Building the Guardian’s Live Elections Notifications

How we put real-time updating AP results data straight on the lock screen

Photo by Xinhua / Barcroft Images

Well. That happened.

November 8th was a big day. Big in a lot of ways we won’t get into here, but it was unquestionably consequential. For the Mobile Innovation Lab, election night was big in not-so-obvious ways: we, along with the Guardian’s US news desk and interactives team in New York, and the apps product and development teams based in London, broke new ground with news notifications by offering a live updating alert of election results as part of the Guardian’s election coverage.

Working alongside the excellent Guardian teams, we used a modified AP data feed to show up-to-the-moment elections results in a single alert that automatically updated with the latest results throughout election night. We offered these worldwide to users of the Guardian iOS and Android apps, and offered a quick sign-up experience to US-based users on both platforms.

This was, to our knowledge, the first use of the new iOS 10 notifications technology in this way, and absolutely the first time live notifications that auto-update have been used on a US election night. It was also the first time the lab’s work was available on iOS, and the first time work we had done was integrated into our parent organization’s main coverage and core products.

From the outset of the lab, we knew we wanted to experiment with a variety of ways we might tell stories or pieces of them through notifications, and to see what would resonate with audiences. We also wanted to experiment on the two biggest mobile platforms, Android and iOS. After much experimentation on Android through web notifications, and once more of the specifics of iOS 10 were announced last summer, the plan to create live updating notifications for both platforms became more of a reality.

The elections alert was in many ways the scaled-up version of the the experiments we ran during Brexit and the US Primaries. While we were only able to serve a relatively small audience on Android during those projects, the lessons we learned from them greatly informed the plans for and scope of the election night notifications. In particular, we adapted lessons learned in how to present and order different data points, how to signal that there was new information, and how to allow users to opt out.

While these notifications were an extension of the lab’s earlier experiments and allowed us to test many of the same assumptions with a much larger audience, they were also a part of the Guardian US’s overall package of coverage on election night, which also included a live blog, a detailed map of results built by the interactives team (also built around AP data and which offered detail down to the county level) and the more traditional text-based alerts sent to Guardian app users at key moments.

Because the results alerts were integrated into the apps, we were able to attract a far bigger audience than we have had in earlier experiments, which allowed us to learn far more about how the alerts were used and how well the format served the readers who received them.

Below is more detail on what we did and what it amounted to:

Presenting a hierarchy of information for users across platform and screen size:

When planning these notifications, we started by figuring out what information was the most important to be displayed without any user interaction required in the collapsed view of the alerts.

We also needed to think across platforms. iOS and Android alerts show different amounts of information in their collapsed states. On iOS we were able to get a bit more information in without having to expand the alert, whereas on Android, we needed to make some tough choices on what information took precedence.

The collapsed view of the alerts as they appeared on Android (top) and iOS (bottom)

We first sent the alert with a title — e.g. LIVE NOW: US election results — that would give a clear signal that the results were now coming in. Both alerts featured the Guardian logo, but with the news desk also sending update alerts with the same logo, we wanted to to differentiate between them by indicating that this was a live results alert. We did that in the title, which on each platform had a version of the words live election results.

Second, because the electoral vote is ultimately what matters most in determining a winner (deep breath), it was the next piece of information that was critical to appear on the collapsed version of the alert on the user’s lock screen. We provided a rolling count of electoral votes as they were reported by the AP.

On Android, that’s all we were able to show in the collapsed alert. On iOS we were able to display two additional bullet points to give more context about the electoral vote: reminding users what number of electoral votes we were counting towards, and indicating how many states had been called.

In the expanded view of the notification, which you get to by pulling down on the notification on Android or hard pressing on iOS, we were better able to achieve parity of information across platform.

The expanded view on Android

Each of the expanded notifications included a data visualization of the electoral votes and the amount needed to reach the winning threshold, accompanied by the delightful 8-bit illustrations of the candidates, created by the Guardian US interactives team.

In the expanded view we were able to show four bullet points of information on each platform. (If we had been less worried about parity between platforms, we could have had more on iOS). In the four lines we were able to have on both, we looked to present several pieces of information.

The expanded view on iOS
  • Number of states called: This provided context of how much more was to come in the electoral count, as well as the popular vote.
  • Number of swing states called: We thought indicating how many of these states’ results had been called might give an idea of the direction of the electoral count. We also displayed the abbreviation of as many of these states as possible. We were able to fit in up to three abbreviations, or two and “incl.” per line.
  • Number of precincts reporting: Similar to the number of states called, this number was intended to give context and indicate the progress of the popular vote.
  • Popular vote percentage: Again, we thought that the percentage of all votes that each candidate had won, according to the precincts reported, would provide a sense of how Americans were voting.

We chose these categories based on our own knowledge of US election coverage, talks with others in the newsroom about what they would be reporting, and what information we received from the AP.

In addition to determining how the data would display in the notification, we also took two important user experience features into account, and built controls for them into the tool used to customize the alert.

  • Major news updates: Through a tool built by Connor Jennings, one of the lab’s developers, we had the ability to change the alert title (and also to interrupt the bullet points, though we never used that feature). As major states and swing states were called, we changed the title of the notification to show the most recent and important news.
Screenshot of the notification editing tool Connor built for election night
  • Notification noise: There’s rarely a use case for a news organization to send a silent alert given the way that push notifications are usually used for breaking news. But for an updating notification, taking advantage of what has been a latent opportunity on Android and iOS (the ability to send a silent notification) suddenly made a lot more sense. We decided to have our initial notification to make sound, but all updates thereafter would be made silently. If the election had been called by midnight ET or so, we intended to have the alert buzz and make noise as the electoral count reached 270. Such as it was, the race was called at just around 3AM ET, and we opted not to.

Finding our audience:

Part of what we discussed and worked with the UK apps and product teams on was how to opt users into the alert. Typically in our experiments we get people to sign up by pointing them to web page. Since the feature was built into the Guardian apps this time we relied on an onboarding screen — a recruitment tool the UK teams had successfully implemented in the apps for two earlier alert-driven features in the apps, the Campaign Minute and the Euro 2016 Minute. The onboarding screen was a full-screen message to app users explaining what the alert was, and how to sign up or opt out. For election alerts, users saw the screen the second time they opened the latest version of the Guardian’s iOS or Android app.

The onboarding screen, as it appeared to iOS users

This screen gave a short description of the alert and let the user toggle on or off the alerts. The onboarding screen ran on iOS and Android in the US, and on Android in the rest of the world. In the US, we had the toggle set to opt users in, in recognition that they were geographically more predisposed to be interested in the results than those using the Guardian app internationally. We reflected that assumption by having the toggle set to off in the onboarding screen users saw outside the US.

How they did:

More than 230,000 Guardian app users subscribed to our alerts. For earlier experiments, our largest audience had been about 14,000 (for our Brexit web notifications). The increase, though largely due to the integration into the Guardian’s core apps, was a huge coup for the mobile lab. Thank you to everyone who tried it out.

In addition to those who already had the app installed and saw the onboarding screen, the UK apps team registered 20K to 25K additional Android app downloads during recruitment time.

Our users were located across the globe, with the greatest number of engaged users in the US, followed by the UK, with Australia, Germany and Canada occupying the third through fifth spots respectively.

The differences in global sign-ups were clearly reflected in the platform split we saw. Overall we had 170,000 Android subscribers worldwide. The onboarding screen was served worldwide on Android, providing a one-tap solution for users. By contrast, the onboarding screen was only served on iOS within the United States and we saw fewer worldwide subscribers as a result — 60,000 users opted in.

While we’ve worked to build a comprehensive experience within the notifications themselves, we are also interested in learning more about how users tap through on the alerts, and how they use the additional functionalities we programmed into the action buttons. Overall, we were really pleased to see the amount of traffic the notifications sent into the app, and the degree to which users used the new button-activated functionalities.

Here are some details, which, due to a mid-launch migration of the Guardian analytics platform, these statistics are extrapolated from Android user data:

— Over the course of the night, the alerts drove over 800K page views to the election night live blog and pages containing a full breakdown of results.

— Users could engage with the alert in three ways, which 74% of them did. Of that, here’s what they tapped on:

  • The alert itself, which led to the live blog. Users tapped an average of 4.5 times throughout the night.
  • The Live Blog button. Users tapped an average of 2 times each.
  • The Full Results button. Users tapped an average of 2.5 times each.

In keeping with previous experiments, we also sent out a follow-up survey to our users seeking qualitative feedback. We sent the survey at 3pm ET the Friday following the election via an app alert to all subscribers to the election-night notification who had not since opted out. We received more than 6,000 responses.

Overall, the alert received extremely high marks in the survey. An overwhelming majority of users thought the alert was useful, clear, effective and convenient. Nearly all respondents (98%) said they would definitely or maybe sign up for the alerts again.

In terms of usability, around 70% of respondents said they expanded the alert and also knew how to stop it if they wanted to opt out. Even more respondents, around 80%, reported knowing that the alert would automatically update, which is incredibly encouraging to hear when we’re considering integrating even more interactivity and content into the expanded view.

The data visualization graphic was popular among respondents, with 70% saying that it was useful, and another 56% saying that it was clear and scannable.

In terms of what users would have liked to have seen more of, they said they would have appreciated more data overall — including state-level information, projections and popular vote count.

On effectiveness:

  • 84.2% of our survey respondents told us that they found the notification effective or very effective at providing a convenient view of the results’ progress.

On engagement:

  • Upwards of 70% of users tapped through on the alert at some point throughout the night, either by tapping the alert itself, or one of the action buttons. That’s very similar to the engagement we saw with the Brexit live data alerts, when 77% of subscribers tapped on the alert at some point during the event.

We’ll be following this post up with more about what we learned creating and deploying this new notification format. Stay tuned.


How did you follow the US elections? Let us know what you think. Send us an email at innovationlab@theguardian.com.

The Guardian Mobile Innovation Lab operates with the generous support of the John S. and James L. Knight Foundation.