Google Analytics and AMP
This is the second part of a series on AMP audience measurement; see Part 1: What’s Going On With AMP Analytics?
Here’s a summary of how Google Analytics records certain common metrics, and how those metrics may register differently for AMP content vs. standard web content. While the “How it works” sections are specific to a standard out-of-the-box Google Analytics implementation, the general concepts may be applicable to other analytics services with similar methodologies.
HOW IT WORKS: A Client ID (cookie) recognizes returning browsers. In general, this method is inexact when users access content from multiple devices and browsers. Unless a publisher enables User ID with logged-in authentication, cross-device, cross-domain and cross-browser user tracking is challenged in general (not just with AMP).
- Overcount of users.
WHY: AMP pages are often cached and delivered by third party domains; for example articles in the Google AMP Carousel are delivered by cdn.ampproject.org. The portability of AMP pages is a game-changer for frictionless content distribution, but contributes to the existing challenge of cross-domain audience tracking. Because Client ID cookies are domain-specific, a user who visits a publisher’s AMP content on multiple domains won’t be recognized as the same user.
Additionally, a user who visits AMP and non-AMP pages on a publisher’s own domain will be counted as two users, as the Client ID cookie is stored differently for AMP vs. non-AMP pages.
HOW IT WORKS: Google Analytics records a session as ongoing interaction with a website. A session ends: (a) when the user exits; (b) after 30 inactive minutes; (c) at midnight; or (d) with a campaign change. Sessions are subject to the same cross-device and cross-browser limitations as Users.
- Overcount of sessions.
- Undercount of views per session.
WHY: In a typical experience, a user discovers a publisher’s AMP article in the Google AMP Carousel (on Google’s domain) and then clicks an in-article link to a second AMP article on the publisher’s own domain or another domain. Even though the user was continuously engaged with the publishers content, this activity would register as two one-page sessions on each domain.
If a user encounters multiple articles from a publisher in the Google AMP Carousel in the same active period, those pages will be counted as a single session on the cdn.ampproject.org domain.
Single-Page Sessions, Bounce Rate
HOW IT WORKS: Single-page sessions are sessions in which only one page was viewed. A bounce session is a single-page session with no engagement events. Bounce rate is the percentage of bounce sessions.
- Large number of single-page sessions.
- Appearance of high bounce rate.
WHY: AMP content is likely to register more single-page sessions and bounce sessions for two reasons:
1) A user session traversing multiple domains can look like a series of single-page sessions on each domain.
2) The interface of the Google AMP Carousel 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 a single-page session is registered in Google Analytics — unless the user encounters another article from the same publisher. If no engagement events were registered, the session is a bounce.
HOW IT WORKS: Session duration is the time from the first page load to the exit page load. Time spent on the exit page is excluded unless the user engages with the exit page — in which case only the time from page load to the final engagement is counted for that page. Single-page sessions with no engagement will register 0 time spent.
Average time on page is calculated as total time spent divided by total views, omitting the exit page.
Average duration is the total duration of all sessions divided by the number of sessions.
- Large number of 0-second sessions.
- Inflated average time on page.
- Deflated average session duration.
WHY: The Google AMP Carousel swipe interface results in a large number of single-page sessions — either with 0 time spent, or some time spent when engagement occurs. These single-page sessions can skew duration calculations:
1) Average time on page may be higher than expected due to a large number of single-page visits with some engagement, which results in more time being added to the numerator but no pages added to the denominator since exit pages are not counted.
2) Average session duration may be lower than expected due the large number of sessions with no time spent.