Tech

Here's how it's possible for Google to sell ads next to offensive content and not know about it

Vox Media CEO: Google is going to be just fine
VIDEO1:5601:56
Vox Media CEO: Google is going to be just fine

Google is under fire after several ads were seen running next to offensive content in the U.K. — including neo-Nazi and jihadist videos — on YouTube and other websites it serves ads on.

The revelation, first reported in early February by The Times of London, led media buying agency Havas to pull all its YouTube and Google Digital Network ads in the U.K. Johnson & Johnson, JPMorgan, AT&T and Verizon, among others, have suspended or pulled advertising with Google as well.

How online ad buying works

To better understand how these ads wind up next to offensive content, it helps to understand how companies buy ads online.

  • A brand has a product it wants to advertise. Its internal research has identified the demographic most likely to buy the product — for example 18- to 25-year-old men.

  • The brand goes to a media agency and explains the objective. It gives the media agency a budget.

  • The media agency does more research and determines the best ways to reach this demographic. It may find that these hypothetical 18- to 25-year-old men also are more likely to be cat lovers and eat pizza. Then it determines the best places to run the ads to reach those people.

Say that media buyers decide YouTube is the best place. Here are the main ways they could buy an ad there:

Want an ad on Youtube? Here’s how to do it
VIDEO1:0101:01
Want an ad on Youtube? Here’s how to do it
  • Talk to a Google sales representative to buy ads on Google Preferred videos. These ads run on the top 2 to 5 percent of Google's videos based on such criteria as the number of views and subscribers. Typically, brands that buy Google Preferred have bigger budgets of six figures or more, according to Matt Borchard, media director at media agency Noble People. These ads are also bought in a similar manner to TV upfronts — meaning brands commit money before the videos are posted.

  • Those with smaller budgets can still get their hands on Google Preferred-quality videos. Smaller players can do this by buying inventory that's left over after the big budget group is finished. They also contact Google sales reps to get their deals done. The budgets don't have to be as big and the ads may be cheaper, but they get the second go-around on premium videos. These sales are comparable to TV scatter markets — the sales on the remainder of programming after upfront deals are made.

  • Most remaining ad space gets sold on Google's ad sales platform AdWords or DoubleClick. These are automated platforms that allow buyers to bid on demographics or topics, and ads are priced based on supply and demand. Media buyers set the maximum price they want to pay and which groups of people they want to reach. The platform then sells them the videos that fit their needs, based on internal Google algorithms.

  • Buyers can also buy through third-party managed services, like Zefr or Ad Parlor. These companies have people and technology who "watch" YouTube videos and label what they are about. Media buyers can ask for a specific type of video, like videos with cats or tied to events like March Madness.

    The third-party service makes informed purchases on AdWords, and can provide a list of video URLs where the ads ran. However, advertisers pay more for these services — about 5 to 6 cents more for each completed view, or $3 to $4 more per 1,000 viewers, according to Borchard. This metric is called "cost per mille," or CPM. (YouTube's average CPM is $2, but it varies widely, especially when the videos reach in-demand demographics.)
  • YouTube video creators and Google split the revenue. Splits are typically 45/55 percent, with the bigger share going to the creator. However, channels with large viewerships can sometimes negotiate better shares. It's also important to note that advertisers who buy "TrueView" ads — which are the majority of YouTube ads sold — only pay if the user watches the whole ad or after 30 seconds have passed.

On the other hand, a media buyer could decide that a simple display ad is better. Display ads are text, image or video ads that appear on websites, essentially the online equivalent of print ads. Google is one of the biggest companies that "serves" online ads, meaning it owns technology that places ads on websites.

Here's how those media buyers can buy display ads through Google:

Here’s how to buy a display ad through Google
VIDEO0:2500:25
Here’s how to buy a display ad through Google
  • Media buyers use Google's AdWords or DoubleClick services or direct through a Google salesperson. They can also buy "programmatically" through a third-party ad tech firm. Similar to YouTube, media buyers name their price and which demographics they want to reach.

  • Google algorithms determine which websites would be appropriate for the goals. Ad tech firms have their own algorithms that find the best places considering cost and exposure to the right groups.

  • The websites and Google split the ad revenue.

Why Google ads can be better ... and worse

In the old days, if a brand wanted to reach a certain demographic, media buyers would turn to a TV channel that they knew had a high number of viewers in that group, said Mikkel If Hansen, partner and product owner at media analytics platform Blackwood Seven. The company uses artificial intelligence and machine learning to help brands make more efficient ad buys.

But those ads would appear to anybody watching the TV channel, including people outside the desired demographic. Advertisers, in a sense, overpaid to reach their desired audience.

Digital platforms allow brands to target only the people they want. If people are logged into their Google accounts and have entered their personal information, the company can tell their age, gender, and location data, among other aspects like search history and websites they visit.

Though digital ads are more targeted and can reach more people than TV commercials, Google and other ad tech companies use computer programs rather than people to decide where to place ads, because media buyers are purchasing type of person not a specific website. It also means no one knows all the instances the ad appeared.

"It's difficult for them to see what they bought because you are buying behavior and not actual placements," Hansen said.

In addition, technology isn't quite good enough yet to block objectionable content instantaneously. Google can block websites tagged with offensive words, but it's harder to "see" what the videos are about, said Borchard. About 400 minutes of video is uploaded to YouTube every minute.

"The sheer amount of video that is being uploaded to YouTube is mindboggling," Borchard said.

In short, there's far too much content online to scan and immediately remove if deemed offensive. Companies still rely on people to tag bad content that may have slipped through their sensors. Plus, there are potential legal issues that prevent blocking content.

In a blog post on Monday, Google's chief business officer Philipp Schindler apologized for the ad adjacency issues, and said the company is working to make sure that ads only show up next to creators the company can vouch for, and make it easier for advertisers to control which kinds of content their ads appear next to. Schindler also said the company will share more about where ads run, as well as act faster to take down questionable videos.