{"id":21675,"date":"2026-03-12T12:50:25","date_gmt":"2026-03-12T12:50:25","guid":{"rendered":"https:\/\/relevantgroup.media\/?p=21675"},"modified":"2026-03-12T12:50:36","modified_gmt":"2026-03-12T12:50:36","slug":"meta-gem-the-new-ai-model-transforming-digital-advertising","status":"publish","type":"post","link":"https:\/\/relevantgroup.media\/en\/meta-gem-the-new-ai-model-transforming-digital-advertising","title":{"rendered":"Meta GEM: The New AI Model Transforming Digital Advertising"},"content":{"rendered":"\n

Meta has introduced GEM (Generative Ads Recommendation Model)<\/strong>, the most advanced foundational ad recommendation model the company has developed to date.<\/p>\n\n\n\n

This system represents a significant shift in the way Meta selects and delivers adverts to users across platforms such as Facebook and Instagram<\/strong>, introducing an approach inspired by large language models (LLMs)<\/strong> applied to recommendation systems.<\/p>\n\n\n\n

The objective of GEM is clear: to improve advertising performance and increase return on investment (ROI) for advertisers through a deeper understanding of user behaviour.<\/strong><\/p>\n\n\n\n

\n\n\n\n

Improvements in Campaign Performance<\/h2>\n\n\n\n

Since its launch, Meta has observed notable improvements in campaign performance thanks to this model.<\/p>\n\n\n\n

Specifically:<\/p>\n\n\n\n

    \n
  • Conversions on Instagram have increased by 5%<\/strong><\/li>\n\n\n\n
  • Conversions in the Facebook Feed have grown by 3%<\/strong><\/li>\n<\/ul>\n\n\n\n

    In addition, GEM\u2019s architecture is four times more efficient<\/strong>, driving performance improvements compared with the previous ranking models used by the platform.<\/p>\n\n\n\n

    This increase in efficiency enables more effective optimisation of ad delivery at scale<\/strong>.<\/p>\n\n\n\n

    \n\n\n\n

    Innovations in the GEM Architecture<\/h2>\n\n\n\n

    One of the main challenges for recommendation systems is processing the enormous volume of interactions that occur between users and adverts.<\/p>\n\n\n\n

    GEM introduces several technical innovations to address this challenge.<\/p>\n\n\n\n

    Advanced Feature Modelling<\/h3>\n\n\n\n

    The model divides data into two main groups:<\/p>\n\n\n\n

      \n
    • Sequential features<\/strong>, such as a user\u2019s interaction history<\/li>\n\n\n\n
    • Non-sequential features<\/strong>, such as age, location, or ad format<\/li>\n<\/ul>\n\n\n\n

      Each group is processed using attention mechanisms specifically designed for that type of information<\/strong>.<\/p>\n\n\n\n

      \n\n\n\n

      Processing Long Interaction Histories<\/h3>\n\n\n\n

      Understanding user intent requires analysing long sequences of events, such as clicks, views, or interactions with content.<\/p>\n\n\n\n

      To address this challenge, GEM uses a parallel pyramid structure<\/strong> that can process thousands of historical events with reduced storage costs.<\/p>\n\n\n\n

      This allows the system to better capture signals indicating purchase intent or genuine interest in a product<\/strong>.<\/p>\n\n\n\n

      \n\n\n\n

      InterFormer: Learning Between Signals<\/h3>\n\n\n\n

      Another key innovation is InterFormer<\/strong>, a system that enables learning from the interaction between multiple features without losing information from the user\u2019s complete history.<\/p>\n\n\n\n

      This helps identify complex patterns that influence the probability of conversion<\/strong>.<\/p>\n\n\n\n

      \n\n\n\n

      Cross-Platform Learning Across the Meta Ecosystem<\/h2>\n\n\n\n

      GEM also introduces an important concept: multi-domain learning<\/strong>.<\/p>\n\n\n\n

      The model learns from interactions occurring across different platforms within the Meta ecosystem:<\/p>\n\n\n\n

        \n
      • Facebook<\/li>\n\n\n\n
      • Instagram<\/li>\n\n\n\n
      • Business Messaging<\/li>\n<\/ul>\n\n\n\n

        For example, if a user frequently interacts with video content on Instagram, that information can help improve ad predictions in the Facebook Feed<\/strong>.<\/p>\n\n\n\n

        This approach enables the creation of far more comprehensive and accurate recommendation models<\/strong>.<\/p>\n\n\n\n

        \n\n\n\n

        Knowledge Transfer to Other Models<\/h2>\n\n\n\n

        GEM also acts as a central model that trains and enhances other smaller recommendation systems<\/strong>.<\/p>\n\n\n\n

        To achieve this, it uses a knowledge transfer framework based on:<\/p>\n\n\n\n

          \n
        • Knowledge distillation<\/strong><\/li>\n\n\n\n
        • Representation learning<\/strong><\/li>\n\n\n\n
        • Parameter sharing<\/strong><\/li>\n<\/ul>\n\n\n\n

          This system is twice as effective as traditional methods<\/strong>, enabling hundreds of specialised models to benefit from the learnings generated by GEM.<\/p>\n\n\n\n

          \n\n\n\n

          Large-Scale Training<\/h2>\n\n\n\n

          Training a model of this magnitude requires highly advanced technological infrastructure.<\/p>\n\n\n\n

          Meta has redesigned its training system for GEM using:<\/p>\n\n\n\n

            \n
          • Thousands of GPUs<\/strong><\/li>\n\n\n\n
          • Advanced compilation with PyTorch 2.0<\/strong><\/li>\n\n\n\n
          • Custom GPU kernels<\/strong><\/li>\n<\/ul>\n\n\n\n

            Thanks to these optimisations, Meta has achieved a 23-fold increase in effective training FLOPs<\/strong>, allowing the model to scale to unprecedented levels.<\/p>\n\n\n\n

            \n\n\n\n

            The Future of GEM: Multimodal Models<\/h2>\n\n\n\n

            The next step in GEM\u2019s evolution will be to transform it into a multimodal model<\/strong>.<\/p>\n\n\n\n

            This means it will be capable of analysing and integrating information from:<\/p>\n\n\n\n

              \n
            • Text<\/li>\n\n\n\n
            • Images<\/li>\n\n\n\n
            • Audio<\/li>\n\n\n\n
            • Video<\/li>\n<\/ul>\n\n\n\n

              With this approach, Meta aims to build a system capable of understanding both organic content and advertisements within a single artificial intelligence model<\/strong>.<\/p>\n\n\n\n

              \n\n\n\n

              Conclusion<\/h2>\n\n\n\n

              The introduction of GEM marks an important step forward in the evolution of advertising recommendation systems.<\/p>\n\n\n\n

              As models become larger and more capable of processing vast volumes of behavioural data, advertising platforms are moving towards more intelligent systems that better understand user intent and interests<\/strong>.<\/p>\n\n\n\n

              For advertisers, this represents a clear opportunity: more relevant campaigns, more efficient optimisation, and stronger performance outcomes.<\/strong><\/p>\n","protected":false},"excerpt":{"rendered":"

              Meta has introduced GEM (Generative Ads Recommendation Model), the most advanced foundational ad recommendation model the company has developed to date. This system represents a significant shift in the way Meta selects and delivers adverts to users across platforms such as Facebook and Instagram, introducing an approach inspired by large language models (LLMs) applied to […]<\/p>\n","protected":false},"author":13,"featured_media":21673,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[1],"tags":[389,388,348],"class_list":["post-21675","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized","tag-bussines","tag-meta-2","tag-optimize"],"acf":[],"_links":{"self":[{"href":"https:\/\/relevantgroup.media\/en\/wp-json\/wp\/v2\/posts\/21675","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/relevantgroup.media\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/relevantgroup.media\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/relevantgroup.media\/en\/wp-json\/wp\/v2\/users\/13"}],"replies":[{"embeddable":true,"href":"https:\/\/relevantgroup.media\/en\/wp-json\/wp\/v2\/comments?post=21675"}],"version-history":[{"count":1,"href":"https:\/\/relevantgroup.media\/en\/wp-json\/wp\/v2\/posts\/21675\/revisions"}],"predecessor-version":[{"id":21676,"href":"https:\/\/relevantgroup.media\/en\/wp-json\/wp\/v2\/posts\/21675\/revisions\/21676"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/relevantgroup.media\/en\/wp-json\/wp\/v2\/media\/21673"}],"wp:attachment":[{"href":"https:\/\/relevantgroup.media\/en\/wp-json\/wp\/v2\/media?parent=21675"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/relevantgroup.media\/en\/wp-json\/wp\/v2\/categories?post=21675"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/relevantgroup.media\/en\/wp-json\/wp\/v2\/tags?post=21675"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}