4 Strategies for Targeting Children on YouTube Despite Advertising Restrictions

Table of Links

Abstract and Introduction

Background

Legislation on Advertising to Children

Mechanisms for Targeting Children

Usage of Placement-Based Targeting

Limitations

Related Works

Conclusion, Acknowledgements and References

Appendix

4 MECHANISMS FOR TARGETING CHILDREN

When advertising, Google does not allow advertisers to choose audiences of users under 18. Hence, the advertising interface does not allow advertisers to target children directly. However, placementbased advertising can be easily exploited to target children with ads: an advertiser simply needs to aggregate a list of videos or channels with content for children and target its ads at these videos. Next, we evaluate whether such targeting works in practice. We first describe how one can aggregate a list of children-focused videos. We then describe how we performed ad experiments to test whether Google allows placement-based advertising on children-focused videos.

4.1 Curated children-focused videos

We compiled two lists of children-focused videos and one list of general-audience videos (as control):

Children YTB Kids List–To compile a list of children-focused videos, we first take advantage of YouTube Kids and, hence, YouTube’s own characterization of videos as being dedicated to children. For this, we created an adult account, and we used it to create three children’s accounts corresponding to the three age groups proposed by YouTube Kids (under 4, 5 to 8, and 9 to 12). We then manually browse the different categories of videos recommended when navigating to the broadcasts, music, learning, and discovery tabs in YouTube Kids and collect the corresponding channels. We collected 22 YouTube Kids channels. For each channel, we then collect the ten most recent videos from its YouTube Kids channel page.

Children Seed List–We observed that not all children-focused videos available on YouTube are part of YouTube Kids. To build a complementary list, we took three channels with a large number of subscribers: Swan & Néo (5.97 M subscribers), Madame Récré FR (3.81 M subscribers), and Studio Bubble Tea (1.82 M subscribers), and did a snowball sample on the “Channels” section. The “Channels” section contains channels with similar content according to YouTube or that have been manually added by the channel’s owner. We added the channels collected (avoiding duplicates) at the end of our list and browsed this list in FIFO. Our stopping point was to have 200 YouTube channels in total. We then manually filtered out channels present on YouTube Kids (one can copy-paste the YouTube link on YouTube Kids to check if it exists) and channels not mainly intended for children according to the FTC guidelines [27]. Finally, we kept a list of 20 children-focused channels, and we collected the ten most recent videos from their YouTube channel pages.

Adult List–We compiled a list of general-audience channels to have a point of reference. For this, we created a profile with no search history or preferences. We collected the channels corresponding to videos recommended on YouTube when logging in for the first time. We compiled a list of 20 adult-focused channels (the intersection with the children-focused lists is empty), and we collected the ten most recent videos from their YouTube channel pages. The lists we compiled are not representative and are biased toward videos recommended by YouTube and more popular videos on the platform. Nevertheless, representativeness is not necessary if the goal of the ad experiments is just to test whether something is possible.

4.2 Targeting ad experiments

To check whether Google allows advertisers to target their ads to children-focused videos, we created seven real-world ad campaigns in which we instructed Google to place our ads on children-focused videos. Section 4.3 describes how we chose the content of the ads to minimize risks for children. As campaign parameters we chose the “Brand Awareness and Reach” campaign objective, instructed Google only to show the ad on YouTube (and not on its Display Network), and disabled the “audience expansion” option. We left all other parameters on the default value. We launched three ad campaigns where we only used placement-based targeting (CID 1 to 3 in Table 1) and four ad campaigns combining both placementbased and interest-based targeting (CID 4 to 7 in Table 1). For each ad campaign, we chose various combinations of placements: videos or channels from Children YTB Kids List, Children Seed List, and Adult List. CID 1 and 5-7 only use children-focused videos as placement, while CID 2-4 combine adult-focused with childrenfocused videos as placement. Our ad experiments were launched between Sept. and Nov. 2022. All our seven ad campaigns have been validated by Google and delivered to users. Nevertheless, this information alone is not enough to prove that Google delivered our ads on children-focused videos, as the platform could have potentially not respected our criteria and delivered our ads on videos that are not on our list.

Luckily, Google provides advertisers with an interface where they can check statistics about the delivery of their ad campaigns and allows us to check for each ad campaign the total number of ad impressions as well as the number of ad impressions on each video the ad was placed on [32]. Table 1 reports the total number of impressions as well as impressions on Children YTB Kids List and Children Seed List videos only. The table shows that all our ad campaigns resulted in our ads being shown on children-focused videos including videos sourced from YouTube Kids–these videos are marked by the platform itself as being primarily intended for children. Hence, this shows that advertisers can target children with ads on YouTube by asking the platform to place their ads on children-focused videos.

Placement-based targeting is done independently of the identity of the connected user and their Google profile. Hence, we performed four experiments where we instructed Google to combine placement-based with interest-based targeting in order to target ads both based on the video watched and based on the profile of the user watching the video. We perform this experiment because both the DSA and COPPA ban or restrict the use of profiling in advertising to children. Table 1 (CID 4 to 7) shows the placements and the interests we used. To check if Google has a different treatment for videos on YouTube Kids (as the platform knows they are intended for children), for CID 6 and CID 7, we used as placement only videos on YouTube Kids. We also checked if Google has different policies across different countries, maybe due to differences in legislation. CID 6 was targeted at the U.S., while CID 7 was targeted at 11 countries in the European Union. All four ad campaigns were validated and shown to users. Again, this is not enough to prove that Google is serving ads based on the profiles of users on children-focused videos, as Google might have ignored the profiling part of our request. Luckily, Google also provides statistics about the number of impressions shown to users with particular characteristics. For example, the statistics provided by Google for CID 4 say it had 393 impressions on users interested in “Travel”, 387 in “Sports & Fitness” and 161 in “News & Politics”. The statistics provided by Google indicate that for all four ad campaigns (CID 4 to 7), Google delivered the ads to only users with the interests we asked for (i.e., the total number of impressions across placements = the total number of impressions across interests). Hence, Google allows advertisers to perform targeting based on profiling when placing their ads on children-focused videos. Such a combination of targeting is allowed both in the U.S. and E.U.; it is also allowed on videos marked by YouTube itself as intended for children.

4.3 Ethics statement

We ran ad experiments on real ad platforms and targeted childrenfocused videos with our ads. Hence, our ads have probably been seen by children. We chose the content of the video ad and the landing page to limit potentially harmful effects on children. We first created a short video that records trees while walking in a park (the video does not capture any visible human face). Second, we used an already existing children-focused video on YouTube under a Creative Commons License that allows redistribution (Figure 3, in the appendix, show a screenshot of our two ads). Finally, the landing URL of the ad was either the tourism office of the city we lived in or the YouTube Channel page corresponding to the video we used as an ad. We set small ad budgets to limit the number of children that see our ads and stopped the ad campaign as soon as we saw impressions. We did an informal validation with colleagues (including legal scholars, behavioral economists, and cognitive scientists), and they did not express any concern regarding potential harm from our video ads.

We filed a request with our institute’s DPO (Data Protection Officer) for the data collection in Section 4.2, and we were exempt as we do not collect personal information from users in these experiments. The data collection gathered through a browser extension (Section 5) and the ad experiments to assess risks with online platforms (Section 4.2) were covered by a project-level IRB.

4.4 Implications

Our experiments show that is possible for advertisers to directly market to children on YouTube through placement-based targeting on children-focused videos. In light of the texts applicable in the E.U. and the U.S., it appears that the online targeting of children by contextual advertising is permitted. Hence, placement-based advertising is also permitted as it is a form of contextual advertising. Nevertheless, contextual advertising targeting children online is less secure than contextual advertising targeting children on television. Online advertisers can place an ad under a single piece of content (e.g., a particular cartoon or a particular video clip dedicated to children), which can make verification of the content of the ad very difficult in practice, as it will only be seen by children who have seen that content, with very little possibility of effective verification of the content of the ad by adults.

In contrast, the online targeting of children by profiling (such as interest, location, and behavior) is clearly banned in the European (DSA) and might be restricted in the American (COPPA). In fact, COPPA restricts the collection of personal information on content primarily intended to children, however, it is not clear from the text if it also restricts the use of previously complied user profiles (e.g., profiles complied when the users was browsing adult-focused videos) when serving an ad on a children-focused video. Hence, it appears from the experiments that there is a need to better distinguish between cases where targeting is carried out solely on the basis of context or where targeting is carried out by mixing contextual placement and profiling. This distinction is necessary to effectively implement the DSA and COPPA texts protecting children against advertising based on profiling.

Nevertheless, in 2019, the Federal Trade Commission (FTC) served YouTube with a $170 million fine for violating COPPA because it had illegally collected information on children for its personalized ads. To avoid future fines, YouTube committed to better distinguish content intended primarily for children by relying on a combination of self-reports from creators and algorithms. Since then, content creators have been obligated to label their videos as “Made for Kids” (or risks $45k fines otherwise). YouTube said they would assume any viewer of children-focused videos is underage, no personal data collection will be performed on such content, and no personalized ads will be sent on such content (as per the COPPA rule) [26, 60, 63, 64]. Hence, contrary to this declaration, we were able to send personalized ads on such content.