessed 2026-10-04
13. Following FAQs - X Help Center, https://help.x.com/en/using-x/following-faqs, accessed 2026-10-04
14. TrustAudience public reports: 15 reports on method audience-context-2026-10-03-v3 (model gpt-6-luna), checked 2026-10-03 to 2026-10-04; 75 Repeated reply pattern labels among 984 followers with readable posts - [TrustAudience audience board](https://trustaudience.com/), accessed 2026-10-04

---

TrustAudience is independent. It is not affiliated with, endorsed by or sponsored by X Corp. Other tools and companies are named for comparison only and have not reviewed this guide.

Source: https://trustaudience.com/guides/ai-reply-bots-on-x