Other platforms are experimenting with layouts that integrate both channels, and Dong expects further refinement by platforms seeking to maximize the engagement that drives their revenue—primarily clicks on paid ads.
“The platform’s goals don’t always perfectly match the interests of advertisers or users,” Dong said. “Their focus is on increasing users’ stickiness to the platform, or in other words, to hook you.”
That is, the algorithms aim to serve up recommended content that will keep users online and clicking on ads.
The research quantified the differences in how users perceived and responded to these two content streams.
The study, “Tales of Two Channels: Digital Advertising Performance Between AI Recommendation and User Subscription Channels,” was published in the Journal of Marketing in July 2023.
“Source credibility and perceived control are higher in the subscription channel, and because of this, the level of engagement is different. Users are spending more cognitive effort in the subscription channel,” Dong said.
By contrast, the AI-driven content stream is akin to “comfort food,” feeding content that feels familiar and requires less cognitive effort from the user. Over time, this narrow diet of content may lead to a breakdown in users’ critical thinking skills and cognitive engagement while browsing, Dong said.
In this less cognitively engaged state, ads feel less intrusive, and users are more likely to click on them. The study found that native ads served in the AI recommendation channel had a 20.08% higher click-thru rate than ads served in the subscription channel.
“This higher click rate is good for platforms’ revenue, but it may not be good for advertisers. Users click more but have lower conversion rates,” Dong said.
The study’s field experiment tallied more than 300 million exposures to 297 ads over three months on Weibo, a Chinese social media platform. It demonstrated conversion rates for ads in the recommendation channel were 15.63% lower than in the subscription channel.
Since platforms are paid per click, regardless of whether that click ultimately leads to a sale, advertisers must create ads that most resonate with users.
The study examined the efficacy of ads in relation to two attributes: whether they were crafted for “emotional” or “informational” appeal, and whether they used direct links (e.g. “Buy Now”) or indirect links (e.g. “Learn More”).
In general, informational ads and direct links are seen by users as more intrusive. However, in the less-engaged cognitive state engendered by the recommendation channel, users are more tolerant of these intrusions and more likely to click but less likely to buy.
Within the recommendation channel, informational ads with direct links increased click-thru rates 31.13% but decreased conversion rates by 25.55%; while emotional ads with indirect links decreased click-thru rates by 20.63% and increased conversion rates by 74.19%.
Within the subscription channel, emotional ads with indirect links receive the highest click-thru rate, but information ads with indirect links are optimal for driving conversions.
Beyond click-and-conversion rates that concern advertisers, a steady diet of recommended content can have negative effects on users, Dong said. As algorithms develop a more specific profile of a user, recommended content becomes narrower in perspective and more similar to what’s been viewed before. Ultimately this pattern can create what Dong calls a “filter bubble.”
“This can make your world smaller and smaller until you’re surrounded by a perfectly comfortable world,” she said.
No longer encountering new and different ideas, users slip into a “low engagement mode,” Dong said. This cognitive state decreases users’ critical thinking skills, reduces their ability to process new information and ultimately harms their creativity and capacity for deep learning.
She encouraged individuals to fight back against this pattern and take back control of their online media diets.
“You can exert a two-way influence on what AI learns about you,” she said. “Be more selective in interacting with subscription channels. Just like eating a well-rounded diet is good for you, intentionally seeking out different ideas and perspectives in the media can help fight the negative effects of the filter bubble.”
Story by Dan Armstrong
Optimal Ad Design
To maximize Click-thru Rate, advertisers should use emotional ads with indirect links for both subscription (1.34%) and recommendation (1.41%) channels.
To maximize Conversion Rates, they should use indirect links with informational ads (3.85%) for the subscription channel and indirect links with emotional ads (3.94%) in the recommendation channel.
To optimize overall efficiency (conversion/impression), they should consider using emotional ads with indirect links in the recommendation channel (.056%), a 40% increase over subscription (.040%).
Behind the Study
As social media platforms continue seeking to increase their revenues, they are turning more to AI-generated algorithms to feed users’ appetite for in-feed advertising.
While corporations are poring over mountains of user data to optimize their revenues, there exists relatively little academic research on the effects of this trend. According to Beibei Dong, associate professor of marketing in the College of Business, that’s because companies are reluctant to share proprietary data or partner with researchers on large-scale field studies, and universities simply lack the resources to run such large-scale studies through other means.
For this study, the authors partnered with a Chinese video game company to run a three-month field study using real advertisements placed on the social media platform Weibo. In total, 297 ads placed received more than 300 million impressions.
The project was expanded and validated using additional lab-based studies, including a controlled experiment using a sample of 132 students and a controlled experiment with 300 research subjects recruited via the Mechanical Turk.
An additional controlled experiment used cutting-edge eye-tracking technology with a sample of 53 students as a method of verification of prior empirical findings.