On April 23, a federal judge in Oregon blocked the enforcement of a new rule banning abortion referrals at taxpayer-funded women’s health clinics. The next morning, Apple News featured coverage of the decision by The Hill, a national outlet, rather than the local newspaper, The Oregonian. The Oregonian had even published its story first—the Hill article linked back to it—just not in the proprietary format that Apple News requires.
Since its launch in 2015, Apple News has become a massive driver of news attention, reporting 85 million active users as of January 2019. The Top Stories section of the app sends a surge of traffic to the stories featured. On some days, inclusion in Apple News’ Top Stories accounts for more than half of traffic at Vox.com. Slate reported in late 2018 that Apple News was driving more readership than Facebook.
But the flood of attention and traffic unleashed by the app isn’t distributed evenly.
This spring, Jack Bandy and I audited the content in Apple News as part of our work at the Computational Journalism Lab at Northwestern. Our audit focused on the Top Stories section, which is shown prominently at the top of the screen when the app is first opened. These stories are known to be selected by a team of editors. We compared the selections to the Trending Stories section, usually the next section visible when scrolling down in the app, which runs articles that are automatically selected and ranked by an algorithm.
A study conducted by the Tow Center for Digital Journalism last year showed that Apple News’s Top Stories section in the UK funnels attention to a small handful of publishers. In our audit we wanted to compare how Apple News’ two approaches to curation—editorial and algorithmic—might affect which news outlets get attention. Like many algorithmic news curation systems, little is known about how Apple’s algorithm works or what types of stories it is designed to surface.
Our automated audit script sat on the Apple News US app for 62 days, from March 9 to May 9, 2019. It checked for new headlines every five minutes, and by the end we gathered 1,268 articles from Top Stories and 3,144 articles from Trending Stories. (The full data set is here, and we’ve open-sourced the code if you’d like to re-run the audit).
We found that both sections selected articles from only a handful of sources. In the human-edited Top Stories section, ten news outlets (The Washington Post and CNN foremost among them) accounted for 55.7 percent of all articles, and in the algorithmic Trending section ten outlets (led by CNN, Fox News, and People) accounted for 74.8 percent of articles. Both sections roughly follow the Pareto principle, where the top 20 percent of sources account for about 80 percent of articles (76 percent for Top and 84 percent for Trending, to be exact). Overall, attention in the Top Stories section is somewhat more evenly distributed across outlets.
In the 62-day period we audited, the Top Stories section ran articles from a total of 87 different news outlets, slightly more than the 83 news outlets found in the Trending Stories section. But between the two there were only 40 in common. Of the ten most popular news outlets featured in either section, only two outlets overlapped: CNN and The Washington Post.
From the above chart, you can see that the editors of Top Stories overwhelmingly included stories from national outlets. They did at times pull from local and regional publications such as The Chicago Tribune or The Miami Herald; this accounted for 20 of the 87 total sources but only 8.3 percent of the articles featured in the section. Moreover, half of those articles were from the LA Times (one of the few news organizations that has opted in to the Apple News+ program, coincidentally). Meanwhile, in the Trending Stories section, not a single locally or regionally specific source was cited.
The next chart compares how frequently the top twenty news sources appear on Top Stories and Trending Stories. Trending tends to favors softer sources such as People or BuzzFeed. And, notably, Fox News dominates the Trending section, but appears in the Top Stories section far less frequently.
The headlines in the Trending section tended to refer to celebrities, Donald Trump and his family, and shocking or sensational stories (“florida man” appeared in no fewer than 11 headlines in the two-month span). In contrast, headlines in the Top Stories section often reflected more substantive political issues on topics such as health (“affordable care act,” “measles cases”), immigration (“border wall,” “sanctuary cities”), and international politics and events (“new zealand mosque”).
In terms of pace, articles typically turn over in the Trending section every 2.9 hours, much faster than Top Stories, which had an average turnover of 7.2 hours. Trending Stories showed a consistent churn: new stories trickled in fairly uniformly throughout the day. By contrast, Top Stories received punctuated updates in the morning, midday, afternoon, and evening.
Apple makes no mystery of the fact that its recommendations are “based on both editorial curation and personalized suggestions.” The app allows users to follow or block specific topics and publications that influence the selection of content shown. Implicit personalizations can also draw on a user’s “reading habits” as well as patterns of use of “Safari and other apps”. “The more you read,” Apples states, “the better News understands your interests.” What’s less clear is which sections are implicitly personalized are and which aren’t. The For You section includes “recommendations based on topics & channels you read”—a clear signal for personalization—but Top Stories appears not to be personalized as it is “selected by Apple News editors.” What about Trending?
In our audit, we sought to determine if Trending Stories is personalized. To do this we crowdsourced synchronized screenshots of the section from different people in different locations. We then compared the screenshots to automatically scraped results. We found no evidence that the articles in Trending are personalized or adapted based on location. This means that unless someone is blocking a particular source, everyone sees the same thing, possibly with a delay of a few minutes as Apple’s servers update.
The various sections of Apple News appear to be generated by different editorial perspectives in mind: editorial, trending, and personalized. Top Stories is hand-curated by editors and tends to be newsy, while Trending Stories traffics in popular entertainment, celebrity, and the sensational, and remaining sections appear to be influenced through implicit personalization. While no single section would be satisfying as an app on its own, the composition of the three distinct editorial perspectives bundle different kinds of content in a way that recalls traditional newspaper sections. And, just like a paper, the app offers opportunities for incidental exposure: to get to the softer news in Trending, for instance, users need to first scroll past the harder news selected by editors in the Top section.
At the same time, the hard news selected for inclusion in the Top Stories section is relatively concentrated in terms of sources. Few sources are local, regional, or international in scope. Is it really good for the public when only 20 sources account for more than 80% of articles? And, given the traffic boost enjoyed by those few sources, is it fair to smaller news organizations producing relevant content that often predates the stories produced by big newsrooms?
Nicholas Diakopoulos is an assistant professor at Northwestern University School of Communication, author of the book “Automating the News: How Algorithms are Rewriting the Media” on automation and algorithms in news media, and regular contributor to CJR on these topics.