“Was this an Ad?!”: An Investigation of Paid Social Media Endorsements
This post summarizes our paper Endorsements on Social Media: An Empirical Study of Affiliate Marketing Disclosures on YouTube and Pinterest. This paper will be presented at the ACM Conference on Computer-Supported Cooperative Work and Social Computing (CSCW 2018) on Wednesday, November 7, 2018.
When Casey Neistat, a popular YouTuber, first posted this video showcasing his elaborate summertime fantasy, he did not anticipate the enormous backlash it would receive. While Casey stated that the video was filmed using a Samsung smartphone, he failed to state that Samsung had sponsored the video itself.
Disguised advertising of this kind is increasingly common on social media platforms, where content creators — colloquially called Influencers — endorse brands and products in order to make money. However, this trend has drawn concerns about transparency because Internet users — teens particularly — often fail to distinguish between sponsored and real content, and therefore may be misled or deceived.
To prevent such harm to users, content creators are required to disclose these advertising relationships to users according to the Federal Trade Commission’s (FTC) guidelines. We studied whether content creators who engage in such advertising disclose these to users, focusing our work on those using affiliate marketing links — a marketing strategy which pays content creators when users click on their customized links. To do so, we collected and analyzed a sample of about 500,000 YouTube videos and 2.1 million Pinterest, and in a surprising result, discovered that only about 10% of affiliate marketing content on either platform is actually disclosed to users. The vast majority of disclosures are in a format the FTC specifically discourages, and which users do not understand.
Identifying affiliate marketing content
We began our analysis by detecting affiliate marketing content in our dataset of YouTube videos and Pinterest pins. Detecting this content is a challenge in itself since there does not exist a comprehensive public repository of affiliate marketing companies and their links.
Using methods we describe in the paper, we gathered all the links embedded in the videos and pins, and used these to identify affiliate marketing links and content. Across both platforms, we found that affiliate marketing was most commonly found in technology, fashion, and beauty related content — e-commerce product types for which consumers proactively seek out online reviews.
Discovering content creators’ disclosures
Having identified affiliate marketing content, we analyzed the videos and pins for affiliate marketing disclosures. We extracted disclosures from the videos’ and pin’s descriptions using natural language processing and machine learning techniques.
Of all the YouTube videos and Pinterest pins that contained affiliate links, we found that only ~10% and ~7% respectively contained accompanying disclosures. When disclosures were present, we could classify them into three types:
- Affiliate Link disclosures: The first type of disclosures simply stated that the link was an “affiliate link”, or that “affiliate links were included”. The FTC’s endorsement guidelines explicitly ask content creators to refrain from using these disclosures in their content. These disclosures were the most prevalent across both YouTube and Pinterest.
- Explanation disclosures: The second type of disclosures attempted to explain what an affiliate link was, along the lines of “This is an affiliate link and I receive a commission for the sales”.
- Support channel disclosures: Finally, the third type of disclosures — exclusive to YouTube — told users that they would be supporting the channel by clicking on the links in the description, without exactly specifying how.
Validating content creators’ disclosures
Finally, we conducted experiments to validate the disclosures we discovered. We designed two experiments — one each for YouTube and Pinterest — through which we elicited whether participants notice the different disclosure types, and understand their underlying message. A total of 1,791 participants from Amazon Mechanical Turk participated in our experiments.
We discovered that users fail to notice and correctly interpret Affiliate Link type disclosures — the most prevalent type of disclosures and those that the FTC advocates against using — on both platforms. Users correctly interpret the more descriptive Explanation disclosures but only when they can notice them; in our study, users were unable to discover these on YouTube videos, which tended to have longer descriptions than Pinterest pins on average.
Improving disclosure practices
While our findings paint a grim picture of the state of endorsement-based advertising disclosures, they also raise practical policy and design solutions to enable better disclosures.
Thus far, the FTC’s enforcement action has targeted content creators and in only the last year, the agency has sent warning letters to those who have failed to disclose their advertising relationships. We recommend that the FTC could also target affiliate marketing companies and hold them accountable when their content creators fail to disclose. Doing so would incentivize affiliate marketing companies to hold their content creators to higher standards. Future work could also aim to better understand why content creators fail to disclose in the first place.
From a design perspective, we recommend that social media platforms can make it easy for content creators to disclose their relationships with advertisers to end-users. Recently, YouTube and Instagram have taken steps in this direction, releasing tools that enable automatic paid advertising disclosures on posts. Such tools can help standardize the disclosures across advertisement posts, but will likely have to further account for various other kinds of advertising strategies.
Arunesh Mathur, Arvind Narayanan, and Marshini Chetty. 2018. Endorsements on Social Media: An Empirical Study of Affiliate Marketing Disclosures on YouTube and Pinterest. In Proceedings of the ACM on Human-Computer Interaction, Vol. 2, CSCW, Article 119 (November 2018). ACM, New York, NY. 26 pages. https://doi.org/10.1145/3274388
Code and Data
The data and analysis scripts are available on GitHub.