Running an X Bot Account for a Month: Reaching 4,000 Followers and Analyzing the SPAM_HIGH_RECALL 1.91% Filter Rate
A developer who scaled an X account from 15 to 4,175 followers using a script-and-agent automation pipeline analyzes platform reach limits and the SPAM_HIGH_REC
The results of a one-month experiment combining automation scripts and large language models (LLMs) to run a social media account on X (formerly Twitter) have been published. Starting from just 15 initial followers, the developer managed to grow the audience to over 4,000, while closely analyzing how X's automated spam detection systems flagged posts for reach limitations.

Image source: https://dev.to/alexlee7359/i-let-a-bot-run-my-x-account-for-a-month-15-to-4000-followers-37-posts-flagged-as-spam-2cg2
The experiment ran from October 23 to November 13, 2026. The developer deployed a bot pipeline consisting of scripts executing repetitive delivery tasks and an LLM agent that read timeline posts to draft contextual, personalized replies. Using X's platform transparency report tool, "Under the Hood," the operator was able to examine the exact behavior of the platform's spam filters and evaluate the real-world effectiveness and limitations of mass follow-back growth strategies.
Division of Labor Between Scripts and LLM Agents
The core of the experiment lay in separating mechanical tasks from intelligent operations requiring contextual understanding.
- The Script's Role: Mechanical actions repeating fixed steps were handled entirely by scripts. These tasks included automatically scraping follow-back call-out posts published by verified users, dropping likes, and dispatching batches up to a daily cap.
- The Agent's Role: Reading posts and writing natural, context-aware replies was delegated to the LLM agent. Under strict rules set by the operator, the agent first addressed the substance of the target post, introduced the author briefly, and requested a follow-back.
While the agent initially handled reply dispatches as well, the process was streamlined by November so that scripts executed the actual sending step, recognizing that transmitting a message is purely mechanical and requires no active reasoning. Between October 23 and October 31, records show 741 friend-making comments were sent over nine days.
Surpassing 4,000 Followers and the Reality of SPAM_HIGH_RECALL Labeling
The growth velocity of the follow-back pipeline was striking. Starting with 15 followers in late October, the count reached 2,923 by November 6, crossed 4,000 on the night of November 12, and hit 4,175 by November 13.
However, this aggressive growth collided directly with X's platform detection systems. According to the monthly transparency report generated by X's "Under the Hood" feature, out of 1,937 posts sent in October, 37 posts (1.91%) were flagged with a SPAM_HIGH_RECALL label. Account-level penalties remained at zero, meaning the platform did not throttle the account as a whole.
- Meaning of the Label: The
SPAM_HIGH_RECALLtag indicates that automated systems detected a post as potentially containing spam. The consequence is singular: the post is blocked from recommendations to non-followers, while existing followers can still view it normally in their timelines. - Concentration of Flags: Of the 37 flagged posts, 31 (84%) were concentrated on October 24 (13 posts) and October 25 (18 posts)—precisely the days when friend-making comment volumes hit the daily ceiling of 120.
Inspecting the original texts of these flagged posts revealed that repeating similar comment templates combined with a uniform hashtag (#蓝V互关) across a high-frequency burst served as the primary trigger for the filter.
Limits of Reach Efficiency and Revisions to the Automation Pipeline
Beneath the rapid follower growth lay structural limitations. A significant portion of the users acquired via follow-back methods were themselves bot-operated accounts or passive users who rarely engaged with home timelines.
Consequently, even when the platform pushed posts out in volume, actual interaction rates remained low. Although certain posts achieved between 17,000 and 20,000 views, actual clicks and engagement stagnated at 200 to 400 interactions. Furthermore, the developer suspected that excessive burst messaging likely triggered unpublished internal weighting rules, throttling post reach to a fraction of its normal baseline.
Recognizing these bottlenecks, the developer shut down the follow-back pipeline on the night of November 12. Because continuing to accumulate followers through automated call-outs required maintaining the exact posting patterns flagged by the transparency report, the scripts were updated to discard all follow-back and recruitment posts entirely. The reply pipeline remained active, but shifted its focus strictly toward substantive technical and open-source discussions.
Sources
- DEV Community: I Let a Bot Run My X Account for a Month: 15 to 4,000 Followers, 37 Posts Flagged as Spam
- X Under the Hood Report: X Under the Hood Transparency Report
- GitHub: X Open Source Recommendation Algorithm Code
- Alex Lee Blog: X Algorithm Teardown
- GitHub Skills Library: For You Skills