4 min read
Meta Platforms has revealed fresh details about its advanced AI powered ad targeting architecture, spotlighting the role of its new foundation model GEM, short for Generative Ads Recommendation Model, in delivering improved results for advertisers across Facebook, Instagram, and its broader family of apps. The company says this model and its supporting infrastructure represent a significant evolution in how ads are targeted.
Meta explained it as:
The Generative Ads Recommendation Model (GEM) is Meta’s most advanced ads foundation model, built on an LLM-inspired paradigm and trained across thousands of GPUs. It is the largest foundation model for recommendation systems (RecSys) in the industry, trained at the scale of large language models.
What Meta is Actually Doing
Meta explains that GEM forms part of a broader ad technology ecosystem that also includes models named Lattice and Andromeda. GEM is described as the largest recommendation foundation model ever used in advertising, trained at the scale of large language models using thousands of GPUs.
According to Meta, GEM enables:
- Four times greater efficiency in driving advertiser performance gains with the same amount of computing power and data.
- Twice the effectiveness in knowledge transfer, allowing improvements from one ad model or context to be applied across others more rapidly.
- Smarter cross feature learning that combines both sequence features such as user activity history and non sequence features like ad attributes, location, and creative format to deliver more relevant ad placement and audience reach.
In further detail, Meta’s Andromeda model focuses on ad retrieval improvements. The company reports that Andromeda significantly increased recall and ad quality on selected segments by redesigning its retrieval architecture, co developing it with hardware accelerators, and optimizing the hierarchical indexing of ad candidates.
The Lattice model, meanwhile, is described as a broad ranking system that learns across ad objectives and transfers insights between campaigns. Industry analysts note that Lattice has become central to Meta’s advertising strategy, especially in data limited environments where transfer learning provides major efficiency gains.
Why Advertisers Should Care
For advertisers, the implications are significant. Meta claims its automated targeting systems can now surface highly relevant audiences without requiring extensive manual setup. With GEM and its supporting AI stack, brands may benefit from:
- Reduced need for manual audience segmentation and refinement
- Improved reach among users who were previously overlooked
- Higher return on ad spend through more accurate and efficient targeting
This development may mark a turning point in automated advertising across major social platforms, particularly as privacy regulations and data restrictions push marketers to depend more on algorithmic intelligence than manual targeting.
Meta also reports that advertisers who use its Advantage Plus creative tools alongside these AI models have seen meaningful increases in both return on ad spend and conversion rates.
Context and Future Moves
Meta’s announcement forms part of a broader transformation of its advertising ecosystem. The company aims to achieve fully automated AI ad creation and targeting by late 2026. Under this system, advertisers will simply provide a product image and budget, while Meta’s AI will automatically generate the creative, select the target audience, and allocate the budget.
Starting December 16, 2025, Meta will also incorporate user interactions with its Meta AI chatbot into ad targeting and content recommendations across Facebook and Instagram. This update reflects how deeply AI is now embedded in Meta’s advertising framework and how conversational data will fuel smarter ad personalization in the future.
However, the increasing sophistication of these systems has also drawn scrutiny. Researchers have warned that AI driven ad targeting models can sometimes reproduce demographic or socioeconomic biases and may lack transparency in explaining how ads are selected or displayed. The findings have reignited calls for clearer oversight and accountability in automated ad delivery.
