Algorithmic Slop Throttling (Posting for the X Algorithm)

The open-sourced X algorithm ranks posts with a Grok transformer and runs a VLM that assigns a slop score to low-effort / AI-generated-looking content; high-slop posts get throttled downstream. The practical rule: posts that read like raw LLM output lose reach, so run anything before publishing through Security and Review Skills. Source: xai-org/x-algorithm grox/classifiers/content/banger_initial_screen.py, 2026-05-15

This is the actionable, posting-side reading of X For You Algorithm (2026 Open Source). Keep it in mind whenever drafting or scheduling X content.

What the algorithm actually optimizes

1. Weighted sum over predicted actions. Rank ≈ Σ (weight × P(action)). Positive weights for likes, replies, reposts, shares, dwell time, profile clicks, follows; negative weights for not_interested, block, mute, report. A post that provokes negative signals is pushed down, not just left unboosted. Source: xai-org/x-algorithm README § Scoring, 2026-05-15
2. Slop / banger screening. A Grok VLM evaluates text formatting and vocabulary and assigns a slop_score; a separate "banger" screen rewards high-quality posts. Classic LLM syntax (signposting, em-dash overuse, rule-of-three, "it's not X, it's Y", boldface lists, generic positive conclusions) reads as slop and throttles reach. Source: xai-org/x-algorithm grox/classifiers/content/banger_initial_screen.py + task_pub.py, 2026-05-15
3. No follower-count crutch. Hand-engineered features (follower count, account age) were removed; the model ranks on predicted engagement from your interaction history. Reach is earned per-post, not granted by audience size. Source: xai-org/x-algorithm README, 2026-05-15

Posting checklist (apply when xposting)

• De-slop the copy. Run final text through Security and Review Skills (no em dashes, no negative parallelism, no "increasingly", no signposting, no generic conclusions). This maps directly onto what the slop_score VLM penalizes.
• Optimize for replies + dwell, not just likes. Replies, reposts, quotes, long dwell, and profile clicks carry weight. Hooks, specificity, and a reason to stop scrolling beat broad bland takes.
• Avoid block/mute/report bait. Rage-bait and spammy patterns are scored negatively and demote the post.
• Lead with the artifact. Concrete proof (screenshot, diagram, code, video) reads as a "banger", not slop — consistent with the Frontend and Design Skills artifact-led approach and the Content Backlog "needs artifact" rows.
• Match Cloud, Data, and Service Skills / Security and Review Skills. Short declarative lines, first person, concrete over abstract — the voice profile already trends anti-slop.

Why this is durable knowledge

Detection runs on a near-deterministic VLM judge (temperature = 0.000001) with conditional reasoning for hard cases, so it is not a brittle keyword filter — surface-level tricks won't fool it; genuinely human, specific, artifact-backed writing is the only reliable counter. The same humanizer pass Kevin already runs for AI-detection now doubles as reach optimization. Source: xai-org/x-algorithm grox/classifiers/content/safety_ptos.py, 2026-05-15

Concept Position

---

Timeline

• 2026-07-01 | Concepts category refresh added this page to the Product, business, distribution, and positioning family, linked it to Concept System Map, and kept it standalone because it owns this reusable mental model: The open-sourced X algorithm ranks posts with a Grok transformer and runs a VLM that assigns a slop score to low-effort / AI-generated-... Source: User request, 2026-07-01
• 2026-06-17 | Concept captured from the open-sourced X algorithm (Kevin: "keep in mind when xposting"). Ties the slop_score / banger VLM screen and weighted-action ranking to a concrete posting checklist; reinforces the existing Security and Review Skills gate as reach optimization, not just AI-tell removal. Source: User, 2026-06-17; xai-org/x-algorithm, 2026-05-15