How ILLIXIS Learns What Social Content Works: Preference Learning for Social

What Gets Tracked

Every social post published through ILLIXIS generates engagement data that feeds the preference learning system. The platform captures four core metrics plus contextual features.

Engagement Metrics Synced from Social Platforms

Primary metrics:

  • Likes — Total reactions/likes
  • Comments — Conversation volume
  • Shares — Retweets, shares, reposts
  • Impressions — How many people saw the post

Sync schedule: Engagement data syncs from the social publishing integration every 4 hours for published posts. New posts may take 4-6 hours to show metrics.

Contextual Features Captured

When engagement data syncs, ILLIXIS records these features for learning:

  1. Platform — Twitter/X, LinkedIn, Instagram, Facebook, TikTok, etc.
  2. Post type — Announcement, feature highlight, social proof, behind-the-scenes, urgency, educational, storytelling, promotional, question, custom
  3. Media presence — Has image/video or text-only
  4. Link presence — Links to article or no link
  5. Hashtag count — Number of hashtags used
  6. Text length — Character count of caption

Why these features matter: If LinkedIn posts with 2-3 hashtags consistently outperform posts with 10+ hashtags, the system learns your audience prefers minimal hashtags. If Instagram posts with images get 4x engagement vs text-only, it learns to recommend visual content.

How Engagement Converts to Preference Signals

After syncing engagement metrics, ILLIXIS calculates an engagement rate and creates a preference signal for learning.

Engagement Rate Formula

Engagement Rate = (Likes + Comments + Shares) / Impressions × 100

Example:

  • Post has 45 likes, 12 comments, 8 shares = 65 total engagements
  • Post impressions: 1,200
  • Engagement rate: (65 / 1,200) × 100 = 5.4%

Signal Outcome Mapping

Based on engagement rate, the system assigns an outcome score:

| Engagement Rate | Outcome Score | Interpretation |
|-----------------|---------------|----------------|
| ≥5% | 1.0 | High engagement — positive signal |
| 1-5% | 0.5 | Medium engagement — neutral signal |
| <1% | 0.2 | Low engagement — negative signal |

What this means: A post with 6% engagement creates a strong positive signal. The system learns which features (platform, post type, hashtag count, etc.) contributed to that success. A post with 0.5% engagement creates a negative signal, teaching the system to avoid similar combinations.

What the System Learns Over Time

As you publish more social posts, the preference learning system identifies patterns across four dimensions.

1. Template Performance Learning

If you use Campaign Planner templates (e.g., "Product Launch," "Weekly Roundup"), ILLIXIS tracks engagement rates per template and calculates a success score.

Success score formula:

  • Usage score: (Template usage count / 10) × 30 points (max 30 pts)
  • Engagement score: (Avg engagement rate / 5) × 70 points (max 70 pts)
  • Total: Usage score + Engagement score (0-100 scale)

Example:

  • Product Launch Template
  • Used 15 times (15/10 = 1.0 → 30 usage points)
  • Avg engagement rate: 4.2% (4.2/5 = 0.84 → 58.8 engagement points)
  • Success score: 30 + 58.8 = 88.8
  • Weekly Roundup Template
  • Used 6 times (6/10 = 0.6 → 18 usage points)
  • Avg engagement rate: 1.8% (1.8/5 = 0.36 → 25.2 engagement points)
  • Success score: 18 + 25.2 = 43.2

How this impacts recommendations: When creating a new campaign, ILLIXIS ranks templates by success score. The Product Launch template appears first, Weekly Roundup appears lower. Templates with proven engagement rates get priority.

2. Platform Performance Patterns

The system identifies which platforms perform best for your audience.

Pattern detected:

  • LinkedIn posts: Avg engagement rate 6.2% (1,200 posts)
  • Instagram posts: Avg engagement rate 3.1% (800 posts)
  • Twitter/X posts: Avg engagement rate 1.8% (1,500 posts)

Recommendation adjustment: When suggesting social distribution, the system prioritizes LinkedIn. If you're planning 10 posts, it recommends 5 LinkedIn, 3 Instagram, 2 Twitter instead of equal distribution.

3. Post Type Effectiveness

If "educational" posts consistently outperform "promotional" posts, the system learns your audience prefers education over sales pitches.

Example pattern:

  • Educational posts: Avg engagement 5.8%
  • Social proof posts: Avg engagement 4.9%
  • Behind-the-scenes posts: Avg engagement 4.1%
  • Promotional posts: Avg engagement 2.3%

How this affects recommendations: Campaign templates with more educational posts rank higher. The system recommends post type distribution: 50% educational, 25% social proof, 15% behind-the-scenes, 10% promotional.

4. Content Format Preferences

The system learns whether your audience prefers visual content (images/video) or text-only posts.

Pattern detected:

  • Posts with media: Avg engagement 5.2% (1,200 posts)
  • Text-only posts: Avg engagement 2.8% (400 posts)

Recommendation: When generating social captions, ILLIXIS suggests using images or video more frequently. If you create a campaign, it recommends including image generation or quote graphics in the plan.

Checking Social Learning Status

Navigate to Campaigns → Campaign Planner → Templates to see performance data.

Each template displays:

  • Usage count — How many campaigns used this template
  • Avg engagement rate — Average engagement across all campaigns using this template
  • Success score — Weighted score combining usage and engagement (0-100)
  • Campaigns completed — How many campaigns using this template finished
  • Performance updated — Last time metrics were recalculated

Sort by success score to see which templates perform best for your audience.

When Learning Activates

Minimum requirement: 20 published social posts with engagement data

Until you publish 20 posts, templates use default success scores based on usage count only. Once you cross 20 posts, engagement-based scoring activates.

Timeline expectations:

Week 1 (0-20 posts): No engagement-based learning yet. Templates ranked by usage count. Publish consistently to build your dataset.

Weeks 2-3 (20-50 posts): First engagement scoring activates. You'll start seeing templates with 4%+ engagement rates ranked above templates with 2% rates.

Month 2 (50-100 posts): Clear patterns emerge. You'll know which platforms, post types, and templates work for your audience. Recommendations become reliable.

Month 3+ (100+ posts): Strong predictive accuracy. The system anticipates which campaign templates will perform well before you launch them.

How Recommendations Improve Over Time

As preference signals accumulate, ILLIXIS adjusts recommendations in three areas.

1. Template Ranking

When creating a new campaign, the template selection screen sorts by success score. Templates with proven engagement appear first.

Before learning (Week 1):

  • Templates sorted alphabetically or by usage count only
  • No indication which templates drive engagement

After learning (Month 2):

  • Templates sorted by success score
  • High-engagement templates surface first
  • Low-engagement templates ranked lower or hidden

2. Post Type Distribution

Campaign templates include post type guidance (e.g., "Day 0: Announcement, Day 2: Feature highlight"). As the system learns which post types perform, it adjusts these recommendations.

Before learning:

  • Generic distribution: 5 posts spread evenly (announcement, feature, social proof, behind-scenes, urgency)

After learning (if educational posts perform best):

  • Adjusted distribution: 3 educational posts, 1 social proof, 1 feature highlight
  • Templates updated to reflect performance data

3. Platform Prioritization

When Maya recommends a campaign or you create a distribution plan, platform suggestions reflect engagement data.

Before learning:

  • Equal distribution: "Post to LinkedIn, Instagram, Twitter/X"

After learning (if LinkedIn performs 3x better):

  • Prioritized distribution: "Focus on LinkedIn (5 posts), supplement with Instagram (2 posts), minimal Twitter/X (1 post)"

Nightly Metric Updates

Every night at 2:30 AM, ILLIXIS recalculates template performance metrics using the latest engagement data.

What happens during the update:

  1. System fetches all campaigns using each template
  2. Aggregates engagement metrics from published posts
  3. Calculates average engagement rate per template
  4. Computes success score (usage + engagement)
  5. Updates template performance fields
  6. Increments performance_updated_at timestamp

Logs: Check admin logs to verify nightly updates ran successfully. If metrics haven't updated in 24+ hours, the scheduled task may need attention -- contact support.

Improving Recommendation Quality

1. Publish consistently across platforms Don't post only to LinkedIn. Publish to 3-4 platforms so the system learns platform-specific patterns. If you only publish to LinkedIn, you can't learn whether Instagram would perform better.

2. Test different post types Don't use only "announcement" posts. Mix in educational, social proof, behind-the-scenes, and urgency posts. The system needs variety to learn preferences.

3. Use campaign templates Templates aggregate data across multiple campaigns. If you manually schedule posts without templates, the system can't track template-level performance.

4. Let posts age before judging Engagement data needs 4-6 hours to sync. A post published at 9 AM won't show metrics until 1-3 PM. Give it 24 hours before evaluating performance.

5. Review analytics monthly Check Campaigns → Campaign Planner → Templates monthly. Archive or update low-performing templates. Create new templates based on discovered patterns.

What Social Learning Does NOT Do

Does NOT learn:

  • Best posting times (not yet implemented — coming in Phase 39)
  • Hashtag effectiveness per hashtag (tracks count, not which hashtags)
  • Individual caption performance (tracks template + post type, not exact wording)
  • Image vs video performance (tracks "has media" boolean, not media type)

What you must still do manually:

  • Experiment with posting times to find optimal windows
  • Test different hashtags and track which ones drive clicks
  • Review individual captions to identify wording patterns
  • A/B test image formats (carousel vs single image)

Common Questions

Q: Does this replace the general Preference Learning system? A: No. These are separate systems. General Preference Learning (Settings → Preference Learning) learns which briefs and opportunities you approve. Social Preference Learning learns which social content performs best. Both run independently.

Q: Do I need to approve every social post for learning to work? A: No. Learning happens automatically when posts publish and engagement syncs. You don't need to manually approve or rate posts.

Q: What happens if I delete a campaign? A: Engagement signals from published posts persist even if you delete the campaign. Historical data remains for learning. Only the campaign structure gets deleted.

Q: Can I disable social learning for specific templates? A: Not currently. All templates with published posts contribute to learning. If you don't want a template to influence recommendations, archive it.

Q: How do I reset social learning if my strategy changes? A: Delete or archive old templates and create new ones. Template performance metrics are template-specific, so new templates start with a clean slate.

Q: Why don't my new templates show engagement rates? A: New templates require at least one completed campaign with published posts. Engagement rate remains null until the first post syncs data.

Q: Does social learning work with Social Studio (Freestyle mode)? A: Not yet. Social learning currently tracks Campaign Planner posts only. Social Studio posts (quote graphics, video scripts) don't create preference signals. Planned for future phase.

Q: What if I publish posts directly to platforms (not through ILLIXIS)? A: Those posts don't create preference signals. Only posts scheduled/published via ILLIXIS Social Hub generate learning data.

How It Works

Algorithm: Weighted success scoring (usage + engagement) Engagement rate calculation: (Likes + Comments + Shares) / Impressions x 100 Update schedule: Automated task every night at 2:30 AM Sync frequency: Engagement data syncs from the social publishing integration every 4 hours

Related Features

  • Campaign Planner — Uses template performance metrics for recommendations (see Campaign Planner help guide)
  • Social Hub — Publishes posts that generate engagement data (see Social Hub help guide)
  • General Preference Learning — Learns brief/opportunity preferences, separate from social (see Preference Learning help guide)
  • Cross-Channel Analytics — Shows aggregated social performance across platforms (see Cross-Channel Analytics help guide)

Questions? Email support@illixis.io or ask Maya (bottom-right chat icon).

Ready to lose the stack?

One platform. You approve. ILLIXIS executes. Marketing that just happens.

Join the waitlistNo spam, everUnsubscribe anytime
First 20 founding members: 50% off any plan for your first year.

Marketing, Unstacked.