Some AI-powered podcast ad tools are labeling Black culture focused podcasts as “unsafe” for advertising.

🚨 Diverse Voices Under Threat?

New research released today by Sounder, a podcasting management platform, and Urban One, a Black-owned media organization in the U.S., found that traditional keyword-based approaches to brand safety are often inaccurate, resulting in under monetization of diverse content. For instance, a podcast episode that discusses race or social justice might be flagged as unsafe solely based on certain keywords. 

🔍 The Study's Findings

The study found that applying a standard blocklist tool to Urban One's podcast content removed 90% of episodes from the available inventory. The tools didn’t understand that the term “bomb” means “cool”, not a literal explosive device.

However, when using a semantically and contextually focused machine learning model to classify brand safety, only 10% of episodes were potentially available for monetization.

Transcription accuracy was also identified as a critical factor in correctly classifying content. The study said errors in transcribing common Black vernacular (AAVE) can lead to misclassifications and further marginalize BIPOC creators.