Polls, Clickbait, and Commemorative $2 Bills:
Problematic Political Advertising on News and Media Websites Around the 2020 U.S. Elections

Eric Zeng, Miranda Wei, Theo Gregersen, Tadayoshi Kohno, Franziska Roesner
This paper appears at the 2021 Internet Measurement Conference (IMC 2021).

Paper with Screenshots (PDF) Dataset https://doi.org/10.1145/3487552.3487850


The 2020 U.S. Elections were one of the most important and contentious elections in recent history. Spending on online political advertising broke records during this election cycle, with over $200 million being spent by both the Biden and Trump campaigns.

In this paper, we investigated the following research questions about online political advertising:

We collected data on online political ads from September 25th, 2020 to January 19th, 2021 using web crawlers. We conducted daily crawls on 745 online news and media websites, including both mainstream news outlets and websites labeled by fact checkers as "misinformation". Using VPNs, we performed these crawls in six locations across the U.S.:

In total, we collected 1.4 million online ads, ~56,000 of which contained political content.

Key Findings

Using Politics for Profit

Certain online ads used inflammatory political content as a means of attracting attention and clicks, sometimes using deceptive techniques. However, the true purpose of these ads were to ultimately to earn a profit, either by selling products, harvesting personal information like email addresses, or earning revenue through ads on the landing page.

Misleading Political Polls
This ad, from the Democratic Strategy Institute, shows a picture of Anthony Fauci, and asks the reader to thank him for “telling the truth” by signing a card. This ad, from “rightwing.org”, asks the reader vote in a public opinion poll on whether illegal immigrants deserve unemployment benefits.
Examples of ads that appear to be political polls or petitions, as a tactic for collecting email addresses.
Democratic Strategy Institute / rightwing.org

One example of problematic political ads we observed were misleading political polls or petitions. Though they looked like polls or petitions that urged you to vote on a political issue, their purpose wasn't actually to measure public opinion. Instead, submitting a vote almost always required giving an email address that be used to subscribe you to mailing lists. Such lists ask for campaign donations, advertise questionable products, and in some cases, spread political misinformation.

Political Product Ads
... This ad, from Legacy Research, claims that they know the stock that would perform well if Biden won the election.
Examples of a political memorabilia ad, and a product ad that uses political content to attract attention.
Proud Patriots / Legacy Research

Political content was also used in advertisements for tangible products. Some ads for political memorabilia used classic bait-and-switch tactics like "free" products... if you pay for shipping and handling. Other ads used political figures and events to attract attention for their products, like ads promising get-rich-quick investments.

Clickbait Political News Ads
This ad shows a picture of Joe Biden, with a headline about him going on a head-turning rant. This ad shows a picture of Jill and Joe Biden, with a headline about Jill’s ex-husband making a surprising allegation. This ad shows a picture of Vanessa and Don Jr. Trump, with a headline about Vanessa changing dramatically after leaving Don Jr.
Examples political "clickbait" ads. These ads are in the "native ad" format, which imitate links to actual articles on the parent website.

Many news websites will run "native ads", which imitate the look and feel of native links to news articles on their site, but are in fact ads for articles on third party websites. (See our previous work investigating native ads). These ads are often written with "clickbait" headlines.

We observed many examples of native ads with political content, which often used sensationalist headlines to imply outrage and controversy about politicians. However, their landing pages were almost always less shocking and outrageous than the headlines typically claimed.

Political Ads are Targeted at Partisan Websites

Political ads appeared more frequently on websites with a higher level of partisan bias. Using data from AllSides and Media Bias/Fact Check, we categorized the political bias of the sites we crawled, as Left, Lean Left, center, Lean Right, and Right.

Our data shows that significantly more political ads appeared on sites categorized as Left and Right, compared to Lean Left and Lean Right, as well as centrist and uncategorized sites.

This graph shows the percent of ads on a site that are political, based on the site's political lean. For mainstream news sites, the percentages of political ads are as follows: 6.85% for left sites, 4.38% of lean left sites, 2.26% of center sites, 9.02% of lean right sites, and 10.33% of right sites. For misinformation sites, the percentages are: 26.03% of left sites, 0.59% of lean left sites, 0% of center cites, 5.75% of lean right sites, and 7.89% of lean right sites.
The percentage of ads, out of all ads on those sites, that were political, by sites' political bias and misinformation label. Higher percentages of ads on partisan sites were political, compared to centrist/uncategorized sites.

Looking specifically at campaign ads, political advertisers also tend to target sites matching their affiliation: Democratic and liberal groups ran most of their ads on left-leaning sites, while Republican and conservative groups ran most of their ads on right-leaning sites. For example, a large number of ads on Daily Kos and Occupy Democrats (classified as Left, Misinformation by Allsides and Media Bias/Fact Check) were from Democratic affiliated advertisers.

This graph shows the percentage of ads on websites that are political, for ads from each political affiliation. It shows that ads from Democratic Party advertisers primarily are on Left and Lean Left sites, and ads from Republican Party advertisers are primarily on Right and Lean Right sites.
The percentage of ads observed on websites from advertisers of different political affiliations, split out by the political bias and misinformation label of the website. Advertisers tended to run ads aligned with their politics.

We also observe this effect for the types of problematic ads we found above, such as political polls and political products. For example, misleading political polls appeared more on right leaning sites than other sites:

Political Lean % of Ads that are Political Polls and Petitions
Right 2.2%
Lean Right 1.1%
Center 0.2%
Lean Left 0.2%
Left 1.1%

These results suggest that political ads are contextually targeted at partisan political news outlets, meaning that people who read from these sources are more likely to be exposed to political advertising, including misleading, outrage-provoking, and potentially harmful political ads.


Download Dataset (220MB)

Our dataset of 1.4 million political and non-political ads is available at the link above. This dataset contains the following information:

The dataset does not currently include the raw screenshots or HTML of ads that we collected. The full dataset is over 2TB in size, and we are working on ways to share this more easily. If you would like the full dataset, please contact us at badads@cs.washington.edu.


This project was a collaborative effort of Eric Zeng, Miranda Wei, Theo Gregersen, Tadayoshi Kohno, and Franziska Roesner.

This work was supported in part by the National Science Foundation under grant CNS-2041894, the UW Center for an Informed Public, and the John S. and James L. Knight Foundation.

Read More

More research on problematic online advertising from our lab: