Coverage of the 2020 U.S. presidential election will mark the beginning of gerrymandered news coverage, as hyper-personalization and geotargeting are applied at scale in news apps and in mobile push notifications.

Personalization and social filtering have both been widely leveraged to maximize user engagement with digital news. The communication of unique messages to specific audiences has capitalized on various personalization strategies that target individuals within populations based on: the channel or platform; behavioral information, such as click history; social network information; and users’ self-identified preferences. Location-based personalization is also common in targeting specific consumers, yet has only recently emerged in a news context.

The 2018 midterm elections provided a first glimpse of location-based targeting in journalism, with major news apps and aggregators using location data to personalize push notifications. This was most prominent in the Florida gubernatorial race: National news outlets provided not only different push notifications at different times to local and national audiences — they also fragmented the state, with different alerts sent to users in Orlando and Tallahassee, Miami and Fort Myers.

For example, following the primary elections nominating Republican Ron DeSantis and Democrat Andrew Gillum, some locations only received alerts from NBC News about one party’s candidate — no matter the ideology or previous reading behavior of the user:

In Miami: DeSantis wins Florida GOP primary for Governor, will face Democrat Andrew Gillum after upset. (August, 29, 2018, 6:13 AM)

In Orlando and Tallahassee: Andrew Gillum wins Florida Democratic governor primary. (August, 29, 2018, 6:37 AM)

National: Bernie Sanders-backed Democrat Gillum to take on Trump-endorsed Republican DeSantis in Florida gov race, NBC News projects. (August 29, 2018, 5:05 PM)

Local elections can have national interest and implications, making election coverage applicable to multiple audiences based on their location. Push notifications about these stories increasingly frame the same topic in ways that are unique to local and national audiences, often even when they link to the same story. Personalizing to different local audiences could have the measurable outcome of impacting name recognition for candidates in particular districts.

This location-based personalization, however, is distinctly different from historic differences between local and national news coverage. Local affiliates have traditionally provided further details on stories that may impact local audiences, and smaller news outlets have provided more coverage than national organizations for local-specific stories. While the use of geo-location targeted push notifications has parallels to the nested models of local affiliations and national organizations, the human-information interactions and overall implications — particularly relative to how this applies to communications around elections — are very different.

Personalized push notifications are somewhat unique in that, while the majority of them are paired — with frames for local and national audiences, but pointing to the same story — there’s no guarantee that a user ever taps on the alert to read the story. That means the tailored frame might be all that the user sees. As a result, geo-targeted notifications have the potential to manipulate the information environment, even through subtle choices about detail or sentiment, potentially exacerbating social-geographic fragmentation and gerrymandering understanding of elections.

In a 2018 study we conducted, we found that location-based personalization had only recently emerged, with hundreds of examples in a few states related to political news — a very small subset of push notifications overall. In some cases, this personalization was not only harmless but also highly useful and effective. In other cases, however, it had the potential of amplifying problems like filter bubble, manipulation, and loss of readers’ trust.

Given the preview provided by the 2018 midterm elections, we anticipate that location-based information will be used to send alerts with different frames for the same story; at different times, based on relevance; selectively, with different details, based on local relevance; and from local affiliates, using a national platform.

Recent studies have cast doubts on the extent to which filter bubbles generate completely different perceptions through news coverage. But push notifications present unique personalization opportunities, in terms of location and of the controlled capture of behavioral patterns. We believe that, even if some concerns about filter bubbles have been overstated, location-based personalization of push notifications could present a different and more significant challenge in this context — especially given the hyper-partisan geographic divides in the United States.