x-content-optimizer

# X Algorithm Optimization Reference

Based on X's open-source recommendation system (Phoenix + Home Mixer).

## Scoring System Overview

X uses a Grok-based transformer (Phoenix) to predict engagement:

```
Final Score = Σ (weight × P(action)) + offset_adjustment
```

Content ranked by **predicted engagement**, not chronological order.

---

## Positive Engagement Signals

| Signal | What It Measures | Optimization |
|--------|------------------|--------------|
| **Favorite** | P(like) | Emotional resonance, valuable insight |
| **Reply** | P(reply) | Questions, discussion hooks |
| **Retweet** | P(repost) | Share-worthy, quotable content |
| **Quote** | P(quote tweet) | Hot takes worth commenting on |
| **Share** | P(share) | "Send to someone" content |
| **Share via DM** | P(DM share) | Personal relevance |
| **Share via Copy Link** | P(copy link) | Reference-worthy |
| **Click** | P(expand tweet) | Curiosity-inducing hooks |
| **Profile Click** | P(visit profile) | Expertise signals |
| **Photo Expand** | P(expand image) | Intriguing visuals |
| **Video Quality View** | P(quality view)* | Engaging video content |
| **Dwell** | P(dwell on tweet) | Engaging content |
| **Dwell Time** | Predicted time spent | Depth of content |
| **Follow Author** | P(follow) | Consistent value |

*VQV only for videos exceeding minimum duration

---

## Negative Signals (Penalties)

| Signal | Trigger | Impact |
|--------|---------|--------|
| **Not Interested** | Irrelevant content | Negative weight |
| **Block Author** | Harassment/spam | Severe penalty |
| **Mute Author** | Annoying/excessive | Moderate penalty |
| **Report** | Policy violations | Severe penalty |

---

## Learned Penalties (Indirect)

These aren't explicit in code but learned by the ML model:

| Factor | Why It Hurts | Mechanism |
|--------|--------------|-----------|
| **External Links** | Takes users OFF platform | Lower dwell_time, fewer subsequent engagements |
| **Link-only tweets** | No native content value | Low dwell, users leave immediately |
| **Clickbait without payoff** | Users bounce quickly | Low dwell_time, possible "not interested" |

**External Links Impact:**
- No `ClientExternalLinkClick` as positive signal
- Users clicking external links = leaving X = lower dwell
- Model learns: external links → lower engagement prediction
- **Mitigation**: Provide value IN the tweet, link as supplement

---

## Special Scoring Factors

### Author Diversity Decay
Repeated posts from same author get progressively reduced scores.

### In-Network Boost
Posts from followed accounts > out-of-network discovery.

### Video Quality View
Only for videos exceeding minimum duration threshold.

---

## Single Tweet Optimization

### Structure
```
[Hook] - First 50 chars visible in preview
[Core Value] - Main insight delivered IN tweet
[Engagement Driver] - Question or CTA
```

### Priority Signals
1. **Favorite/Like** - Emotional or valuable
2. **Reply** - Discussion-worthy
3. **Retweet** - Share-worthy
4. **Dwell** - Holds attention

### External Link Strategy
- Deliver core value IN the tweet
- Link as optional "learn more"
- Never make link the only value
- Consider: screenshot + context > raw link

---

## X Article Optimization

### Why Articles Win Over Links
- Content stays ON platform (dwell time)
- Native format = algorithm friendly
- Full engagement tracking possible

### Structure
```
[Title] - Curiosity + value promise
[Preview] - Expands on value
[Body] - Structured, valuable sections
[Closing] - Summary + CTA
```

---

## Thread Optimization

### Position Strategy
| Position | Focus | Signal Target |
|----------|-------|---------------|
| Tweet 1 | Hook | Click, Dwell |
| Tweet 2-3 | Engage | Reply |
| Middle | Value | Dwell Time |
| Final | CTA | Repost, Follow |

---

## Audit Checklist

### Positive Signal Check
- [ ] Like potential: Emotionally resonant?
- [ ] Reply potential: Discussion-worthy?
- [ ] Repost potential: Share-worthy?
- [ ] Dwell potential: Holds attention?

### Negative Signal Risk
- [ ] Not Interested: Relevant to audience?
- [ ] Mute risk: Frequency reasonable?
- [ ] Block risk: Respectful content?
- [ ] Report risk: Policy compliant?

### External Link Check
- [ ] Core value delivered IN tweet?
- [ ] Link supplementary, not essential?
- [ ] Consider native alternative (article, screenshot)?
- [ ] If link required: context + value before link?

### Format Check
- [ ] Content complete and polished
- [ ] Key information emphasized
- [ ] Readable formatting
- [ ] No unnecessary padding

---
name: x-content-optimizer
description: Audit and optimize tweets, X articles, and threads for X's recommendation algorithm. Use when user wants to review content before posting, improve engagement potential, or get algorithm-friendly suggestions. Triggers on /x-content-optimizer or requests to "review tweet", "optimize for X algorithm", "audit my post", or "improve engagement".
---

# X Optimizer

Audit content against X's recommendation algorithm (Phoenix/Grok-based) and provide optimization suggestions.

## Workflow

```
Input Content → Algorithm Audit → Issue Report → Fact Check → Suggested Revision → User Approval → Final Output
```

## Step 1: Receive Content

Accept content in any format:
- Direct text paste
- Markdown file
- Multiple tweets (thread)

Identify content type:
- **Single Tweet**: Under 280 chars, one post
- **X Article**: Long-form native content
- **Thread**: Multiple connected tweets

## Step 2: Algorithm Audit

Reference: [algorithm-rules.md](references/algorithm-rules.md)

### Audit Categories

**A. Positive Signal Potential**
| Signal | Check |
|--------|-------|
| Favorite | Emotional resonance or valuable insight? |
| Reply | Discussion hook or question? |
| Repost | Share-worthy content? |
| Dwell | Depth to hold attention? |
| Click | Curiosity-inducing hook? |

**B. Negative Signal Risk**
| Signal | Check |
|--------|-------|
| Not Interested | Relevant to target audience? |
| Mute/Block | Respectful, not spammy? |
| Report | Policy compliant? |

**C. Learned Penalties**
| Factor | Check |
|--------|-------|
| External Links | Core value IN content, not behind link? |
| Clickbait | Payoff matches promise? |

**D. Format & Structure**
- Hook strength (first line/50 chars)
- Content completeness
- Engagement driver (CTA/question)
- Readability (line breaks, emphasis)

## Step 3: Generate Audit Report

Format:
```
## Audit Report

### Score Summary
- Positive Signal Potential: [HIGH/MEDIUM/LOW]
- Negative Signal Risk: [HIGH/MEDIUM/LOW]
- Overall Algorithm Fit: [EXCELLENT/GOOD/NEEDS WORK/POOR]

### Issues Found
1. [Issue]: [Explanation]
   - Impact: [Which signal affected]
   - Fix: [Specific suggestion]

2. [Issue]: [Explanation]
   ...

### Strengths
- [What works well]
```

## Step 4: Fact Check

**CRITICAL**: Before adding or suggesting any factual information (numbers, dates, statistics, claims about products/companies), you MUST:

1. **Identify factual claims** in the content that need verification
2. **Search and verify** using WebSearch tool
3. **Only include verified facts** in the optimized version
4. **Remove or flag unverifiable claims** - never fabricate data

Examples requiring verification:
- Product launch dates ("X was released in...")
- Statistics ("X% of users...")
- Company announcements ("Company just launched...")
- Performance metrics ("reduces time by X%")

If no verifiable source exists, either:
- Remove the claim entirely
- Replace with qualitative language ("significantly improves" instead of fake percentages)
- Ask user if they have a source

## Step 5: Generate Optimized Version

Provide complete rewritten version addressing all issues.

Format:
```
## Optimized Version

[Full rewritten content]

### Changes Made
1. [Change]: [Why it improves algorithm score]
2. [Change]: [Why it improves algorithm score]
```

## Step 6: Offer One-Click Apply

If content is from a file:
```
Would you like me to apply these changes to the file?
```

If direct text:
```
Here's your optimized content ready to copy.
```

## Content-Specific Guidelines

### Single Tweet
- Hook in first 50 chars (preview visibility)
- Core value delivered completely
- End with engagement driver
- External links: provide context first

### X Article
- Title: curiosity + specific value
- Opening: validate the click
- Body: structured sections
- Closing: summary + CTA
- Advantage: keeps users on platform (dwell time)

### Thread
- Tweet 1: Standalone hook (must work alone)
- Tweet 2-3: Engagement point (question)
- Middle: Value delivery
- Final: CTA (repost, follow)
- Each tweet: Complete and independent

## Quick Reference

### High-Impact Improvements
1. Add question/discussion hook → ↑ Reply
2. Add quotable statement → ↑ Repost
3. Strengthen opening hook → ↑ Click, Dwell
4. Remove/contextualize external links → ↑ Dwell Time
5. Add specific data/examples → ↑ Credibility, Share

### Common Issues
| Issue | Impact | Fix |
|-------|--------|-----|
| Weak hook | Low click/dwell | Rewrite first line |
| Link-only value | User leaves platform | Add value IN content |
| No engagement driver | Low reply | Add question/CTA |
| Wall of text | Low dwell | Add line breaks |
| Off-topic | "Not interested" signal | Clarify audience fit |