YouTube, AI, and the ‘Adpocalpyse’

Article ImageIt’s a company’s worst nightmare: Disgruntled clients pull back due to dissatisfaction with the company’s business practices and do so publicly, with government officials urging them on. But when the company being targeted by clients’ ire is the video and search behemoth YouTube—and the business practice is the airing of revenue-generating advertisements for creators of extremist content with dangerous real-world implications—both the impact and the stakes are especially high.

Over the course of a few weeks in March 2017—after an article in The Times highlighted instances in which U.K.-government funded advertisements appeared on YouTube alongside videos extolling rape, anti-Semitism, and terrorism—one major advertiser after another began pulling its advertisements off the platform. Within a week of the The Times’ story, the so-called Adpocalypse was nigh, with brands such as AT&T, Verizon, and L’Oreal pulling ads off YouTube, rightfully concerned about guilt by association to ideological content their consumers might find offensive. Government officials, such as U.K. Home Secretary Amber Rudd, called for the company to take ownership for its role in monetizing hate.

YouTube moved fast to get ahead of the story, with parent company Google’s chief business officer Philipp Schindler pledging to increase brand safety measures and offer advertisers more control. Given the sheer volume of content created and consumed on the platform (with 400 hours of new content uploaded every minute) and the complexity of contextualization of video content, it was no surprise when Google’s general counsel Kevin Walker announced, in a June blog post, that artificial intelligence (AI) was going to be a major part of the solution.

“When you look at the exponential content being loaded up [to YouTube], it’s not even possible to scale to that with humans,” says Kurt Kratchman, COO of Grapeshot, which uses adaptive machine learning algorithms to improve ad targeting. Walker noted that Google would devote more resources to video analysis models that would assess previously removed terror-related and extremist content to help train AI “content classifiers” to find and pull such content more quickly.

Walker also announced other changes. These include a “trusted flagger” program that would enable human intervention to identify emerging areas of concern and negotiate the line between newsworthy free speech and violent propaganda. Rounding out the plan was the creation of a limbo-like category to be applied to content that technically abides by YouTube policies but is still problematic, making such content less discoverable and ineligible for monetization, and a commitment by YouTube to engage in counter-radicalization efforts.

For now, YouTube’s machine learning approach seems to be paying off. According to a company blog post from August 2017, its new AI-driven detection and removal tools are proving effective, with 75% of the videos its machine systems removed for violent extremism being identified without human intervention. YouTube acknowledged that the AI accuracy needs improvement; for instance, the Real Women Real Stories channel—which promotes awareness of the damage of sexual abuse on women—was flagged as problematic, although it has now been cleared. YouTube contends that its AI approach is already more effective than humans doing the same job and that it is scaling to cope with the vast volume of content in question.

And advertisers are coming back to YouTube. Advertising intelligence company MediaRadar says that while 5% of YouTube’s advertisers dropped off the platform in April 2017, its efforts to assuage advertisers are paying off. The company created a Google Preferred platform to give advertisers a curated collection of brand-safe videos. The number of Google Preferred advertisers grew from 217 in January to more than 500 in June.

Other social platforms are taking notice. When Facebook announced its plans to ramp up in-stream video advertising in the fall, Carolyn Everson, Facebook’s VP of global marketing solutions, quickly issued reassurances to advertisers that brand safety is built in, with placement and category opt-outs as well as pre-campaign reporting. Facebook is also hiring an additional 3,000 content reviewers, adding a human touch to flag content adjacency problems.

Unfortunately, the creation of extremist, violent content shows no signs of slowing, and it’s the nature of the threat that symbols, language, and images evolve to elude the kinds of speed bumps that YouTube is trying to set. The same week in September that Facebook announced its new brand-safety video plans, for instance, mega-popular YouTuber PewDiePie used a racial slur during a YouTube livestream (not his first such offensive issuance), and advertisers were once again on alert to make sure their ads didn’t appear on his channel.

Such situations are why companies and organizations are taking steps to address the need for brand-safe video content, turning to brand-safety solution providers such as Grapeshot. Grapeshot’s algorithm analyzes word weight within context, is language-agnostic, and allows brands to dial up and down settings for individual products, for a custom fit. “We’re not an either/or option to what YouTube is doing,” says Kratchman. “We’re an excellent additive option, especially when you consider that it takes forever to build up brand equity.” After all, it only takes one problem—an ad for a beach vacation served up alongside a video of Hurricane Irma devastating the Caribbean or a children’s charity advertisement superimposed on a video extolling gun violence—and your brand equity can evaporate overnight.

“And if you have a problem like that, it can take a year or two to recover,” says Kratchman. “That’s why, when it comes to brand safety, we recommend a zero-tolerance policy.” He points out how quickly the press and social media can amplify an unfortunate ad placement, long after it’s been fixed or removed. Anyone who follows @slpng_giants on Twitter, which channels public outrage at brands spending their ad budget at the alt-right Breitbart News, knows the effectiveness of its reach.

 “This has gone beyond the CMO and become a board issue, especially if it becomes a distraction from core business,” says Kratchman. “The cost to ensure data safety is so small to turn on and leave on,” he adds, especially compared to the damage that a problem, quickly amplified by the press and social media, can do.

Want more proof? Just ask YouTube about the Adpocalypse. 

Related Articles

From a meow to a roar: That's a simplistic, but spot-on, way to describe the evolution of online video, a realm previously ruled by cute and curious cat videos. But it's now populated by countless contemporary iterations of playable visual content—including those of the short-form, live-streamed, silent, branded, vertical, and viral variety.
On the web series Royal Crush, sophisticated teens fall in love with each other on cruise ships that bring them to some of the world's most majestic destinations and have playful onboard attractions such as a rock climbing wall, a spa, and a teen lounge. It's perfectly delightful entertainment, but that's not all. It's also a grand marketing effort from Royal Caribbean International. Royal Caribbean is one of many brands that have embraced episodic content, which refers to a single narrative developed over multiple segments. It's an elaborate content marketing play that employs the power of compelling stories to build stronger brand connections with viewers.
I bet Facebook is a really stressful place to work these days. Last week, Facebook announced the latest step it's taking to combat "fake news." This time Facebook is focusing on transparency in advertising. But what is Google doing to solve this issue before the next election?
Webinars are still widely promoted online, but does the format still resonate with end users and, if so, which ones? Can webinars reach, and influence, a B2C audience or are they primarily used for B2B outreach? What emerging technologies are moving into the webinar space? These are questions marketers need to be asking themselves.
Hold the phone, Mr. and Mrs. Marketer, and not so fast, John Q. Publisher: That mobile video content you're preparing to post better be future-proofed beyond just the message and branding. Viewable convenience and an amenable aspect ratio are also key criteria today and likely will be for the foreseeable future—which means that your video better be vertical.
Many businesses struggle to get their blogs going because they believe their industry doesn't have good potential to craft interesting posts. In reality, it all comes down to how creative you get with your writing. The more unique and compelling your posts are, the more powerful your blog is in growing your brand. Regardless of what line of work you are in, there will always be ways you can use your content to stand out among the competition.