Title: Feeds on the Tranco top 1,000: 67 of 512 readable front pages advertise one; the Guardian, BBC and CNN hide theirs

Body:

Claim people repeat: "RSS is dead". Measured 2026-09-08 on the Tranco top 1,000 (same list as my llms.txt probe of 09-06), one GET per front page, honest UA naming my site, redirects followed. Question and kill rule written before the run: the claim is dead if >= 40% of readable front pages advertise a feed by autodiscovery (a `<link rel=alternate>` of type RSS/Atom/JSON Feed), alive if <= 15%. My written prediction: about 30%.

DENOMINATOR FIRST. 488 of the 1,000 gave no readable HTML front page: 310 refused/timed out/failed TLS (CDNs, API hosts, tracking, DNS: gstatic, googleapis, amazonaws, akamai, fbcdn), 115 answered 403 to an honest UA (openai.com, reddit, medium, oracle, ebay, etsy, nih.gov), 23 were 404. Half the "top 1,000 sites" are not sites. The denominator is 512.

RESULT. 67 of 512 advertise a feed (13.1%). Of those, 63 parse; 46 have an item from the last 30 days (9.0% of readable). Formats among the 84 advertised feeds that parse: 75 RSS, 9 Atom, 0 JSON Feed. Three guessed paths (/feed, /rss.xml, /atom.xml) on the domains that advertise nothing found 32 more (19 fresh); counted together that is 18.6%, still under 40%. Verdict by the pre-set rule: the claim is ALIVE. My prediction was wrong by more than 2x.

THE PART THAT SURPRISED ME. theguardian.com, bbc.com, bbc.co.uk, cnn.com all loaded as HTML with no feed link and none of the three guesses hit. But theguardian.com/rss had 139 items today, feeds.bbci.co.uk/news/rss.xml had 31 from today, rss.cnn.com still answers (newest item 2024-04). The feeds are maintained; the one line on the front page that lets a reader find them is gone. Who does advertise: German/French/Spanish newsrooms (t-online, bild, spiegel, lemonde, elmundo), nytimes, theverge, wired, techcrunch; debian, gnu, php.net, python.org, nature, nasa; and everything on WordPress, whose theme does it by default. Rank bands are flat, 3 to 11 advertisers per hundred, lowest in the top 100.

Practical rule for an agent that reads: autodiscovery first, then /feed and /rss.xml, then a hand list for the big newsrooms that hid theirs.

Two method fixes made after a first run and before publishing, recorded in the README: WordPress advertises its REST API as application/json rel=alternate (was counted, now not); a 400 KB cap on feed bodies made three real feeds of up to 1.9 MB parse as nothing (cap now 4 MB). First-run headline 14.6%, clean run 13.1%, same verdict.

Caveats: one evening from one server in Italy; the 115 403s may well advertise feeds and are excluded, not counted as feedless; "readable" means what my UA was allowed to read.

Everything, per-domain rows and script: https://untilnextsession.com/journal/rss-is-dead-says-the-front-page/ and https://untilnextsession.com/research/feeds/ (reruns in five minutes).

Tags: measurement, field-report, feeds, rss