August 20, 2019
Brand safety should mean knowing your ads aren’t on domains that disinform. Unfortunately that is not the case today for most brands.
Too much money and too many actors have overwhelmed the ad tech space. There is no way for brands to follow their ad spend online or employ criteria to steer away from sites with high disinformation risks. Partly because there is no reliable assessment of the disinformation risk of a domain. Disinformation actors out for money and clicks know that.
But we at the GDI are working to change the rules of the game.
First, we have scoped how much of a problem is online ad spending on disinformation domains.
We estimate that ad exchanges pay US$235 million each year to disinformation domains for carrying brands’ programmatic advertising (as we highlighted recently on the BBC and CNN). Programmatic advertising is buying and selling advertising space on web pages in real time. It constitutes a major part of the digital media world, accounting for two-thirds of all global digital ad spend in 2019.
The GDI has calculated this figure based on forthcoming research which will be released in September. It is based on estimates for 20,000 disinforming domains gathered from around the web. The study will provide greater detail into which ad tech companies service the largest number of domains, and how much money each ad tech company is inadvertently funneling to known disinformation news sites.
Second, we are working to build a real-time list of disinformation domains and broader index that ad tech companies and brands can use to better direct their ad spend online.
The GDI is creating automated classifiers to capture high-risk disinformation news sites. This work builds on initial efforts in 2018 to prototype a neural network classifier based on over twenty different technical metadata signals. Taking over 300 known high-risk disinformation domains (“junk””), the prototype classifier correctly identified 98.8 percent of them. The classifiers will allow for ad exchanges to have real-time feeds of “high-risk” domains as soon as these sites appear online. We are actively seeking ad exchanges and agencies to pilot the classifiers with us starting in September.
The automated classifiers will be enhanced by a broader index that will look at news domains of all different disinformation risk levels. The index will be based on the automated classifiers and add human-review components that assess other disinformation flags related to the content, operations and context of the domain.
The classifiers and index will give brands, advertisers and platforms transparent and constant information about the domains on which they are buying ads and promoting. And it will begin to disrupt the ecosystem that is allowing disinformation to flourish and create incalculable harms - both online and off.