--- title: Publisher Session Depth Benchmark description: "Original benchmark data from a continuously measured publisher panel: multi-page session share bands, the 2+/3+/5+/10+ depth distribution, device and channel splits, and the nine-month erosion rate on unchanged titles." url: https://preview.artificialpoets.com/resources/session-depth-benchmark/ site: Artificial Poets type: a13s_content date: 2026-08-09T20:37:16+00:00 modified: 2026-08-18T03:30:39+00:00 author: Matías Sanchez Moises image: https://preview.artificialpoets.com/wp-content/uploads/a13s-cards/1532-social-bac696df.png --- # Publisher Session Depth Benchmark What normal looks like past page two — multi-page share bands, the full depth distribution, device and channel splits, and the rate depth erodes when nothing is done. What normal looks like past page two — measured continuously on a live publisher panel, August 2025 – May 2026. Multi-page share bands, the full depth distribution, device and channel splits, and the rate at which depth erodes when nothing is done. ## Why this benchmark exists Session depth decides how much of a publisher's shrinking external traffic gets multiplied before it leaves — and there is almost no published data on what depth actually looks like. Bounce-rate benchmarks abound; the distribution past the second page, where loyalty and ad economics live, is effectively unpublished. Every audience team we speak to assumes their numbers are uniquely bad, because they have nothing to compare against. This paper publishes the missing reference: the full session-depth profile of a multi-title special-interest publisher network we measure continuously. Sites are anonymised; windows, definitions and methodology are stated in full. **How to use it:** run the same queries on your own data (definitions below), place yourself against each table, and pay more attention to your trend than your level. ## Panel and definitions The panel is a portfolio of special-interest consumer titles under one publisher, measured event-level and continuously from August 2025 through May 2026 for cross-title comparisons. Metrics are computed per site keyed on domain, bot-filtered, on identical calendar weeks. - **Multi-page session share** — sessions with more than one pageview ÷ sessions. - **Depth distribution** — share of sessions reaching N or more *distinct article paths* (distinct matters: auto-loading and pagination inflate raw counts). - **Pre-period** — 4 Aug – 17 Dec 2025. **Post-period** — 22 Dec 2025 – 10 May 2026, bounded by panel data-integrity limits, not the calendar. - Two titles enabled the Artificial Poets Platform mid-series; the remaining titles ran unchanged and serve as the erosion baseline. Both groups are reported, separately. ## Benchmark 1 · Multi-page session share | Band | Multi-page share | Reading | | --- | --- | --- | | Healthy | ≥ 12% | Top of panel | | Typical | 9 – 12% | Most titles sit here | | At-risk | < 9% | Bottom of panel | Panel observations ran **8.6% to 12.3%** before any intervention. In plain terms: on a normal special-interest title, roughly nine sessions in ten end on the page they started on. ## Benchmark 2 · The depth distribution | Articles reached | Representative title | Pooled comparison titles | | --- | --- | --- | | 2 or more | 6.2% | 7.6% | | 3 or more | 1.9% | 2.4% | | 4 or more | 0.6% | 1.1% | | 5 or more | 0.32% | 0.60% | | 10 or more | 0.05% | 0.10% | Self-benchmarking ratios: the 3+ population is typically about **a third** of the 2+ population; the 5+ population runs **5–8%** of it. If your ratios differ wildly, check the measurement before the strategy. ## Benchmarks 3 & 4 · Depth by device and channel | Segment | Multi-page share, pre-period range | | --- | --- | | Desktop | 12.5 – 13.1% | | Mobile | 9.2 – 11.5% | | Tablet | 13.3 – 20.3% | | Direct | 7.0 – 12.7% | | Organic search | 9.6 – 11.4% | | Organic social | 8.3 – 8.6% | | Referral | 6.1 – 23.5% | Desktop starts deeper; mobile — typically two-thirds of traffic — starts shallower and erodes fastest. Social arrivals are the shallowest consistent cohort. Referral's enormous spread is a warning rather than a benchmark: benchmark that channel only against itself. A blended depth number is mostly a device-mix number. ## Benchmark 5 · The erosion rate The panel's most consequential finding. On titles where nothing changed — no redesign, no strategy shift, no new products — depth fell at every level over the nine-month series: | Measure | Change over ~9 months | | --- | --- | | Multi-page share (pooled) | −16% | | Sessions reaching 3+ | −22% | | Sessions reaching 5+ | −17% | | Sessions reaching 10+ | −13% | | Mobile multi-page share | −23% | This is the quantified version of the industry's "every visit is shallower" observation — and it means **a flat depth trend is already outperformance**. Any plan that assumes this year's pool of engaged sessions persists by default is assuming against the measured drift. ## Benchmark 6 · What intervention moved | Measure | Enabled titles | Unchanged titles | | --- | --- | --- | | Multi-page share | +46% (10.3% → 15.1%) | −16% | | Representative title, 3+ | +91% | −22% | | Representative title, 5+ | +221% | −17% | Stated plainly, as always: launch periods overstate steady state — the largest single-title pages-per-session lift ran +19.1% in its first quarter and +4.0% at run rate, both published. The deep-tail gains sit on small bases (under 2% of sessions reach five articles even after tripling). Sessions and users did not grow; the entire effect is depth on the existing audience. Verification — comparison groups, window sweeps, and a four-week serving pause that returned the metric to baseline — is published in the accompanying case-study series. ## Self-benchmark worksheet 1. Compute multi-page share, monthly, per title, per device. Place yourself in Benchmark 1's bands. 2. Run the distribution at 2+/3+/5+/10+ distinct articles. Check your ratios against Benchmark 2. 3. Trend both for at least six months. Compare your slope to Benchmark 5 — flat beats the panel. 4. Before acting on any change, freeze definitions and hold comparable properties back. The erosion baseline is what makes any improvement claim meaningful. ## Method notes Counting uses distinct article paths per session to avoid auto-load inflation. Windows are bounded by panel data-integrity limits (tracking continuity verified per site, per week; unbalanced weeks dropped). Percentages only — the panel's absolute volumes are not disclosed. The intervention figures and their verification, including everything that did *not* move, are published in the case-study series and the companion methodology articles. ## You just estimated. Want it measured? Every tool here approximates. In twenty minutes on your analytics we compute the real distribution: your single-page rate, your depth by device and channel. No deck. [Run my real numbers](/request-a-demo/)