RSS 2.0 vs Atom: Why 88% of Feeds Are Already Both
Pick whichever one your CMS already produces. In the Proofite census of 23 August 2026 — every distinct feed URL actually subscribed to by our users, fetched and classified in a single pass — 93% of feeds declare <rss version="2.0">, 5% are Atom 1.0, 1% still use the RDF syntax of RSS 1.0, and 1% are no longer feeds at all. But 88% of those RSS 2.0 feeds already declare the Atom namespace, and 89% declare Dublin Core. Eight feeds out of nine are both things at once. The question "RSS or Atom" assumes a fork in the road that stopped existing years ago.
What matters is which elements your generator emits. Not the label on the root tag.
Atom won the standards argument and lost eighteen to one
Atom is the only one of these formats that went through a standards body. It has an RFC — 4287, published December 2005 — a working group, a registered media type, and a data model that a careful engineer would actually design. Dates in one format only. Language declared per element. An explicit distinction between plain text and HTML in content.
RSS 2.0 has none of that. It has never passed through any standards organisation. version="2.0" covers eleven separate revisions of the specification published between 2002 and 2009, all shipping under the same declared number.
Ratio in the wild: 18 to 1 against Atom.
That gap has nothing to do with technical merit and everything to do with what publishing software emits by default. Which brings us to the interesting part.
The format war ended in a merger
88% of RSS 2.0 feeds declare xmlns:atom. Almost always for one element: <atom:link rel="self">.
That element does something RSS 2.0 cannot do on its own — it lets a document state its own canonical address. If an aggregator receives your feed by some indirect route — a proxy, an email forward, a cached copy — rel="self" is how it learns where to go for the next fetch. RSS 2.0 has no native way to say that. So publishers borrowed the Atom element and dropped it into an RSS container.
This is the actual outcome of the format war. Not a victory. A merger, with RSS 2.0 as the wrapper and Atom as a parts bin.
RSS 1.0 lost as syntax and won as vocabulary
Only 1% of subscribed feeds use RSS 1.0's RDF syntax. By that measure the RSS-DEV Working Group — an independent group, incidentally, not the W3C, as it is often miscredited — lost decisively.
Then look at the namespace table. Dublin Core, the vocabulary RSS 1.0 adopted, appears in 89% of RSS 2.0 feeds. dc:creator for the byline. dc:date for the timestamp. The losing branch of the December 2000 split survives, dismantled into components, inside the format that beat it.
The full picture across all valid feeds, Atom and RDF documents included: dc 90%, atom 82% (the prefix, which Atom 1.0 feeds have no reason to declare), content 75% (content:encoded, the full article body), sy 49% (update frequency hints), slash 48% (comment counts), wfw 46% (per-article comment feeds), media 35% (Media RSS: thumbnails, durations, licences), itunes 7%.
What looks like an editorial choice is a factory setting
Read those top six again: dc, atom, content, sy, slash, wfw.
Those are exactly the namespaces WordPress emits out of the box, in roughly that order of usefulness. The shape of the modern feed ecosystem is a default configuration file, replicated a few hundred million times. Nobody sat in a meeting and decided that Slashdot's comment-count extension deserved to appear in half the world's feeds.
This is why "which format should I use" is a badly framed question. For most publishers the format was chosen in 2003 by somebody else, and the choice is fine.
The MIME type is wrong in more than a third of cases
63% of valid feeds are served with a correct media type — 61% application/rss+xml, 2% application/atom+xml. The other 37% are not: application/xml at 22%, text/xml at 14%, and one feed served as text/html.
The historical reason is neat. application/rss+xml was never registered with IANA. Two attempts went in as Internet Drafts and both lapsed, because RSS had no recognised body behind it to push the paperwork through. application/atom+xml is registered properly, via RFC 4287. The format nobody uses has the media type everybody borrows.
Does the wrong header break anything? For readers, almost never — parsers sniff the document, not the header. The real consequence is in browsers. Hand a browser application/rss+xml and it will often download the file instead of displaying it. Some publishers fall back to text/xml on purpose for exactly this reason, accepting a technically sloppier header in exchange for a feed URL a human can click.
One more number from the same fetch: 2% of subscribed feeds are still served over plain HTTP.
Podcasting is the part of RSS that never wobbled
21% of subscribed feeds carry at least one <enclosure>. 7% declare the iTunes namespace, which is the narrow definition of a podcast.
<enclosure> entered RSS 0.92 in December 2000 — proposed by Tristan Louis, pushed by Adam Curry — and sat there with no audience for three years. Then it became the transport layer for an entire industry. Apple, Spotify and Overcast do not host episodes. They index public RSS feeds. Every time a show moves hosts without losing subscribers, that is RSS doing the one job it does perfectly.
Yes, people still use RSS, and here is what their setup looks like
A typical active Proofite reader follows 70 sources: 30 RSS feeds, 20 saved web searches, 19 newsletters, one social profile, spread across eight topic boxes. RSS is the largest single category.
It also behaves better than the alternatives. On 40 real newsletters analysed in July 2026, 12 (30%) carried at least one declared sponsored block — "Presented by", "Together with", "(SPONSOR)". In one PetaPixel edition the ad ran 959 characters, 18% of the usable text. In TLDR AI, 624. Feeds don't do that to you, or at least not as often.
The duplication problem is real in both. Across four generated digests covering 37 items, 27% of items came from two or more outlets reporting the same event: a funding round arriving via Coindesk and Wired at once; an AI security incident arriving via Import AI and TLDR AI within hours. Following more feeds does not automatically mean knowing more.
The failure mode that never shows up as an error
A feed that returns 200, parses cleanly, and whose newest item is eight months old is a dead source. It generates no warning anywhere. It reads as "no news".
That is the worst thing that can happen to a reader, and it is invisible. Check the date of the most recent item, not the absence of new items. Past six months, assume the source is gone.
The checklist
- Look at the first tag. If it isn't
<rss,<rdf:RDFor<feed, you are not publishing a feed. That single test is what disqualified the 1% of subscribed URLs now returning HTML pages or nothing. - Add
<atom:link rel="self">. Without it, an aggregator that receives your document indirectly has no way to find it again. - Give every item a
<guid>, withisPermaLink="true"when it matches the article URL. This is what stops a fixed typo in a headline from reappearing as a new item in everyone's reader. - Put autodiscovery in
<head>:<link rel="alternate" type="application/rss+xml" title="RSS" href="...">. Formalised by the RSS Advisory Board in 2006 and still the only way most tools find your feed. - Run it through validator.w3.org/feed. It flags things that don't break parsing but degrade it — ambiguous date formats, mainly.
And if you are writing a generator from scratch, with no CMS defaults to inherit: use Atom. Cleaner data model, one date format, a media type that IANA actually recognises. You will be in the 5%, and every reader worth using will handle you fine.
Frequently asked questions
Which RSS feed format is best?
For a publisher, the one your CMS already emits — every widely used reader parses RSS 2.0, Atom and RSS 1.0. For a new generator written from scratch, Atom 1.0: a single date format, per-element language, an explicit text/HTML distinction, and a media type registered with IANA. In practice 93% of subscribed feeds are RSS 2.0, and 88% of those have already borrowed Atom elements anyway.
Is RSS still used in 2026?
Yes. Among active Proofite users, RSS is the largest source category: 30 feeds in a typical 70-source setup, ahead of saved searches (20) and newsletters (19). Separately, 21% of subscribed feeds carry an <enclosure>, which is the mechanism the entire podcast industry runs on.
Is RSS 1.0 newer than RSS 2.0?
No. This is the most common confusion in the whole subject. RSS 1.0 was published on 6 December 2000 by the RSS-DEV Working Group; RSS 0.92 followed nineteen days later from UserLand; RSS 2.0 arrived in August 2002. They are separate branches with incompatible syntax, born from a split. The higher number is not the later version.
What does version="2.0" actually mean?
"UserLand branch", not "second edition". Eleven revisions of the specification were published between 2002 and 2009 without the declared version number changing once.
Is Atom better than Sublime Text?
Different thing entirely. Atom the feed format is a syndication standard (RFC 4287, 2005) and has nothing to do with Atom the code editor, which GitHub shipped in 2014 and retired in December 2022. If you're comparing editors, Sublime Text is still maintained and Atom is not.
Does the wrong MIME type matter?
Rarely for readers, which sniff the document body. It matters for browsers: application/rss+xml frequently triggers a download instead of a display, which is why 14% of feeds are deliberately served as text/xml.
---
Method. Census of distinct feed URLs actually subscribed to by Proofite users, fetched and classified on 23 August 2026 by reading the root element and the declared namespaces of each document. Recurring measurement; percentages are updated on each run. This is not a representative sample of the web — it is the set of sources an active reading community has chosen to follow, already filtered towards live, maintained sites. Hence the 1% unreachable, a figure that would be far higher on a random crawl. Newsletter and digest figures come from automated analysis of email bodies and generated digests in July 2026, not from publisher statements or surveys. Raw data, one row per feed, available on request.