How Long Does a Story Stay "Most Read"? A Q2 2026 Look Across 9 News Sites
By Rene Roost | Published: July 13, 2026
This is the first quarterly overview from mostreadnews.com, and the anchor of a short series: this page looks across every source at once, and separate vertical deep-dives look closely at the stories within each.
Every ranking on this site derives from one signal: real readers clicking real articles, as reflected in each outlet's own Most Read list. I've been recording these lists continuously since late 2025, so this first report is overdue — the first half of 2026 went into building the site itself. How the recording and scoring work is documented in full on the methodology page.
Key findings
- Across nine major news sites, the median article holds a spot on a Most Read list for 5 hours — half of all charting articles are gone that fast. Only one charting story in ten survives past 23 hours — barely a day.
- Tech publications are the exception. On The Verge the median is 22 hours — up to 4.4× the ecosystem baseline. Ars Technica sits at 16 hours on the strictest count. However, some of its record is reconstructed, and filling those short gaps raises that figure to 23 hours, so read it as roughly 16–23.
- The fastest-moving lists belong to CNBC and The Guardian, where the median article lasts just 3 hours and a list slot turns over 8 times a day.
- "Longest-lived" depends on how you measure it. By total time visible anywhere on a list, the champion was a Business Insider human-interest story that lingered for just over eight days (194 hours), most of it in the lower slots. By time actually spent at #1, it was a tech story on The Verge, which held the top spot for 79 hours — more than three days leading.
The full picture: nine sources compared
The table below counts only directly recorded readings.[1] Sources are ordered by how long their typical article survives — longest-lived at the top. The List size column is the number of slots each publisher shows on its own Most Read list.
| Organization | List size | Median Duration (Hours)[2] | 90th Percentile Duration (Hours)[3] | Turn Rate (Per 24h)[4] |
|---|---|---|---|---|
| The Verge | 5 | 22.0 | 45.0 | 1.09 |
| Ars Technica | 5 | 16.0 | 38.0 | 1.50 |
| Business Insider | 10 | 16.0 | 57.0 | 1.50 |
| Al Jazeera | 10 | 7.0 | 23.0 | 3.43 |
| AP News | 5 | 6.0 | 16.0 | 4.00 |
| BBC | 10 | 6.0 | 17.0 | 4.00 |
| Forbes | 10 | 5.0 | 32.0 | 4.80 |
| CNBC | 5 | 3.0 | 15.0 | 8.00 |
| The Guardian | 10 | 3.0 | 14.0 | 8.00 |
| All 9 sources, pooled | — | 5.0 | 23.0 | 4.80 |
Two sources fall below 90% directly recorded readings: Ars Technica (71.0%) and AP News (67.7%). Including reconstructed readings raises their medians to 23.0 and 10.0 hours. The other seven sources keep the same median under both methods; two of them (BBC and Forbes) edge up only at the 90th percentile, by about an hour and three hours respectively. See Data completeness.
The spread is the story. A reader who checks CNBC or The Guardian twice a day, at morning and at evening, sees an almost entirely fresh list each time. A reader who checks The Verge once a day will often see the same lead story two or three days running. The two behaviours are produced by the same underlying signal — reader clicks — filtered through very different ranking mechanics.
Why the gap? Two mechanisms
Publishers treat these ranking calculations as trade secrets, so what follows is inference, not disclosure. Two mechanisms could plausibly account for the gap — and the list-size data lets me test one of them directly.
1. Ranking windows. The median article across all nine sources holds a spot for 5 hours. On Ars Technica that figure is 16 hours and on The Verge 22 — three to over four times longer. A plausible reading is that the tech outlets score articles over a longer lookback period, which would dampen short-term volatility and let a single story hold its slot across multiple days. That's a hypothesis the data is consistent with, not a conclusion it establishes.
2. List length. The obvious suspect is how many slots each list shows. Five of these sources publish a 10-article list; four — Ars Technica, The Verge, CNBC and AP News — show only five. Fewer slots should mean each surviving article occupies scarcer space and stays visible longer, so a 5-slot list ought to retain stories longer than a 10-slot one.
The data doesn't bear that out. Two pairs make the point. CNBC shows five articles and is tied for the fastest turnover here — a 3-hour median, a slot refreshing eight times a day; The Guardian shows ten and turns over at exactly that same rate. Ars Technica (five slots) and Business Insider (ten slots) each hold their typical article for an identical 16 hours. If slot count drove residency, these pairs would diverge; instead they match. The four 5-slot lists on their own span the entire range of the dataset, from CNBC's 3-hour median to The Verge's 22. A short list is perfectly capable of being the fastest-moving one I track.
So list length is not the explanation. Once it's ruled out, what remains is how each publisher scores its list — the lookback window, eligibility rules and refresh logic I can't see from outside. Topic offers no clean shortcut either: CNBC and Business Insider both cover business, yet CNBC ties for the fastest turnover of all nine sources while Business Insider sits at the slow end, tied with Ars Technica at 16 hours — two outlets on the same beat, at opposite ends of the range. Whatever separates a fast list from a slow one lives inside each publisher's ranking process, not in anything visible from the list itself. What the data does now settle is the narrower question: the gap is not an artefact of how many slots a list shows.
Read the vertical deep-dives
This overview stays at the ecosystem level. The stories themselves — which articles held #1 longest, how they rose and fell, and which kinds of stories each audience kept coming back to — are covered in the vertical reports:
- Tech & Science — Q2 2026 — Ars Technica and The Verge. Available now.
- World News — Q2 2026 — Al Jazeera, BBC, The Guardian. Coming later this quarter.
- Business & Finance — Q2 2026 — Business Insider, CNBC, Forbes. Coming later this quarter.
Data completeness
The main table above counts only readings my scraper recorded directly. Where readings were missed, short gaps are reconstructed by interpolation; I keep those out of the headline numbers and report, per source, how much of each record they would affect. Two different things can weaken a record, and they're worth separating. Coverage is how much of the quarter I was recording a given source at all. Directly recorded is, of the readings I did take, how many were captured first-hand rather than reconstructed for a short gap (the full procedure is on the methodology page). A source can be strong on one axis and weak on the other. The corpus is every article published between 1 April and 30 June 2026 (UTC), matching what the site's own by-source filters display. Only sources that were being tracked from the start of the quarter are listed; the whole dataset runs to 142,252 positional observations, where an observation is one article seen in one slot at one reading.
| Source | Coverage | Directly recorded | Reconstructed |
|---|---|---|---|
| Al Jazeera | 97.21% | 99.76% | 0.24% |
| The Guardian | 97.12% | 99.67% | 0.33% |
| BBC | 96.98% | 94.17% | 5.83% |
| Forbes | 96.66% | 90.56% | 9.44% |
| Business Insider | 96.34% | 99.74% | 0.26% |
| The Verge | 96.29% | 99.62% | 0.38% |
| CNBC | 95.47% | 99.78% | 0.22% |
| Ars Technica | 92.90% | 71.01% | 28.99% |
| AP News | 72.12% | 67.72% | 32.28% |
| All sources | 93.45% | 94.07% | 5.93% |
Eight of the nine sources were tracked across at least 92% of the quarter. AP News is the exception at 72%: for over a quarter of the period I have no readings for it at all. Those missing stretches are gaps too long to reconstruct — I only fill gaps of up to three consecutive missed readings and leave anything longer blank — so they never enter the reconstructed column. They show up here instead, as reduced coverage. That is why AP News carries the widest uncertainty in the report: its figures rest on a thinner and less evenly distributed record than the rest.
Directness is a separate story. Ars Technica sits at 71% directly recorded — the next-lowest after AP News — yet its coverage is a healthy 93%: it was tracked steadily all quarter, but with many short gaps scattered through it, each one short enough to reconstruct. Its uncertainty comes from interpolation, not absence. AP News is the harder case: at 68% it has the lowest directly recorded share of any source, and its coverage is thin too — weak on both axes at once.
For the other seven sources both figures are high, and their medians are identical whether or not reconstructed readings are counted — two of them, BBC and Forbes, nudge up only at the 90th percentile. I've published Ars Technica and AP News alongside them because the reconstruction is bounded — long gaps are never filled — and every interpolated reading is flagged in the data, but both, and AP News especially, should be read as carrying wider uncertainty than the rest. Full detail is on the methodology page.
Notes
Corrections: errors are fixed in place with a dated note at the foot of this page. Spotted something wrong? Get in touch.
A directly recorded reading is one my scraper actually took. Where readings were missed, gaps of up to three consecutive readings are reconstructed by interpolation. Counting only direct readings slightly undercounts time on the list (unobserved hours aren't counted); counting reconstructed readings may overcount. Of the nine sources, only Ars Technica and AP News shift at the median between the two methods; BBC and Forbes move only at the 90th percentile, and only slightly. ↩︎
Median Duration — the median cumulative hours an article stays visible anywhere on a publication's Most Read or Most Popular list. ↩︎
90th Percentile Duration — the duration threshold only the top 10% longest-lasting articles cross. ↩︎
Turn Rate (per 24h) — how many times a single slot refreshes with new content in a day; derived as 24 ÷ median duration, so it restates the first column rather than measuring something new. ↩︎