We loaded 890 news and expert sites in a real browser and counted what each page brought with it. 868 returned a complete measurement, and every percentage here runs on those.
14
third-party domains on the median news site
197
on the single heaviest site we measured
27.3%
of sites load zero ad entities
3.8%
block you with a popup you have to dismiss
The bigger the site, the heavier the page
We accepted every cookie banner, so each page ran the way a real reader sees it. Then we counted ad entities, third-party domains, layout shift, and blocking popups. On ad weight the pattern is blunt. On popups it is not, and the correction below explains why.
Nearly half of independent sites load no ad tech at all. Among the biggest names, 18 percent do. But blocking popups are flat across every tier, and marginally rarer at the top: an indie newsletter overlay interrupts you more often than a mass-reach homepage does. The tiers below split the 890 sites by our published reach score.
One thing to read carefully in this table: the Sites column counts every source with a reach score, 890 of them, while the percentages beside it run on the 868 that returned a complete measurement. That is why a percentage here will not divide cleanly into the count next to it.
Tier
Sites (reach-scored)
Median ad entities
Median third parties
Zero ad tech
Blocking popups
Mass reach reach 8-10
190
12
39
18%
2.7%
Mid reach reach 5-7
555
2
14
26%
4.1%
Indie reach 0-4
145
1
8
45%
4.2%
39vs8
Median third-party domains per page: the mass-reach tier against the indie tier. Reading the same day's news costs five times the surveillance, depending on who wrote it.
Where the weight lives, by beat
Sports and breaking news are the heaviest reads. The lightest beats are written closest to their readers: engineering blogs and longform essays.
Sports
32.5
Breaking News
27.5
Home & Living
18
Culture & Arts
17
Food & Travel
17
Health & Wellbeing
16
Business & Money
16
Style & Beauty
15
World & Politics
14
Science & Environment
13
Tech & AI
9
Ideas & Longform
6
Makers
5
The long tail is long
Most sites are reasonable: 547 stay under twenty third-party domains. But 29 sites load a hundred or more. The heaviest brought 197 separate third-party domains along to show us one article.
We are not naming it. The point is the distribution, not the pillory.
0-9 third parties
298
10-19 third parties
249
20-29 third parties
97
30-39 third parties
45
40-49 third parties
41
50-59 third parties
30
60-69 third parties
37
70-79 third parties
17
80-89 third parties
11
90-99 third parties
14
100+ third parties
29
What the weight is made of
Split the 498 sources we publish per-source measurements for by how many ad entities the homepage loaded. The median third-party count climbs with it, from 4 to 67.
The middle half of each band clears the band below it, so the climb runs through the whole group. Most of these sites have a source page carrying their individual numbers, though not all of them do: we publish measurements for more sources than we have built pages for.
Third-party domains contacted, by ad entities on the page
498 Signal 500 sources, each homepage loaded once in a real browser, August 2026.
Ad entities
Third-party domains
0
160 sites
4
1 to 4
214 sites
13
5 to 9
49 sites
25
10 to 19
25 sites
42
20 or more
50 sites
67
020406080100
Middle half of the band, 25th to 75th percentileMedianfeeds.bar
The cleanest reads in news
The other end of the distribution deserves names. These Signal 500 sources loaded zero ad entities and near-zero third parties, with almost no layout shift. Reading them costs you nothing but attention.
Source
Pillar
Third parties
Nikita Prokopov (tonsky)tonsky.me
Makers
0
Modepilotmodepilot.de
Style & Beauty
0
Vitalik Buterin's Blogvitalik.eth.limo
Business & Money
0
LWN.netlwn.net
Tech & AI
0
EFF Deeplinkseff.org
Tech & AI
0
Rust Blogblog.rust-lang.org
Makers
0
Armin Ronacher (lucumr)lucumr.pocoo.org
Makers
0
Universe Todayuniversetoday.com
Science & Environment
0
Josh Comeaujoshwcomeau.com
Makers
1
Irrational Exuberance (Will Larson)lethain.com
Makers
1
Dan Luudanluu.com
Makers
1
Jake Archibaldjakearchibald.com
Makers
1
overreacted (Dan Abramov)overreacted.io
Makers
1
Homedithomedit.com
Home & Living
1
Apple Newsroomapple.com
Tech & AI
1
3:AM Magazine3ammagazine.com
Culture & Arts
1
Adrian Roselliadrianroselli.com
Makers
1
Momusmomus.ca
Culture & Arts
2
EleutherAI Blogblog.eleuther.ai
Tech & AI
2
Be My Travel Musebemytravelmuse.com
Food & Travel
2
Fred Hebert (ferd.ca)ferd.ca
Makers
2
Fashionistafashionista.com
Style & Beauty
2
Paul Graham: Essayspaulgraham.com
Ideas & Longform
2
LukeW (Luke Wroblewski)lukew.com
Makers
2
Method, honestly
Every number comes from the measurement pass behind the Signal 500, FeedsBar's published source ranking. Each homepage loaded in a real browser on this snapshot's date, cookie banner accepted. Then we counted ad entities, third-party domains, cumulative layout shift, and blocking popups.
Four caveats we would want stated if this were someone else's chart. Reach tiers come from our own published reach score (Tranco-based). A homepage is not every article page. Each site was measured on one page load, from one country, so a per-site number is one sample. And the 890 are the sites we curate plus candidates we scored, not a random sample of the web. Full methodology, weights, and per-source measurements: feeds.bar/signal-500/method.
Cite or reproduce any of this freely with a link back to this page. Next edition: 2027, same instruments.
Correction · September 26, 2026
This report originally said 56.5% of sites throw an interstitial before you read. The threshold behind it counted any fixed or sticky element covering more than 5% of the viewport, which describes a sticky header, not a popup. Of the 495 sites over that line, 198 had class names like header__wrapper-outer or c-parallax__img. Counting only surfaces that cover at least half the viewport and lock page scroll gives 33 sites, 3.8%. That test is stricter and still imperfect: 32 of the 33 are popups, and one is a navigation drawer, which does cover the screen and lock scrolling.
The finding reversed with the definition. The old column fell from 63% to 43% as publishers got smaller, which read as "big media interrupt you more". Blocking popups are flat, and slightly rarer at the top: 2.7% mass reach, 4.1% mid, 4.2% indie. The old number had found that large publishers ship more sticky navigation.
Percentages now run on the 868 sites carrying a complete measurement. They previously ran on the 890 carrying a reach score, so each is about 2.5% higher than before. Zero ad tech moves from 26.5% to 27.3%. Every other figure on this page is unchanged and was re-verified against the source table on September 26, 2026.
You cannot fully reproduce this page. The per-source measurements we publish cover theSignal 500 and its 498 sources. This pass covered 890, so the aggregates above cannot be recomputed from anything on this site. The distributions by ad band further up can be, because they are computed from that published file. If you want a specific figure checked, ask and we will run it and show the query.
Two caveats if you do. Nine Apple-press sources were added to the catalog on September 26, 2026 and are deliberately not part of this edition. And the stored interstitial value is a fraction of the viewport, so reading it as a percentage is the unit confusion behind the error above.
This research exists because FeedsBar only carries the clean end of this chart. The whole catalog is free to take: the Signal 500 as OPML.