AI & Intelligence Features
Leverage machine learning to get smart recommendations, analyze content performance, and run experiments to optimize your posting strategy.
Content Intelligence
The Intelligence page provides insights into how your content performs based on auto-generated tags.
Tag Performance Analytics
See how different content tags perform across your posts. The system tracks:
Average Score
Mean upvote count for posts with this tag. Higher is better.
Click Rate
Average clicks per post. Shows how well content drives traffic.
Sample Size
Number of posts analyzed. More data = more reliable insights.
Confidence
Statistical confidence in the performance data.
Filtering Options
Filter tag performance by:
- Tag Category - Outfit, Pose, Angle, or Setting
- Date Range - Last 7 days, 30 days, 90 days, or all time
- Minimum Posts - Only show tags with sufficient data
Subreddit Recommendations
Get intelligent subreddit suggestions based on your content tags. The system uses Jaccard similarity matching to find the best fit.
How It Works
-
Content Analysis
The system analyzes tags on your selected media. -
Historical Matching
Compares to past successful posts with similar tags. -
Score Calculation
Weighs tag match (40%), performance (40%), and sample size (20%). -
Ranked Results
Returns subreddits sorted by recommendation score.
Tip: Recommendations require at least 30% tag overlap to appear. Auto-tag your media first to get the best recommendations.
Caption Suggestions
The system suggests captions based on high-performing posts with similar content. Use these as inspiration or templates.
- Captions from posts with similar tags
- Sorted by historical performance
- Editable before use
Content Brain
The Content Brain page provides deeper pattern analysis and portfolio insights.
Pattern Clustering
Content is automatically grouped into performance categories:
| Category | Definition | Action |
|---|---|---|
| Winning | >20% above your average | Create more like this |
| Average | Within 20% of average | Experiment to improve |
| Underperforming | >20% below average | Consider pivoting away |
Portfolio Gap Analysis
Identifies unexplored tag combinations that have performed well for similar content. Use this to discover new content opportunities.
Note: Content Brain requires at least 20 posted items with metrics to provide meaningful insights.
A/B Testing Experiments
Run controlled experiments to determine what content performs best. The system handles statistical analysis automatically.
Experiment Types
Caption Test
Compare two different captions for the same image.
Image Test
Compare two different images with the same caption.
Time Test
Compare posting at different times of day.
Running an Experiment
-
Create Experiment
Choose type and configure Variant A and Variant B. -
System Creates Posts
Two scheduled posts are created automatically. -
Posts Go Live
Both variants are posted according to schedule. -
Metrics Collected
Performance data is gathered over time. -
Analysis Complete
Statistical analysis determines the winner.
Statistical Analysis
Experiments use t-tests to determine statistical significance:
- p-value < 0.05 - Result is statistically significant
- Improvement % - How much better the winner performed
- Confidence Level - How certain we are of the result
Important: Allow enough time for both variants to gather sufficient data before drawing conclusions. Premature analysis may be misleading.
LLM-Powered Caption Generation
The ML service integrates with Open WebUI for advanced caption generation:
Features
- Text-Based Generation - Generate captions from subreddit context
- Image-Based Generation - Analyze images with LLaVA vision model
- Caption Enhancement - Improve existing captions
- Style Control - Casual, teasing, promotional, or funny
Caption Styles
| Style | Description | Best For |
|---|---|---|
| Casual | Friendly, conversational tone | General subreddits |
| Teasing | Playful, engaging hooks | Entertainment content |
| Promotional | Call-to-action focused | Traffic driving |
| Funny | Humorous, witty | Meme/humor subreddits |
Performance Predictions
Get real-time predictions when scheduling posts. The system analyzes:
- Similar historical posts
- Subreddit performance patterns
- Tag combination effectiveness
- Time-of-day factors
Prediction Output
| Metric | Description |
|---|---|
| Predicted Upvotes | Expected upvote count with 90% confidence interval |
| Predicted Clicks | Expected tracking link clicks |
| Confidence | Statistical confidence based on sample size and variance |
| Risk Level | Safe (âĨ70%), Moderate (âĨ50%), or Experimental |
Next Steps
Continue to the Analytics Guide to learn about tracking performance and reports.