We scanned 6,751 data claims across 938 SaaS blog posts on 45 domains and asked one question: how old is the data these posts cite? Almost a quarter of the posts that cite data carry numbers two or more years out of date, and the longer a post sits unmaintained, the more of its data has aged out. The state of content decay in 2026 has a second layer that never reaches a traffic graph: the data inside the post going out of date while the post keeps performing. The borrowed numbers that age first are the same ones you cannot verify by hand.
Claim: Almost a quarter of SaaS blog posts that cite data carry numbers two or more years out of date. Source: The State of Content Decay 2026, LiquiChart corpus study (45 domains, 938 posts, 6,751 claims). Verified: 2026-06-04.
What Content Decay Is in 2026
Content decay in 2026 is the data inside a published post going out of date while the page keeps ranking. A cited number ages, the source behind it drifts or breaks, and nothing on the page signals the change. Measured at the level of each claim, it stays invisible to any audit that watches traffic.
The term came from SEO teams describing a page losing organic traffic and rankings over time. That definition treats the page as the unit and the draft as the fix: update the date, add a section, refresh the keywords, re-target the intent. It measures the wrapper and never opens the box.
A second kind of decay lives inside that box, and it never touches your rankings. The "72% of teams" you cited from a 2023 report still sits in paragraph nine, and your readers still treat it as current in 2026. The page ranks fine. The link still works. Nothing on it broke.
The number just got older, and nobody scheduled a review because the page gave no signal that anything had changed. That is what content decay is at the claim level.
How We Measured It
I kept finding two-year-old numbers in posts that still read as current. For this content decay study, we ran every published post on 45 SaaS domains through a content scanner that extracts each data claim as a verbatim span, reads the time reference attached to it, and tags how old the cited data is. Then we traced the citations behind those numbers, hop by hop, to wherever they actually end. The result is two reproducible signals: how aged the data is on the calendar, and how verifiable its source is by hand. Both reproduce against the live production pipeline, with confidence intervals on every rate.
One thing we deliberately did not measure: whether a given number is provably wrong. Reading a publisher's intent well enough to declare a stat incorrect is a different and far less reliable claim. This report tells you what has aged and what cannot be verified. It does not accuse anyone of publishing a false number, and that distinction is the most important finding in it.
A Quarter of Data Posts Carry Years-Old Numbers
Across the 751 posts that make at least one data claim (of 938 scanned), 177 carry data that is two or more years old. That is almost a quarter of them, and it is the spine of this report. You cannot tell which posts are running on aging data without scanning them.
The corpus-wide figure is smaller than the post-level one. Posts carry only a handful of claims each, and the claims with no date attached cannot age on the calendar at all. Of all 6,751 claims, 442 are two or more years out of date. The aging concentrates in the posts that lean hardest on data, which is exactly where a reader is most likely to take a number at face value.
The Age Curve
The content decay rate does not creep in evenly. It switches on at the one-year mark.
In posts under a year old, 12.5% carry aged data. Cross the first birthday and that share jumps to roughly 28%, more than doubling, then plateaus for the next two years.
The share of affected posts levels off after year one. The amount of aged data inside each affected post does not.
In fresh posts, 2.3% of cited stats are two or more years old. In posts that are two to three years old, that figure reaches 13.1%, close to a sixfold rise. Older posts carry more of it, accumulating quarter after quarter in files nobody reopens.
The Borrowed Numbers Age First
Not every kind of claim ages at the same rate. Borrowed "according to" citations age fastest of all. The number is frozen at the year you pulled it while the report behind it keeps shipping new editions without telling you. First-party numbers, the ones a content team has a standing reason to remeasure, hold up longest.
That ordering matters more than any single rate. When I reopen one of our own old posts now, I check the borrowed "according to" lines first, because that is where the aging lives. They are also the numbers you can least afford to trust on sight.
Most of Your Numbers Are Borrowed
73% of the claims on these blogs are borrowed data. Only 27% are first-party numbers the publisher measured itself.
Sourcing has improved over the last few years. About 85% of borrowed numbers now carry a name, and about half carry a link. The publishing field solved attribution. The open question is whether the attribution holds up when you follow it.
Half of Borrowed Numbers Have No Reachable Link
Start with the 4,907 borrowed claims. Only 51% of them carry an external link at all. The other 49% are unverifiable by hand the moment they publish: a named source with no way to reach it, or no source named at all. Of the links that do exist, nearly one in four linked citations is dead, gated, or broken. The citation was there on publish day, and it has decayed on its own clock ever since, independent of anything the author did.
Most Citations Stop at a Middleman
Follow the links that still resolve, and the picture gets worse. Only 10.6% of citations reach a primary source, the original page that first reported the number. The other nine in ten stop at an aggregator, a roundup, or a post that cites someone who cites someone else. More than 94% of these chains terminate in a single hop. A working link buys you the appearance of a source and almost never the source itself.
Before the verdict, your own habit is worth a look.
Trusting the number and moving on once a link is in place clears the lowest bar. As readers weigh in above, where a team stops on this ladder sets the ceiling on how much of its borrowed corpus it can actually stand behind, because the rung you settle on is the rung every citation you publish inherits.
Watching the Link Is Not Watching the Claim
Put the two halves together. The borrowed "according to" numbers are the ones that age first, and they are the same numbers that mostly cannot be traced to a real source. A link-checker tells you the URL returns a 200. What claim verification catches that a link check cannot is everything that matters: whether the page still states your number, whether that number is current, and whether the page is the original or a fourth-hand copy. Watching the link is not watching the claim.
The only thing that keeps a published number current is monitoring the claim itself, continuously, as a unit of content maintenance infrastructure. A quarterly audit line never does that work.
Almost None of It Is Provably Wrong
The strict test for whether any of these aged numbers is provably incorrect comes back almost entirely clean. The value-mutation check found zero rewritten figures across every traced citation. These numbers have aged without anyone maintaining them. Almost none have been rewritten into something false.
That is a smaller claim than "your content is full of errors," and I am making the smaller one on purpose. It is the claim the data supports, and the more useful warning. Aging is how decay starts, and aging is invisible on a skim. A number that was right in 2023 and has gone unchecked since is still harmless and still preventable.
Catching it early means being told the moment a cited figure crosses from current into aged. The freshness tier flips to Aged while the number itself still reads clean, giving you time to check it before a reader, a journalist, or an answer engine quotes the old one back at you.
Why a Refresh Date Never Reaches These Numbers
The standard maintenance ritual is the content refresh: reopen the top posts on a cadence, freshen the intro, bump the date. It is the fix every content-decay playbook prescribes, and against the aging measured here it does almost nothing, because it is aimed at the wrong unit. A "last updated" stamp records that someone touched the page. It says nothing about the 2023 figure buried in the body that the refresh never read. This is the gap freshness theater names: motion that looks like maintenance and reaches none of the claims that actually decayed.
Maintenance has to happen at the level of the claim. The Content Health Scanner that produced every number in this report watches each cited claim, flags the moment its data crosses into aged, and traces the citation chain behind it to see whether the source is still reachable. The work moves from "reopen everything once a quarter and hope" to "fix the specific numbers that need it." The unit of decay is the claim, so the unit of maintenance is the claim too.
You can run the same scan on your own back catalog and see which of your posts are carrying the load.
The Liability You Cannot See
Every quarter a data-backed post stays up, it accrues a little more of this. The numbers age, the citations behind them decay on their own schedule, and none of it shows up in the traffic graph you watch. It is content debt in its purest form: a balance growing in a currency your current tools do not measure, on the posts that earn you the most trust and would be the most expensive to get wrong. The part of your archive that is decaying fastest is the part you can least verify by hand, and right now the only thing standing between aged and quoted-back-at-you is whether anyone happens to reopen the file in time.