---
title: Nobody Measures Past Page Two
description: Industry benchmarks stop at the second pageview — bounce rate and pages per session say nothing about the readers who go five articles deep. The depth report almost no publisher runs, with measured numbers to compare against.
url: https://preview.artificialpoets.com/blog/nobody-measures-past-page-two/
site: Artificial Poets
type: post
date: 2026-08-10T03:32:09+00:00
modified: 2026-08-14T18:30:48+00:00
categories:
  - Measurement
---
# Nobody Measures Past Page Two

## Why the standard metrics stop at page two

Bounce rate was designed as a landing-page quality check: did the visit produce a second interaction, yes or no. Pages per session was designed as a volume ratio. Both were built for a question the 2010s cared about — *is traffic engaging at all* — and both compress everything past the second page into invisibility.

The compression matters because the value past page two is not evenly distributed. A reader on their fifth article is not "2.5× a two-page reader" — they are in a different behavioural state: habit-forming, subscription-plausible, newsletter-receptive. Chartbeat's return-rate data shows the fork at one extra page (8% → 22% week-one return probability)[[1]](#ref-1); the states beyond it are worth proportionally more, and the standard metrics cannot see them at all.

Meanwhile the average actively misleads. Pages per session on the panel we measure runs between about 1.10 and 1.35 — numbers that look nearly identical — while the share of sessions reaching five articles varies by **multiples** between titles and periods underneath them.

## The report to run instead

One query, four rows: the share of sessions reaching 2+, 3+, 5+ and 10+ **distinct articles** (distinct matters — auto-reloads and pagination inflate raw pageview counts). Trended monthly, split by device.

Here is what it produced on one measured title, before any intervention:

| Articles reached | Share of sessions |
| --- | --- |
| 2 or more | 6.2% |
| 3 or more | 1.9% |
| 5 or more | 0.32% |
| 10 or more | 0.05% |

And here is why trending it matters more than the level: on comparison titles where nothing changed for nine months, every row fell — **−22% at 3+, −17% at 5+, −13% at 10+**. The erosion the industry describes as "shallower visits"[[2]](#ref-2) is real, measurable, and completely invisible in bounce rate, which moved by a rounding error over the same period.

## What you learn once you can see past page two

Running depth as a distribution rather than an average changed three conclusions for us.

**The deep tail responds differently than the shallow one.** When contextual recirculation was enabled on measured titles, the 2+ row moved +62% — and the response *grew* with depth: +91% at 3+, +221% at 5+, +443% at 10+ (small bases, printed with the result). An average would have reported "pages per session up a bit". The distribution showed a compounding mechanism.

**Continuation has a shape.** Per-position telemetry on served articles shows the real curve: about a third of readers accept a second article; half of those accept a third; past the fourth article, continuation runs **70–82%** per step, with attention settling around 19 seconds per article rather than decaying to zero. The expensive step is the first hand-off. Everything after it rides momentum — which is an argument about where to spend product effort that no aggregate metric could have made.

**Device tells you where the ceiling is.** Depth distributions differ sharply by device — desktop readers start deeper, mobile has the most headroom and, on unchanged titles, the fastest erosion. A single blended number hides exactly the segment where the problem and the opportunity are largest.

## The benchmark question

"What's a good pages per session?" is the most common question in publisher analytics and close to meaningless as asked. The usable version is distributional: on the panel we measure, multi-page share before any intervention ranged **8.6% to 12.3%** across titles — call under 9% at-risk, 9–12% typical, above 12% healthy — with the 3+ row typically a third of the 2+ row and the 5+ row around a twentieth of it.

Your own numbers will differ by category and traffic mix, which is the point: benchmark the shape once, then benchmark against yourself, monthly, with the instrumentation frozen. Any depth series is only as honest as its measurement — infinite scroll inverts scroll-depth percentages and raw pageview counts inflate under auto-loading, which is why the distribution counts *distinct articles* and why we published the [measurement traps](#) separately.

- Standard metrics see two states; the distribution on a measured title: **6.2% / 1.9% / 0.32% / 0.05%** at 2+/3+/5+/10+
- Left alone, every row eroded: **−22% / −17% / −13%** over nine months — invisible in bounce rate
- Under intervention the response compounded with depth: **+62% → +443%**
- Continuation past the fourth article: **70–82%** per step, attention settling ~19s/article
- Panel multi-page bands: **<9% at-risk · 9–12% typical · >12% healthy**

## FAQ

Bounce rate reports one bit: second page or not. It cannot distinguish a two-page site from a five-article habit machine, and on the panel it barely moved while the deep rows fell by double-digit percentages. It is a landing-page metric wearing an engagement costume.

Any event-level stack can: count distinct article paths (or content IDs) per session, bucket sessions at ≥2/≥3/≥5/≥10, divide by total sessions, trend monthly. The discipline is using *distinct* content units and freezing the definition before you start comparing periods.

On measured titles, holding the distribution flat already beats the market — unchanged properties declined at every depth. Beyond that, the measured intervention range on the 2+ row was +29% to +63% in launch periods, settling lower at steady state; a plan built on the settled figure is the honest one.

## Your four rows, in thirty seconds

The Session Depth Assessment estimates your distribution from two analytics inputs and places it against the panel bands in this article.

## References

1. Chartbeat, “Encourage multiple page visits, use social video for successful 2026 publishing strategy.” [chartbeat.com](https://chartbeat.com/resources/general/successful-2026-publishing-strategy/)
2. Nitropay, “Protecting Ad Revenue Per Session in the Age of AI Search.” [blog.nitropay.com](https://blog.nitropay.com/protecting-ad-revenue-per-session-in-the-age-of-ai-search/)

## 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/)