Article
You Are Human. Why Are You Being Treated Like a Farm Bot?
14 Oct 2026
X’s systems do not care that you are a real person. They look for patterns. If your behavior matches the patterns they built rules and models to catch, you get treated the same as a farm.
X’s systems do not care that you are a real person. They look for patterns. If your behavior matches the patterns they built rules and models to catch, you get treated the same as a farm.
The open-source code shows the layers they actually use.
What the code flags
Duplicate text
Botmaker rules (BBQDuplicateTextProd and the replies version) scan for identical or near-identical text. When the same sentence or near-copy appears across posts or replies, those posts get labeled COPYPASTA_SPAM. One repeated line is enough to start the label.
Self-reply chains
A filter added in September 2026 specifically looks for chains of replies where you answer your own reply under a larger account (1,000+ followers). An AI model scores the thread. If it flags enough of your replies, those replies receive a spam label and a reply ranking score of 0. They sink.
Low-quality and generic replies
Replies are scored by a model from 0 to 3. Low-effort, generic, or template replies get buried. Labels such as llm_slop_user and SpamHighRecall also exist. Soft versions of these labels last 30 days and limit visibility on the recommendations path.
Bad URL patterns
Rules flag low-quality or malicious link chains, especially when combined with @-mentions to people who do not follow you. These posts can be labeled spam and dropped or limited before most people see them.
Self-engagement
Your own likes, reposts, and copy-link taps on your own posts are ignored for ranking. Coordinated or rapid self-activity is the explicit target of the self-reply filter above.
The published rules cover the mechanical patterns. Broader behavioral signals (reply velocity faster than a human can read, high volume of generic replies only under large accounts, engagement that only appears under verified users) are almost certainly evaluated by models that are not fully open-sourced. The effect is the same: the pattern is treated as farm behavior.
What a real contributor does instead
The systems are looking for mechanical repetition and low-effort volume. Original contribution is the opposite of those patterns.
Write the specific thing
A concrete observation, a real number, a screenshot of something you actually saw, a short breakdown of how something works, or a direct reply that shows you read the post. These do not match the duplicate-text rules and they score higher on reply quality models.
Reply only when you have something to add
One clear sentence that responds to the actual content is better than five generic ones. The quality model scores low-effort replies near zero and buries them. A reply that references a specific part of the post does not look like the template replies the spam filters target.
Do not reply to yourself in a chain
Do not reply to yourself in a chain under someone else’s post. That is the exact pattern the September multi-step filter was written to catch. If you need to clarify, edit or post a new reply that stands alone.
Avoid the phrases that already look like templates
Avoid the phrases and formats that have been used so many times they now look like templates. “This is insane,” “Tag someone who,” “Like if you agree,” “Check the comments,” and near-copies of viral lines are the material the duplicate detectors and quality models were trained on. Write the sentence you would actually say.
Space your posts and replies
Rapid bursts of activity that only appear under large accounts match the behavioral patterns the systems are built to notice. A normal human pace does not.
The code is public. The rules and filters are written to catch repetition, self-reply chains, generic low-quality replies, and bad link patterns. If your activity does not match those patterns, the systems have nothing to label. Write the specific thing you actually think. That is the only reliable way to stay on the human side of the detector.
Blackout! $35. Buy