Meta details next-generation Andromeda, GEM and GenRecs ads ranking systems and reported conversion gains
On October 28, 2026 Meta published details of three AI updates to its ads ranking system that apply automatically to all advertisers. The next generation of Andromeda, its ads retrieval engine co-designed with NVIDIA, routes ads to specialized expert models and scores every eligible ad at over one billion ads per second; Meta says a 10x complexity increase in Q3 lifted conversions by 1.37% on Facebook and 0.24% on Instagram Reels. GEM, its generative ads recommendation model, is now the foundation model across Facebook, Instagram, Reels and off-Meta conversions and transfers knowledge to real-time ranking models, helping low-conversion campaigns. GenRecs uses LLM-based semantic tokens to understand ad content and now powers generative retrieval on Facebook Feed, which Meta says raised conversions by nearly 2% in targeted segments. Meta states there is no setting to enable.
Impact
More of each advertiser's ads, creative variations and products are considered for each impression, which rewards broad creative diversity and accurate conversion signals over manual targeting tweaks. New ads and new products are expected by Meta to exit the learning phase faster thanks to semantic understanding of ad content. Campaigns optimizing for deep-funnel events with few conversions benefit from GEM knowledge transfer. Performance figures are Meta's own reported averages and individual results vary.
Recommendations
- Feed the system with diverse creative and complete catalogs rather than a small number of near-identical ads
- Check Conversions API and pixel signal quality, since Meta stresses data quality as the main lever
- Optimize for the business outcome you actually value and avoid overly narrow targeting constraints
- Monitor new-ad ramp-up times and conversion rates over the coming weeks against prior baselines
Source: Meta for Business