# X Algorithm Skill > Write viral X posts backed by the actual recommendation algorithm source code. No generic social media advice—every recommendation traces to real algorithm behavior. ## Scoring Signals The Phoenix ranking model predicts 19 engagement probabilities, multiplied by weights and summed into a final score. ### High-Value Positive Signals (maximize these) **Reply** - Probability user will reply. Ask genuine questions, make claims that invite pushback, share incomplete thoughts others can build on. **DM Share** - Probability user shares via Direct Message. Share insider knowledge, actionable frameworks, niche-specific content experts share with peers. **Copy Link Share** - Probability user copies link to share externally. Create reference-worthy guides, lists, frameworks. Evergreen content people bookmark. **Dwell Time** - How long user spends viewing. Write longer substantive posts, use formatting that encourages reading, open with intrigue. **Follow Author** - Probability user follows you. Demonstrate unique expertise, promise ongoing value, show personality worth following. **Quote** - Probability user quote-tweets. Make claims worth commenting on, share takes people want to amplify with their own spin. ### Medium Positive Signals - Like/Favorite - Basic approval - Retweet - Pure amplification - Click - Link/media engagement - Profile Click - Author interest - Photo Expand - Image engagement - Video Quality View - Video completion ### Negative Signals (avoid at all costs) - Not Interested - Score penalty. Avoid spam, engagement bait, off-topic content. - Mute Author - Severe penalty. Don't over-post (5-10/day max), stay in your lane. - Block Author - Severe penalty. Don't be hostile or aggressive. - Report - Most severe penalty. Follow platform rules. ## Discovery & Retrieval Two-tower neural network matches post embeddings to user interest embeddings based on engagement history. ### Out-of-Network Penalty Posts from accounts the user doesn't follow get a penalty multiplier. Your post must score significantly higher to overcome this. ### Author Diversity Penalty Multiple posts from same author get exponentially discounted. Don't flood the timeline. ### Age Filter Posts older than threshold are filtered out entirely. Fresh content gets priority. ## Content Structure ### Hook Types (first line is 80% of success) - Question Hook - Opens with genuine question reader wants answered - Contrarian Hook - Challenges conventional wisdom immediately - Insight Hook - Leads with non-obvious observation - Story Hook - Opens mid-action to pull readers in - Resource Hook - Promises immediate value ### Body Structure **Short posts (<280 chars)**: Single punch. Hook + Payoff. No preamble. **Long posts (280-1000 chars)**: HSP structure. Hook (1 line) + Support (2-4 points) + Punch (closing line). **Threads**: Each post must stand alone AND continue the story. Label explicitly: "Thread:" or "1/" ### Formatting for Dwell Time - Every sentence gets its own line - Short lines create rhythm - Lists work - Questions break monotony ## Viral Patterns ### The Insight Thread Non-obvious observation + varied transition (not always "Here's what I learned:") + 2-7 numbered insights (vary the count) + natural closer ### The Contrarian Take Bold claim challenging consensus + brief explanation + optional invitation to disagree or personal note ### The Resource Post Clear value promise (skip "ultimate/comprehensive") + practical scannable content + natural closer with personal touch ### The Curiosity Gap Create specific curiosity + brief delay (don't milk it) + deliver satisfying non-obvious answer ### The Personal Story Start mid-action + quick context + the turn + optional lesson (don't force one if there isn't one) ### The Effort Summary What you did (specific) + what you found (varied count, not always 3) + casual closer ## Anti-Patterns to Avoid ### Engagement Farmer "Like + RT + Follow!" triggers Not Interested signals. ### Reply Guy Self-promotional replies on every viral post trigger Mute. ### Thread Maximalist Every thought as 20-tweet thread hits author diversity penalty. ### Controversy Farmer Rage bait gets short-term replies but long-term blocks/mutes. ### Copy-Paste Recycler Same content reposted hits deduplication filters. ## Writing Like a Human, Not a Bot Algorithm optimization means nothing if your posts scream "AI wrote this." The algorithm favors genuine engagement, and users can smell inauthenticity. ### AI Vocabulary Blacklist - NEVER use these words **Corporate buzzwords**: leverage, utilize, optimize, synergy, stakeholders, robust, comprehensive, streamlined, scalable, cutting-edge, best-in-class, world-class **AI verbal tics**: Additionally, Furthermore, Moreover, crucial, essential, vital, key (as adjectives), delve, dive into, unpack, landscape, ecosystem, space (for industries), journey, adventure (for experiences), tapestry, myriad, plethora, multifaceted, nuanced, paradigm, paradigm shift, game-changing, groundbreaking, transformative, revolutionary, unlock, unleash, empower, pivotal, pivotal moment, broader implications, overarching, "It's worth noting", "It's important to note", "at the end of the day" **Sycophantic openers** (never start with): "Great question!", "Absolutely!", "That's a fantastic point!", "I love this!" **Generic positive closers** (never end with): "Success awaits!", "The possibilities are endless!", "I hope this helps!", "Let me know if you have any questions!" ### Structural Tells to Avoid **Rule of Three Trap**: AI loves exactly 3 things. Real humans sometimes have 2 things. Or 4. Or 7. Vary your list lengths. **Negative Parallelism**: "Not just X, but Y" and "It's not about X, it's about Y" are AI favorites. Use sparingly. **Significance Inflation**: "Pivotal moments." "Broader implications." "Fundamental shifts." Real insights don't need inflation. **Em Dash Overuse**: Em dashes are fine occasionally. AI uses them constantly. If you have more than one em dash per post, cut some. **Formulaic Transitions**: "Here's the thing:", "Here's why:", "Here's what I learned:" — these work but are overused. Vary your transitions or skip them entirely. **The Perfect Structure**: AI posts have suspiciously clean structure. Real posts meander sometimes. That's okay. ### Adding Soul to Your Writing **Express Genuine Opinions**: AI hedges. Humans commit. Say "Option B is wrong. I've tried it." not "There's something to be said for both approaches..." **Use Contractions**: "Don't" not "Do not". "It's" not "It is". "Can't" not "Cannot". **Vary Rhythm Unpredictably**: Short sentence. Then maybe a longer one that goes on for a bit. Then short again. Fragment. Works. **Include Specific Details**: "Watched three companies blow $2M+ on this exact mistake last year" not "I've seen many companies struggle with this challenge" **Allow Imperfection**: Real posts have tangents that don't quite connect, incomplete thoughts, opinions stated too strongly, casual asides. AI never includes these. **Write Like You Talk**: Read your post out loud. Would you say this to a friend at a bar? If not, rewrite it until you would. ### Humanization Checklist - [ ] Zero words from the AI blacklist? - [ ] Varied list lengths (not always 3)? - [ ] No more than one em dash? - [ ] Using contractions naturally? - [ ] At least one specific detail? - [ ] Would you actually say this out loud? - [ ] Something slightly imperfect or casual? - [ ] No sycophantic opener or generic closer? ## Pre-Post Checklist (6-Point Audit) - [ ] Reply: Contains question or debate-worthy claim? - [ ] Share: Reference-worthy value someone would send to specific person? - [ ] Dwell: Hook creates curiosity, formatting encourages reading? - [ ] Follow: Demonstrates unique expertise? - [ ] Negative: Not asking for engagement, not hostile, on-brand? - [ ] Humanization: No AI vocabulary, varied structure, sounds like a real person? ## Signal Hierarchy **Tier 1 (Highest)**: DM Share, Copy Link Share, Reply, Follow **Tier 2 (High)**: Quote, Dwell Time, Video Quality View **Tier 3 (Medium)**: Retweet, Like, Profile Click **Tier 4 (Lower)**: Click, Photo Expand **Tier 5 (Negative)**: Not Interested < Mute < Block < Report ## Key Insight The algorithm doesn't care about "engagement" generically. It cares about specific actions: replies, shares, dwell time. A post that gets 100 likes but no replies scores worse than a post with 50 likes and 20 replies. Write for the high-value signals: replies, shares, follows. The rest follows. And make it sound human. The best algorithm optimization in the world won't help if your post reads like AI slop.