The U.S. public is consistently exposed to incendiary rhetoric and misinformation online, and a great deal of that is directed at the Black community. From dog whistles to unconscious bias, this has left the Black community vulnerable.

Advertisers fund billions of dollars in content every year ($263 billion in the U.S. alone), and so brands have a responsibility to address the toxicity crisis online and its real, human impact on people. We need to be accountable for the content we surround and sponsor.  

We had to teach the algorithms of the Impact Index exactly what was good and bad in terms of human impact, and used cutting-edge, supervised machine learning to do so. 

We started with a definition for each news filter that we built, based on academic research. There were four main overall categories: Positive Impact, Neutral Impact, Negative Impact, and Toxic Impact, each geared towards the Black community.  

Then we captured and identified thousands of examples from all around the web to match that definition. Those articles were then used to train deep learning models, with each algorithm containing millions of features/inputs. 

We validated each algorithm and tested its performance. Before it was ready, each algorithm had to perform well on news it had never seen. The models get periodically updated with new data to make sure they stay accurate across time.  

The Impact Index ranks individual news articles, based on their content, across thousands of publisher sites across four categories:  

- Positive Impact: Uplifting or celebrating the Black community.  

- Neutral Impact: Neither positive nor negative; oftentimes lifestyle content.  

- Negative Impact: Refers to criminal, anti-social, or non-normative behavior.   

- Toxic Impact: Invokes negative stereotypes, belittling, or is purposely distrustful of the Black community.  

Through these classifications, we gained insights about the type of content to surround, and what to avoid. For example, we learned the most common form of bias is in local news, where some local news publishers often “boost the signal” of negative news pertaining to the Black community. Meanwhile, lifestyle sites boosted positive stories, but generally with little impact on the Black community. 

We then used this data to optimize current media plans and inform future activations, allowing us to prioritize Positive Impact sites across campaigns.  

Using this tool, we adjusted our investment and increased Tyson’s media volume on Positive Impact sites by 20%.

And as a result, we doubled the amount of brand engagements vs. historical benchmarks.

With paid media and AI we began creating a safer internet for the Black community, increasing brand performance along the way.

"Tyson Foods has one of the most diverse workforces, and we're proud to advance diversity, equity and inclusion by committing to actions like the Impact Index," said Paul Davis, vice president and chief diversity, equity, and inclusion officer for Tyson Foods. "The Impact Index ensures our actions in brand responsibility ladder up to our core values as a company and individuals."