Why ‘bad’ ads appear on ‘good’ websites – a computer scientist explains

Examples of “bad ads” found on the web include clickbait articles, potentially unwanted programs, miracle weight loss supplements, crude images and investment pitches. Screenshot by Eric Zeng

Sketchy advertisements, such as miracle diet pills and suspicious software, sometimes appear on legitimate and reputable websites. It turns out that most websites don’t actually decide who can serve ads to their viewers. Instead, most sites outsource this task to a complex network of ad technology companies that determine which ads are served to each specific person.

The online advertising ecosystem is largely built around “programmatic advertising”, a system for placing advertisements from millions of advertisers on millions of websites. The system uses computers to automate advertisers’ bidding on available ad space, often with faster transactions than would be possible manually.

Programmatic advertising is a powerful tool that allows advertisers to target and reach people on a wide range of websites. As a PhD student in computer science, I study how malicious online advertisers take advantage of this system and use online advertisements to deliver scams or malware to millions of people. This means that online advertising companies have a great responsibility to prevent harmful ads from reaching users, but sometimes they fail.

Programmatic advertising, explained

The modern online advertising market is supposed to solve one problem: to match the high volume of advertisements with the large number of advertising spaces. Websites want to keep their ad slots full and at the best prices, and advertisers want to target their ads to relevant sites and users.

Rather than each website and each advertiser teaming up to serve ads together, advertisers work with demand-side platforms, technology companies that allow advertisers to buy ads. Websites work with supply-side platforms, technology companies that pay sites to put ads on their page. These companies manage the details to determine which websites and users should be matched with specific advertisements.

Most of the time, ad tech companies decide which ads to show through a real-time auction. Whenever a person loads a website and the website has space for an ad, the website’s supply-side platform will request bids for advertisements from demand-side platforms through a system of auction called ad exchange. The demand-side platform will decide which ad in its inventory best targets the particular user, based on the information it has collected about the user’s interests and web history from user browsing tracking. , then submit an offer. The winner of this auction can place his ad in front of the user. It all happens in an instant.

When you see an ad on a web page, behind the scenes, an ad network has just automatically launched an auction to decide which advertiser has won the right to show you their ad. Eric Zeng, CC BY-ND

Major players in this market include Google, which operates a supply-side platform, a demand-side platform, and an exchange. These three components make up an ad network. Various smaller companies such as Criteo, Pubmatic, Rubicon and AppNexus also operate in the online advertising market.

This system allows an advertiser to serve ads to millions of potential users, on millions of websites, without needing to know the details of how this happens. And it allows websites to solicit advertisements from countless potential advertisers without needing to contact or enter into an agreement with any of them.

Filtering bad ads: an imperfect system

Malicious advertisers, like any other advertiser, can take advantage of the scale and reach of programmatic advertising to send scams and links to malware to millions of potential users on any website.

There are checks against bad publicity at several levels. Ad networks, supply-side platforms, and demand-side platforms usually have content policies limiting harmful ads. For example, Google Ads has an extensive content policy that prohibits illegal and dangerous products, inappropriate and offensive content, and a long list of deceptive techniques, such as phishing, clickbait, false advertising, and doctored images.

However, other ad networks have less strict policies. For example, MGID, a native ad network that my colleagues and I looked at in a study and found to serve a lot of substandard ads, has a much shorter content policy that prohibits ads illegal, offensive, and malicious, and a single line about “misleading, inaccurate, or misleading information”. Native advertising is designed to mimic the look and feel of the website it appears on and is generally responsible for sketchy advertisements at the bottom of news articles Another native ad network, content.ad, has no content policy on its website.

These 2020 election political ads are examples of potentially deceptive techniques to trick you into clicking on them. The ad on the left uses Trump’s name and a clickbait headline promising money. The ad in the center purports to be a thank you card for Dr. Fauci but is actually intended to collect email addresses for political mailing lists. The ad on the right looks like an opinion poll, but links to a page selling a product. Screenshots by Eric Zeng

Websites can block specific advertisers and ad categories. For example, a site can block a particular advertiser who served fraudulent ads on their page, or specific ad networks who served low-quality ads.

However, these policies are only as good as their application. Ad networks typically use a combination of manual content moderators and automated tools to verify that each ad campaign complies with their policies. Their effectiveness is unclear, but a report from ad quality firm Confiant suggests that between 0.14% and 1.29% of ads served by various supply-side platforms in Q3 2020 were of poor quality.

Malicious advertisers adapt to countermeasures and find ways to evade automated or manual auditing of their ads, or exploit gray areas in content policies. For example, in a study my colleagues and I conducted on misleading political ads in the 2020 US election, we found numerous examples of fake political polls, which claimed to be public opinion polls but asked for an email address. -email to vote. Voting in the poll got the user on political mailing lists. Despite this deception, ads like these may not have violated Google’s content policies on content policy, data collection, or misrepresentation, or may have simply been missed in the review process. .

Bad Ads by Design: Native Advertising on News Websites

Finally, some examples of “bad” advertisements are intentionally designed to be misleading and misleading by both the website and the advertising network. Native ads are a great example. They are apparently effective because native advertising companies claim higher click-through rates and revenue for sites. Studies have shown that this is likely due to users having a hard time telling the difference between native ads and website content.

A grid of three native advertisements that look like news articles.  One ad sells CBD gummies, another is a clickbait story, and the last tries to sell financial advice.
Here are examples of native ads found on news websites. They mimic the look and feel of links to news articles and often contain clickbait, scams and questionable products. Screenshot by Eric Zeng

You may have seen native ads on many news and media websites, including major sites like CNN, USA Today, and Vox. If you scroll to the bottom of a news article, there may be a section called “sponsored content” or “on the web”, containing what looks like news articles. However, all these contents are paid. My colleagues and I conducted a study of native advertising on news and misinformation websites and found that these native ads disproportionately contained potentially misleading and misleading content, such as advertisements for unregulated health supplements. , misleadingly written infomercials, investment pitches, and content from content farms.

This highlights an unfortunate situation. Even reputable news and media websites are struggling to generate revenue and are turning to serving deceptive and misleading ads on their sites to earn more revenue, despite the risks this poses to their users and the cost to their reputation.

Eric Zeng, PhD candidate in computer science and engineering, University of Washington

This article is republished from The Conversation under a Creative Commons license. Read the original article.


Comments are closed.