What’s Going On With AMP Analytics?

Accelerated Mobile Pages are designed to drive engagement, but your analytics tools might be telling a different story. Don’t panic; read this.

As publishers take a first, close look at audience analytics for Accelerated Mobile Pages, they might find that certain engagement measures appear to under-perform standard mobile web content. AMP was specifically engineered to remedy the low engagement and high bounce rates plaguing the mobile web… So what’s going on?

Keep in mind that AMP reach is still nascent and there hasn’t been much time to test and develop industry best practices for discovery and navigation of AMP content. There’s much learning ahead.

Also, AMP is a standard, not a rigid template — leaving room for wide variation in execution. Some publishers are merely “checking the box” by rendering bare-bones AMP pages, limiting the opportunity for engagement (and revenue). Others with robust AMP implementations are reporting encouraging results; a recent Washington Post case study credits AMP for a 23% increase in mobile search users who return within a week.

Learning curve and execution factors aside, there’s a more basic explanation for underwhelming metrics in analytics reports: AMP behaves differently than standard web content, and requires a different measurement approach.

The Unique Properties of AMP

Some of AMP’s most innovative features are curve balls for traditional audience metrics. The following observations are based on a standard Google Analytics implementation using <amp-analytics>, but the concepts should apply to other tools with similar methodologies.

First, AMP is cross-domain by design. AMP pages can be cached and delivered from any domain, as when Google presents articles in the AMP carousel viewer via cdn.ampproject.org. This means a user could — and likely will — traverse multiple domains while engaging with a publisher’s content.

Analytics services which rely on anonymous browser cookies* to track users won’t recognize this cross-domain activity as part of a continuous session. A user who visits a publisher’s article on cdn.ampproject.org and then clicks to a second article on the publisher’s AMP domain will appear as two users, engaging in two sessions of one view each. As a result, AMP activity is likely to register more users and sessions, with more single-page visits and fewer views per session than analogous interactions with standard web content. See Google Analytics and AMP for more detail.

In this typical user path, the first article accessed from the AMP carousel is delivered by cdn.ampproject.org inside the Google viewer. When a user clicks through an internal link to a second article on the publisher’s domain, Google Anatlyics should capture both views but will not recognize them as a continuous session by the same user.

*Publishers with user authentication services may have the ability to reconcile cross-domain and cross-platform activity of logged-in users.

Second, AMP introduces new user interfaces. The AMP carousel viewer in Google mobile search results is the primary example. The carousel viewer encourages users to browse content from multiple publishers by swiping from one article to the next. Each time a user loads and swipes past a publisher article in the viewer, a single-page session is registered in Google Analytics. If no interaction events are recorded, the session is also a bounce. This activity can produce a large number of single-page sessions, with an inflationary effect on time on page and a negative effect on session duration. See Google Analytics and AMP for more detail.

Each time a user loads and swipes past a publisher article in the viewer, a single-page session is registered in Google Analytics. If no interaction events are recorded, the session is also a bounce.

Initial Recommendations for AMP Measurement

Understanding how AMP activity is recorded is the first step toward benchmarking and managing AMP performance. The situation presents an opportunity to think about KPIs for the AMP and non-AMP web. After all, AMP is just one facet of an increasingly distributed, cross-platform content ecosystem.

Here are some initial thoughts on AMP performance measurement:

  1. Track AMP separately. Use separate tags for tracking AMP audience in order to benchmark AMP trends and prevent AMP data from skewing standard web reporting. In Google Analytics this means setting up a separate property for AMP; Google recommends this approach.
  2. Look at new metrics. For example, scroll depth may be one of the most meaningful and most overlooked metrics available to publishers. In a world where users bounce between social platforms and publisher pages, average scroll depth on article pages could be a leading indicator of user experience — and in certain contexts, a better engagement measure than the number of pages consumed in a continuous session. Think of it as “viewability” for content.
  3. Look at the bottom line. User engagement and ad performance should generally track together. While the underpinnings of a robust AMP advertising economy are just now being established, it’s not too early to benchmark certain advertising KPIs such as viewability, CTR and programmatic rates for AMP inventory. AMP is designed to drive emerging performance indicators (e.g. viewability) rather than declining, opaque KPIs such as raw impression volume.
  4. Expect more change. This fall, Google is expected to start surfacing AMP pages more broadly in search links. Other platforms such as Twitter could start linking to AMP content more frequently. The AMP carousel could change, or new user interfaces could be introduced. Any of those events could alter the way audience engagement with AMP content is tracked and reported.

As AMP evolves from an experiment to a material factor in overall mobile audience and revenue, publishers will need an analytics framework that reflects AMP’s unique qualities and aligns with the goals AMP is meant to achieve. Relay Media is eager to contribute to that framework.

(For more nitty gritty, continue on to Part 2: Google Analytics and AMP.)