Around this time of year, Google shares how many “bad ads” it killed the year before. And every year, the number grows. But not this time.
“Bad ads” are those that violate Google’s advertising policies, including ad fraud, phishing scams, and malware. That includes everything from a one-off accident to a coordinated action by scammers trying to make money.
Google shared today that the company’s ads team killed 2.3 billion bad ads in 2018, a decrease of 28 percent from the previous year, when it removed 3.2 billion bad ads. Put another way, Google took down more than 100 bad ads per second in 2017, and fewer than 75 bad ads per second in 2018.
Going after bad actors with machine learning
Google wouldn’t go on the record when we asked why it removed fewer ads last year, but a few factors combine to paint a pretty good picture. The first is that the company is increasingly blocking bad ad experiences before the scams impact people. Google says it made a “concerted effort to go after the bad actors behind numerous bad ads, not just the ads themselves” in 2018.
Next, Google says it used “improved machine learning technology” to identify and terminate almost 1 million bad advertiser accounts, in addition to nearly 734,000 publishers and app developers. That figure is approximately double the number in 2017. Acting at the account level naturally has a bigger impact in terms of addressing the root cause of bad ads — one bad account can run thousands of bad ads.
Also in 2018, Google launched 330 detection classifiers to better detect “badness” at the page level — nearly 3 times the number of classifiers launched in 2017. Google thus completely removed ads that violated its publisher policies from nearly 1.5 million apps and nearly 28 million pages. These violations are caught using a combination of manual reviews and machine learning.
Lastly, Google worked with cybersecurity firm White Ops, the FBI, and others in the industry to take down an international ad fraud operation. The operation, codenamed 3ve, exploited datacenters, computers infected with malware, spoofed fraudulent domains, and fake websites to produce more than 10,000 counterfeit domains and generate over 3 billion daily bid requests at its peak. Late last year, the FBI announced charges against eight individuals for crimes that included aggravated identity theft and money laundering.
Policies and a Policy manager
During 2018, Google introduced 31 new ads policies to address abuses in such areas as third-party tech support, ticket resellers, cryptocurrency, and local services. As a result, the company removed 207,000 ads for ticket resellers, over 531,000 ads for bail bonds, and approximately 58.8 million phishing ads.
Google also launched a new policy for election ads in the U.S. ahead of the 2018 midterm elections, verifying nearly 143,000 such ads. The company additionally released a political ads transparency report that provides more information about who bought election ads. And 2019 will bring similar tools ahead of elections in the EU and India, Google promised.
As for misinformation and low-quality sites, Google removed ads from approximately 1.2 million pages, more than 22,000 apps, and nearly 15,000 sites for violations of policies directed at misrepresentative, hateful, or other low-quality content. The “dangerous or derogatory” content policy, for example, includes a prohibition on hate speech and hateful content.
Speaking of policies, Google plans to launch a Policy manager in Google Ads, like its AdSense Policy Center, next month. The Policy manager will help advertisers ensure their creatives are policy-compliant and offer tips on common policy mistakes to avoid.
Google needs to continue fighting bad ads to ensure its users don’t associate its network with fraud, scams, and malware. Given that the majority of Google’s revenue comes from ads, that scenario is simply not an option.