--- title: Your Scroll Depth Metric Lies When You Run Infinite Scroll description: Scroll depth is a percentage of a document that grows as readers scroll. On infinite-scroll pages the metric falls exactly when engagement rises. The math, a real example from live publisher data, and what to measure instead. url: https://preview.artificialpoets.com/blog/scroll-depth-infinite-scroll/ site: Artificial Poets type: post date: 2026-08-10T03:32:09+00:00 modified: 2026-08-14T18:30:48+00:00 categories: - Measurement --- # Your Scroll Depth Metric Lies When You Run Infinite Scroll ## What does scroll depth actually measure? Scroll depth is the maximum scroll position a reader reached, divided by the total scrollable height of the document, expressed as a percentage. That denominator is the entire problem. The metric was designed for a world where a page's height is fixed at load time: a reader who reaches the end of the article scores 100%, a reader who bounces at the headline scores near zero. On a static page, the number means what everyone assumes it means. Analytics tools report it at percentile thresholds — 25%, 50%, 75%, 90% — and publishers read those thresholds as a proxy for "did they actually read it." For static pages, that reading is fair. ## What happens to scroll depth on an infinite-scroll page? The document grows as the reader scrolls, so the denominator grows with the reader — and the percentage moves down precisely when engagement moves up. Walk through the arithmetic with one reader and a 2,000-pixel article: | Reader behaviour | Static page | Infinite-scroll page | | --- | --- | --- | | Reads the full article, leaves | 2,000 ÷ 2,000 = **100%** | A second article loaded below. 2,000 ÷ 4,000 = **50%** | | Reads the article *and* the next one | n/a — no next one | A third loaded. 4,000 ÷ 6,000 = **67%** | | Reads four full articles | n/a | A fifth loaded. 8,000 ÷ 10,000 = **80%** | The same reader, behaving strictly better each row, never gets back to the score they would have had for reading a single page on a static site. The best readers on an infinite-scroll site are mathematically prevented from scoring 100% — there is always more document below them. Aggregate that across every session and the site-level average does something perverse: **the more successful the scroll experience is, the further average scroll depth falls.** ## We watched it happen on live publisher sites On two titles inside a multi-title special-interest network we measure, average max-scroll fell from roughly **48% to 29–33%** after a continuous-scroll experience was enabled — over the same window in which distinct pages per session rose **+12.2% and +19.1%**, and sessions reaching five or more articles roughly **tripled**. Readers were reaching more articles, spending more time in served content, and going measurably deeper into the site. Every independent metric said engagement rose. Scroll depth said it collapsed. If we had evaluated the deployment on scroll depth, we would have concluded it failed. If a publisher evaluates *any* continuous-scroll product on scroll depth — including ours — they will conclude it failed, no matter how well it worked. The metric is not noisy in this situation. It is inverted. This is not unique to infinite scroll. Anything that changes document height after load — lazy-loaded galleries, expanding modules, "read more" reveals, comment sections that hydrate late — moves the denominator and distorts the percentage. Infinite scroll is just the extreme case, because it moves the denominator *continuously and on purpose*. A vendor's own documentation should tell you this. Ours does — our measurement docs flag scroll depth as distorted under infinite scroll and say not to read those sessions on document percentage. If a vendor is selling you a scroll product and quoting scroll-depth gains, ask them to explain the denominator. It is a short conversation. ## Which metrics survive infinite scroll? Count things whose definition does not depend on document height. Three replacements cover what scroll depth was trying to tell you. **Distinct articles per session.** Count unique article paths a session touched, not pixels travelled. It answers the question scroll depth was really being asked — *did they read more than the thing they landed on?* — and it is immune to how the pages were stitched together. This is the metric we publish our own results in. **Per-article time.** Instrument each served article with its own timer — time from when the article enters the viewport to when the reader moves past it. On our own deployments this telemetry shows a median of about **50 seconds** of added reading per engaged session, roughly **30 seconds per served article** — numbers a document-relative metric cannot produce at all. **Reading-stage milestones per article.** Fire discrete events per article unit — started, engaged, finished — with the article as the denominator, not the page. 90% of *an article* means something; 90% of *an endless document* does not. The pattern in all three: make the article the unit of measurement, because on a continuous-scroll site the page stopped being one. ## How to check your own reports 1. **Find the discontinuity.** Pull average scroll depth as a weekly series and mark the date any scroll or lazy-load experience shipped. A step *down* at that date, with pages per session flat or up, is the artifact — not a reader problem. 2. **Split by template.** Compare scroll depth on fixed-height templates (about pages, landing pages) against article templates running continuous scroll. If the gap is large and appeared when the experience shipped, you are measuring geometry, not attention. 3. **Cross-check against a height-independent metric.** Distinct pages per session, return rate, or per-article time. When scroll depth disagrees with all of them, the disagreement *is* the finding: your denominator moved. For the same read on your own numbers, the [Session Depth Assessment](#) estimates your multi-page and depth rates from two inputs, benchmarked against a measured publisher panel. - Scroll depth = max scroll ÷ **document height** — the denominator grows under infinite scroll - A reader who finishes one article scores **100%** on a static page and **~50%** on an infinite-scroll page - Live panel: avg max-scroll fell **~48% → 29–33%** while distinct pages/session rose **+12–19%** - Sessions reaching 5+ articles roughly **tripled** over the same window the "depth" metric collapsed ## FAQ No — it is a fixed-height metric. On static templates it still says what it always said. It becomes misleading the moment document height changes after load, and inverted under infinite scroll. Under dynamic document height there is no meaningful benchmark, because two sites with identical reader behaviour can report wildly different percentages depending on how their pages grow. Benchmark distinct articles per session instead — it compares cleanly across sites. Any tool computing scroll as a percentage of document height has the same arithmetic, whatever the vendor. The fix is not a different tool; it is a different denominator — per-article events instead of per-document percentage. ## See what your session depth looks like We will read your analytics with you for twenty minutes and tell you what share of your sessions stop at the first page — and whether your scroll metrics are measuring readers or geometry. No deck. ## See what your session depth looks like We read your analytics with you for twenty minutes and tell you what share of your sessions stop at the first page. You leave with the annotated read, whether or not we ever talk again. 20 minutes, your analytics, no deck. [Book my session-depth read](/request-a-demo/)