Semantic clustering groups content by topic similarity, not just keywords. If you publish articles about "marathon training," "5K preparation," and "running nutrition," ILLIXIS recognizes these all relate to running and clusters them together.
This creates precise retargeting audiences. Instead of targeting "everyone who visited my blog," you target "readers interested in marathon training" or "readers who viewed nutrition content."
Every night at 2:30 AM, ILLIXIS runs automatic clustering:
You don't configure this. It happens automatically after publishing.
ILLIXIS uses topic overlap, not word embeddings:
["marathon training", "running shoes", "endurance"]["marathon training", "race preparation", "running gear"]intersection / union = 2/6 = 0.33If similarity > 0.3, content joins the cluster. Otherwise, a new cluster is created.
This is simpler and faster than vector-based clustering, with no ML library dependencies.
Go to Advertising in your main navigation. The dashboard shows all auto-created clusters:
Click any cluster to see:
Use this to verify clustering accuracy. If unrelated content appears together, those articles may need better topic extraction (ensure they have clear themes).
ILLIXIS generates names from the most frequent topics:
Names are auto-generated. You cannot rename clusters manually (they represent semantic themes, not user-defined categories).
Each cluster becomes one audience on each platform:
Topic: [Cluster Name] (e.g., "Topic: Marathon Training")[ILLIXIS] [Cluster Name] (e.g., "[ILLIXIS] Marathon Training")[ILLIXIS] [Cluster Name]Clusters require at least 1 content item to create an audience. Single-article clusters are allowed (useful for high-value content like pillar posts).
Maximum cluster size: 50 articles (to avoid regex size limits in GA4 filter rules).
Content joins a cluster if similarity > 0.3 (30% topic overlap).
Lower threshold = larger clusters (more diverse content). Higher threshold = smaller clusters (more focused content).
This is not configurable. The 0.3 default balances audience size with relevance.
Clusters update automatically when:
Audiences do NOT update when clusters change. GA4 audiences cannot be modified after creation (would require archive + recreate). Meta/LinkedIn audiences can update, but batch updates are expensive.
For now, clusters are flagged for re-sync when content changes, but re-syncing audiences is manual.
ILLIXIS runs several automated processes to keep your clusters and audiences up to date:
| Process | Schedule | Description |
|---------|----------|-------------|
| Cluster Recalculation | Weekly, Thursdays 4:00 AM UTC | Full semantic analysis of all content, recalculates cluster memberships |
| New Content Assignment | Within 24 hours of publishing | Newly published content is added to appropriate clusters |
| Topic Label Regeneration | Monthly | Cluster names and common topics are regenerated based on current content |
| Internal Linking Suggestions | After each cluster refresh | Linking recommendations update when cluster membership changes |
You don't need to trigger any of these manually. They run automatically in the background.
If you publish content and want immediate clustering (without waiting for 2:30 AM):
This is useful after publishing a batch of articles.
Each cluster creates a GA4 filter based on content URLs:
Uses top 10 keywords from common topics:
["marathon training", "running shoes", "race preparation"]page_path_keywords: ["marathon-training", "running-shoes", "race-preparation"]If cluster has few topics, ILLIXIS extracts keywords from content titles:
If no topics or title keywords exist:
page_path_contains: [cluster-slug]page_path_contains: "marathon-training"This is the weakest match (requires exact slug in URL).
GA4 audiences appear in Google Ads automatically if:
ILLIXIS creates the GA4 audience. Google handles the Ads sync.
Note: Display campaigns accept audiences under 1,000 members. Search campaigns do not.
Meta Custom Audiences sync if:
ILLIXIS creates Custom Audiences using the Meta Marketing API. Pixel events populate the audience with matching users.
Meta audience minimum: 100 members before ads can target it.
LinkedIn Matched Audiences sync if:
ILLIXIS creates DMP Segments using the LinkedIn Marketing API. Insight Tag events populate the segment.
LinkedIn audience minimum: 300 members before ads can target it.
Cause: Content has no extracted topics.
Fix: Ensure articles are published and topics are extracted. Topics are auto-generated during content creation or content inventory sync.
Check any article in Content Hub → View Details → Scroll to "Extracted Topics" field.
Cause: GA4/Meta/LinkedIn not connected, or sync failed.
Fix:
Cause: GA4 and Google Ads are not linked in GA4 Admin.
Fix:
ILLIXIS creates the GA4 audience correctly. The GA4-to-Ads link is external.
Cause: Meta Pixel not installed, or no visitors match the filter.
Fix:
Cause: Missing ad account, expired token, or insufficient permissions.
Fix:
rw_ads, r_ads_reporting, rw_dmp_segmentsClustering performance scales with content volume:
Nightly batch job has a 5-minute timeout. If you have 1,000+ articles, clustering may timeout. Contact support to increase the limit.
Advertising dashboard shows:
Use this to identify:
Content must have clear topics for accurate clustering. If ILLIXIS generates generic topics like "content," "marketing," or "business," the clustering will be weak.
Good topics:
Bad topics:
Fix this by improving your content's theme clarity (clear headings, focused topics, specific keywords).
Check clusters monthly:
You cannot manually move content between clusters. The algorithm must recognize semantic similarity.
Don't just retarget randomly. Plan campaigns by cluster:
Match ad creative to cluster themes for higher CTR.
Don't over-analyze individual clustering decisions. The algorithm optimizes for:
If a few articles seem misclustered, that's okay. Audience targeting is probabilistic. A 95% accuracy rate is sufficient.
Q: Can I manually create clusters? A: No. Clusters are auto-generated based on topic similarity. Manual categorization defeats the purpose of semantic analysis.
Q: Can I rename clusters? A: No. Cluster names are generated from dominant topics. Renaming would disconnect the name from the content.
Q: Can I merge clusters? A: No. Merging is unnecessary. If two clusters should be one, improve topic extraction on those articles so the algorithm recognizes similarity.
Q: What happens to old content? A: Content published before clustering was enabled gets processed on the next nightly run. It's not excluded.
Q: Does this work with draft content? A: No. Only published content is clustered. Drafts are ignored.
Q: Can I exclude content from clustering? A: Not currently. All published content with topics gets clustered. If you don't want content in retargeting audiences, don't publish it.
Q: Does this affect my site's performance? A: No. Clustering runs as a background task. Your site is unaffected.
Q: What if I have 10,000 articles? A: Clustering is O(n*m) where n = unprocessed content, m = existing clusters. For large catalogs (1,000+ articles), contact support to optimize the batch job.
Q: Can I see clustering logs? A: Yes, if you have admin access. Each cluster creation and update is logged with similarity scores.
Q: Does this replace Google Analytics audiences? A: No. This creates GA4 audiences automatically. You can still create manual audiences in GA4 for specific use cases.
Q: What if I delete a cluster? A: Don't delete clusters manually. They'll be recreated on the next nightly run (content still exists). Instead, delete or unpublish the content.
Q: How often do audiences update? A: Clusters update nightly. GA4/Meta/LinkedIn audiences sync when cluster content changes. Audience membership updates happen in each platform (based on Pixel/Tag events).
If clustering isn't working:
Last Updated: February 5, 2026
What This Does
Content Arbitrage Finder discovers "easy win" keywords where competition is weak but search volume and commercial value are strong. These are opportunities where you can rank quickly because competitors are vulnerable.
What This Does
Rising Search Detection identifies keywords with accelerating search volume before they peak. This first-mover advantage lets you create content while interest is building, not after competitors have already captured the traffic.
One platform. You approve. ILLIXIS executes. Marketing that just happens.
Marketing, Unstacked.