NBCUniversal's AI-generated audience platform is a system that uses large language models (LLMs) to analyze every movie and show owned and created by NBCU, along with web-based content. The AI writes summary descriptions for each piece and is trained on data from academic research about behavioral science, human emotion, and motivation theories. It then mathematically maps content descriptions with academic materials to score similarities between human emotions and show episodes, creating over 300 unique audience segments identified by AI for targeted advertising.

In an alpha phase test with clients from various industries, NBCUniversal's AI-generated audiences led to an increase in sales ranging from 22% to 46% more than the existing machine learning models. This test was conducted to compare the performance of AI-generated audience segments against established machine learning models for performance-based campaigns.

Creating generative AI tools for audience targeting involves several challenges, including:

• Cleaning and labeling data: Ensuring the data used to train AI models is accurate and well-organized.
• Scaling insights: Extending the insights from language models to visual AI models that analyze content.
• Privacy protections: Implementing safeguards to protect user privacy.
• Accuracy: Making sure AI models do not produce false results or create incorrect associations for audience segments.
• Bias checking: Evaluating both the data and the AI models for potential biases that may no longer be accurate or appropriate, such as outdated cultural references in older TV series.

Disney's Magic Words advertising product is a feature that analyzes scenes across Disney’s vast library to serve advertising that resonates with a specific emotion or cultural touchpoint. It was introduced as a way to tie mood to messaging, allowing for new products and actions while viewers stream content using a remote or phone.

AI video models like OpenAI's Sora and Google's Veo have the potential to power bespoke, brand-specific models for executing campaigns in CTV (Connected TV). These technologies, once out of beta, could leverage audience data and historical campaign inventory data to produce content that is tailored for the right audience, on the right channel, and with the right message. Although the production capabilities aren't ready for prime time on addressable TV, the pace of innovation in the industry suggests that this could change quickly.