Dwell time: measuring engagement without official data

Dwell time never existed as an official metric in Google Analytics, yet half the agency reports still quote the number as if it were exact. The classic definition is the interval between the SERP click and the user returning to Google, something only Google itself can measure. What GA4 gives you are proxies: engagement time, scroll depth, interaction events and return cohorts. Treating those variables as combined signals rather than magic substitutes is what separates honest analysis from dashboard theater. Before measuring anything, revisit how GA4 for SEO: the reports that actually replace Universal Analytics reorganized what is actually observable.

Start with engagement_time_msec, the event GA4 fires while the tab is in focus. That counter pauses when the user switches tabs, minimizes or idles for more than five seconds, so it already drops most of the noise that used to inflate avg session duration in Universal Analytics. An 1,800 word post with average engagement under 45 seconds almost always has a reading problem, not a traffic problem. Cross that number with average position in Search Console and you can split pages that rank on inertia from pages that actually deliver. The next step is checking Search Console: 7 underused reports and what to extract from them to validate whether the query carries the right intent.

Scroll depth is the second most underrated proxy. Wiring up events at 25, 50, 75 and 90 percent through Google Tag Manager takes fifteen minutes and answers a question engagement time alone cannot: did the reader reach the end, or bail at the third paragraph? In tests I ran with technical blog clients, pages where over 40 percent of users hit 75 percent scroll converted to newsletter 2.3 times more than pages with the same traffic and shallow scroll. When scroll collapses before the main H2, the problem is usually the title promise, a topic I unpack in Title tags that convert: 7 patterns tested on real SERPs.

Microinteraction events round out the trio. Clicks on the table of contents, accordion expansions, embedded video plays, code block copies, sustained hovers on tables: all of those are active reading signals. In BigQuery, a single query grouping user_pseudo_id by session_id and counting distinct events is enough to build a per-session engagement score. Anyone keeping the GA4 export on for six months already has the sample needed to correlate score with organic return. This kind of analysis rarely leaves a standard agency, as I argued in BigQuery + GSC: queries your agency won't run.

Return cohorts are the long-horizon proxy almost nobody looks at. If a user lands through organic, leaves, and comes back direct or via brand search within seven to thirty days, that is the strongest possible signal that the content did its job. In GA4, the weekly cohort retention report shows exactly that once you filter by first_user_medium equal to organic. On an ecommerce audit I ran, PDPs with 14 day return above 8 percent converted 5x better downstream, even when the first session was short. A short session is not always bad, which is why Honest SEO KPIs: beyond rankings and traffic keeps pushing back on isolated metrics.

Combine the four signals into a single index before you conclude anything. I run a weighted average: 35 percent engagement time normalized by word count, 25 percent scroll 75, 20 percent microinteractions per session, 20 percent 14 day return. Reweight by page type, because PLP, PDP and blog behave differently, as covered in E-commerce on-page: PLP vs PDP without cannibalization. Practical takeaway: stop asking GA4 for dwell time. Build your composite proxy, document the formula inside the dashboard itself, and revisit it quarterly. Honest metrics are not the ones that look precise, they are the ones you can defend when the CFO asks where the number came from.