Key Changes to the X Algorithm (July–September) 1️⃣ July: Enhanced Mutual Follow Interaction Increased the weight for replies to original posts between mutually followed accounts from 5.0 to 20.0 (later finalized at 15.0 to prevent feed spamming). Began prioritizing interactions between mutually followed accounts as a high-quality signal. 2️⃣ August: Disclosure of Behavioral Prediction Weights and Revised Feed Quality Standards Weights: Copy and share (20.0), mutual follow replies (20.0), general replies/quotes/DMs (5.0) Follows via post (4.0), reposts (1.0), likes (0.5) Negative weights: Reports (-234.0), mute (-58.8), “Not interested” (-43.2), block (-31.2) Implemented semantic ID-based topic diversity, applied mute word filtering to the Following tab, and transitioned to the Phoenix engine centered on reply spam thresholds and dwell time. 3️⃣ September: Combatting Comment Farming and Abuse Introduced a standalone reply spam LLM classifier based on Gemma (precise text and image analysis). Added direct user reporting for bots and enforced a minimum 10-second click dwell time measurement. Launched “Under the Hood” to transparently disclose legal and jurisdiction-specific account enforcement records.