Organic and Paid Content on Platforms – Discussion Paper 2: The Impact of Social Media Platforms in Elections

During the 2019 general election, citizens used a wide swath of digital and social media platforms to create and engage with a range of content, from memes to leaders' debates. Regulated political entities, including candidates, political parties and third parties, were also active on several major platforms, as were media outlets and civil society groups. Elections Canada used organic content and digital ads to explain where, when and ways to register and vote, and participated in some platforms' special initiatives to increase election awareness and voter registration.^{Footnote 9}

Digital and social media platforms differ from other spaces where Canadians encounter and interact with people, information and ideas. To have a productive discussion about how platforms shape the digital information sphere and impact the democratic process, and to think about addressing regulatory gaps therein, it is important to understand platforms' functionality and architecture. The following section draws on scholarship to describe how digital and social media platforms curate, moderate and disseminate organic and paid content.

Organic content

Organic content on social media platforms—content that the poster (i.e. an individual or organization) does not pay the platform to publish or distribute—is subject to relatively few regulations in Canada.^{Footnote 11} It is subject to existing criminal and civil laws around expression. In the pre-election and election periods, it is also subject to some Canada Elections Act (CEA) provisions that forbid the impersonation of Elections Canada and political entities, the publication of misleading information for the purposes of affecting the election outcome, and some types of false statements about candidates and people associated with parties.

Otherwise, the governance of organic content largely falls to the platforms themselves. Before the 2019 general election, many platforms committed to taking measures to safeguard electoral integrity, such as by offering enhanced security for political campaigns' accounts and by setting up dedicated contact channels for political entities and Elections Canada to report election-related incidents to the platform. Some platforms also signed on to the government-led Canada Declaration on Electoral Integrity Online, a voluntary pledge to remove fake accounts, bots and inauthentic content.^{Footnote 12}

Platforms govern organic content by defining what is possible, permitted and promoted. They determine what actions are available to users, such as by offering buttons that let users "like" or "share" content with a click.^{Footnote 13} Most platforms display engagement metrics (numbers of likes, shares, follows, etc.), identifying the popularity of content and offering users real-time feedback on the "performance" of their posts.^{Footnote 14} Through the creation and enforcement of policies such as terms of use and community standards, platforms moderate accounts and content and remove what they deem unacceptable.^{Footnote 15} Platforms also promote some content over others, through recommendation algorithms.^{Footnote 16}

Recommendation algorithms exercise strong influence: they determine what each user sees, sees first and does not see in their timeline (Facebook News Feed, YouTube queue or equivalent). Though algorithms differ from one platform to another, change often and are largely unknown to users and outside entities, researchers have a general idea of how they work.

Many recommendation algorithms assign a relevancy score to each piece of content, based on each user's personal and behavioural profile (described below), to predict how a given user will respond to each item.^{Footnote 17} They then use this information to rank content in each user's timeline, placing content that generates engagement—clicks, video plays, likes, comments or shares—at the top.^{Footnote 18}

By promoting content that generates strong reactions, algorithms generally reward and increase the reach of content that is provocative, entertaining and shocking.^{Footnote 19} Research suggests that these effects of recommendation algorithms, coupled with the human drive to emotional response, can impair users' ability to detect and access reliable information. For instance, a study of engagement on Twitter found that posts exhibiting "indignant disagreement" received nearly twice as much engagement as other types of content; each moral or emotional word used in a tweet (such as "greed," "evil" or "shame") boosts its reach by 20%.^{Footnote 20} Another study found that users are more likely to believe and share articles with emotional headlines, even when they have not read or evaluated the article.^{Footnote 21} A third study showed that Twitter users are 70% more likely to share untrue news—often sensational and novel—than factual news.^{Footnote 22} While content creators in other media may also skew their content toward the sensational to get better reach, the tendency is more strongly reinforced on social media platforms, because they offer real-time feedback (likes, shares, retweets) on how audiences are reacting.^{Footnote 23}

Users may also have difficulty assessing the validity of information on these platforms because content is, for the most part, presented without obvious clues about its source, authenticity, quality, or the interests of those who created or shared it.^{Footnote 24} Unable to examine the facts and motives behind every post they see, users can take cognitive shortcuts; they tend to believe posts that come from friends, that are repeated or that are accompanied by photos.^{Footnote 25} Users' challenges in assessing information are compounded when an individual they trust, such as a political leader, posts or shares inaccurate or misleading information.^{Footnote 26}

Data-driven digital advertising

The digital display advertising market in Canada is significant: in 2019, 53.5% of the total amount spent on advertising was spent on digital ads, representing $8.8 billion.^{Footnote 28} As with organic content, during the 2019 general election, Canadians encountered political content in the form of digital ads. This section provides a high-level description of how data-driven digital advertising services function, particularly in the Canadian electoral context.

Most social media platforms deliver data-driven advertising; they use automated software that lets advertisers exploit user data to target and optimize ads. Other companies, notably Google, also offer data-driven ads, displayed on search engine results pages and websites.^{Footnote 29} In recent times, virtually all digital ads have been data-driven,^{Footnote 30} with Google and Facebook accounting for almost three-quarters of the Canadian market.^{Footnote 31}

Ads are generally subject to more regulation than organic content. In the pre-election and election periods defined by the CEA, certain ads are subject to spending limits,^{Footnote 32} reporting requirements and taglines disclosing who paid for them.^{Footnote 33} Before the 2019 general election, a new CEA provision came into law: it requires platforms and websites reaching a defined threshold of visitors and running regulated ads posted in the pre-election or election period to create ad registries.^{Footnote 34}

Ad registries must make regulated ads public and provide information on who paid for each ad.^{Footnote 35} Major platforms responded to this new requirement in various ways: Google announced it would ban all issue and partisan ads in Canada;^{Footnote 36} Twitter announced it would accept such ads in the election period only;^{Footnote 37} and Facebook and some other platforms continued to accept these ads and published them in ad registries.^{Footnote 38}

How data-driven digital advertising services work

Data-driven advertising services deliver, optimize and price ads in ways that are different from those of traditional advertising channels, such as broadcast television. Their features, designed for commercial purposes, are very powerful, enabling advertisers to target consumers precisely and optimize their ads based on detailed real-time feedback on their performance with targeted segments of their audience.^{Footnote 39}

At the heart of digital advertising is data: platforms may hold tens of thousands of attributes on a single user.^{Footnote 40} This includes data that users provide directly, inferred data and behavioural data.^{Footnote 41} It can also include psychographic data^{Footnote 42} and highly detailed personal information purchased from data brokers.^{Footnote 43}

Platforms use these rich data to group users into segments for ad buyers—one reportedly offers 29,000 segments based on characteristics that include ethnic affinity, income, support for breastfeeding and smart device usage.^{Footnote 44} Some platforms allow advertisers to upload email addresses, phone numbers or postal codes to target known individuals.^{Footnote 45} Some also permit advertisers to target users who resemble a known audience, based on characteristics that the platform "thinks" are salient, but that are not made explicit to the advertiser or ad targets.^{Footnote 46}

Leveraging their detailed user data sets, machine-learning algorithms then optimize ads automatically and instantly, zeroing in on who is most receptive to their messages, when and in what format.^{Footnote 47} Using these tools, advertisers can profile consumers based on their susceptibility to various appeals, tweaking ads through multiple iterations in an attempt to find the most influential persuasion strategy for each individual user, such as appeals to authority or identity, favourite colours or images.^{Footnote 48} These ad services with machine-learning capabilities enable advertisers to continually iterate ads, often running thousands or tens of thousands at the same time, many to small audiences, in an effort to see which ads "stick."

The pricing model for data-driven advertising is also unique. Many ad services on platforms offer a real-time bidding system, where multiple advertisers compete to reach particular audiences. For example, if company A wishes to advertise to skiers, and company B to women in Alberta, when a female skier from Calgary logs on, the platform weighs the companies' bids and serves the woman the winning bidder's ad, based on calculations it has completed in microseconds.^{Footnote 49}

In choosing bids, some platforms consider factors other than price, such as relevancy—the likelihood that users exposed to the ad will watch, click or share it.^{Footnote 50} As one platform puts it, "we subsidize relevant ads in auctions, so more relevant ads often cost less and see more results."^{Footnote 51} Platforms' data-driven advertising services also let advertisers optimize campaigns to achieve specific behaviours: for example, if an advertiser optimizes for "shares," the platform shows ads to people who are the most likely to share, at a lower cost.^{Footnote 52}

Data-driven advertising services are popular among advertisers precisely due to the services' use of data and ability to optimize ad delivery at a relatively low cost to the advertiser.

Users are often not aware of how and why they see particular ads, nor are they aware of the impact that their clicks, likes and shares will have on the ads they will see in the future. Likewise, advertisers themselves may lack information on how these data-driven ads function. For instance, due to the ad-pricing models that many advertising services use, advertisers may not know how much a competitor is being charged for the delivery of their ads, and may be charged more or less than their competitors, depending on the content of the ad and who they are seeking to reach with it.^{Footnote 53} This raises questions about how level the playing field is for political actors purchasing digital ads.

• Previous
• Table of Contents
• Next