# X Algorithm Playbook (Phoenix Update -- 2026) ## How the Algorithm Ranks Posts The Phoenix algorithm scores 1,500 candidate posts per user per feed refresh. It predicts 19 interaction types and weights them as follows: | Signal | Weight | What It Means | |--------|--------|---------------| | **Author-Engaged Reply** | Exponential multiplier | The author replies back to someone who replied -> triggers cascading reach | | **Bookmark** | High | User saves it -> "Knowledge Node" signal, high utility content | | **Thread Expansion** | High | User clicks to read the full thread -> session depth | | **Substantive Reply** | Moderate-High | A reply with real content (not just an emoji) -> conversation signal | | **Repost (Retweet)** | Moderate | Shares to their followers -> out-of-network discovery | | **Dwell Time (>3 seconds)** | Moderate | User paused scrolling -> "scroll-stopper" worked | | **Like** | Baseline (1x) | Necessary but not sufficient for reach | ### Key Takeaway A post with 5 back-and-forth replies in the comments will outperform a post with 50 likes and zero conversation. **Replies > Likes. Always.** ### CLI Scoring Implementation When using `twitter-cli`, you can mimic the Phoenix algorithm to find engagement opportunities: - Use `twitter search "topic" --filter --json` to sort results algorithmically. - The default CLI scoring formula heavily weights Replies and Bookmarks. - **Tip:** Use `--compact` (`-c`) along with `--json` when fetching large lists to save context tokens while still receiving engagement metrics. --- ## The 30-Minute Golden Window The algorithm makes its primary distribution decision within the first 30 minutes of a post going live. **What to do:** 1. Post the content 2. Stay online for the next 30 minutes 3. Reply to EVERY comment immediately -- this triggers "Author-Engaged Reply" signals 4. Ask follow-up questions in your replies to keep threads going 5. The algorithm sees sustained conversation -> pushes post to "out-of-network" users **What NOT to do:** - Post and close the app - Post and check back 3 hours later - Post and only like the comments without replying --- ## Vector Consistency The algorithm maps every account into a multi-dimensional vector space using "Dynamic Embeddings." This means: - Every post pushes your account vector in a direction - Consistent topics = strong, clear vector = algorithm knows who to show you to - Random, off-topic posts = confused vector = algorithm stops distributing you **For your account:** Every post must reinforce your niche positioning as defined in your user-profile.md. Polluting the vector (posting about unrelated topics) will cause algorithmic isolation. --- ## Link Suppression External links cause a **30-50% reach penalty**. The algorithm wants users to stay on X. **Workaround:** 1. Post the content natively (text + video/image) 2. Build engagement in the first 30-60 minutes 3. Drop the link in the first REPLY (not the main post) 4. The reply's reach is smaller, but followers who engaged will see it --- ## No-Premium Penalties & Workarounds Without X Premium, there's a documented 2-4x visibility penalty. Workarounds: ### 1. First Reply Speed Even without Premium, being the first high-quality reply on a trending post maintains visibility. Set notifications for Tier 1 accounts. ### 2. Media Dwell Time Video and image posts get outsized dwell time signals. If your work naturally produces visual content, this is a structural advantage. ### 3. Zombie Post Tactic 48 hours after posting, reply to the top comment on your best-performing post with additional value. This triggers a second distribution wave. ### 4. Cross-Platform Amplification Post every piece of content on X, Threads, and LinkedIn simultaneously. External engagement signals can indirectly boost X performance through increased profile visits. ### 5. Get Premium When Possible The math: Premium costs ~$8/month. If your content gets 3x the reach, you need 1 client/sale to pay for years of Premium. Once you hit the monetization threshold (500 followers + 5M impressions in 3 months), Premium pays for itself. --- ## Session Depth Signals The algorithm rewards content that keeps users in the app: - **Native video > External links** (users watch on X vs leaving) - **Threads > Single tweets** (users expand to read more) - **Conversations > Broadcasts** (author replies create session depth) - **Long-form text > Short quips** (if Premium, use extended tweets) --- ## Timing Heuristics Research-based best posting times for various niches: | Day | Best Time (UTC) | Why | |-----|----------------|-----| | Weekdays | 13:00-15:00 | US morning + Europe afternoon overlap | | Weekdays | 17:00-19:00 | US afternoon, peak browsing | | Saturday | 14:00-16:00 | Weekend browsing, less competition | | Sunday | 16:00-18:00 | Sunday evening engagement spike | **Note:** These are starting heuristics. Track actual performance in memory.md and adjust based on what works for your specific audience. Convert these to your local timezone using the timezone in user-profile.md. --- ## Negative Signals to Avoid | Signal | Impact | How to Avoid | |--------|--------|-------------| | High block/mute rate | Candidate Isolation (shadowban) | Never rage-bait or be controversial outside niche | | Fast unfollows after post | Negative quality signal | Stay on-topic, maintain vector | | Link-only posts | 30-50% reach penalty | Put links in replies, not main post | | Low reply rate from author | Missed Author-Engaged multiplier | Always reply to comments for 30 min after posting | | Generic replies on others' posts | Zero algorithmic value | Write substantive, multi-sentence replies | --- ## Growth Phase Strategy | Phase | Focus | Daily Engagement | |-------|-------|--------------------| | **0-500 followers** | 80% replying, 20% posting | 20-50 high-value replies/day | | **500-2,000 followers** | 60% replying, 40% posting | 15-30 replies + 1-2 posts/day | | **2,000-10,000 followers** | 40% replying, 60% posting | 10-20 replies + 2-3 posts/day | | **10,000+ followers** | 20% replying, 80% posting | 5-10 replies + daily posting | Check your current phase in memory.md and adjust your effort allocation accordingly.