A link to a source is not proof of a source.
That distinction used to be academic. For years the sourcing problem in SaaS content was absence: a number would appear in the prose with no name and no link, and a reader had no way to check it. Add a citation. Name the study. Link to the report. The fix was obvious, and the industry applied it.
It worked, only partway. We traced 3,299 third-party claims across 961 posts on 46 SaaS domains, and only 30% carry an external link in the claim's own paragraph, the place a reader actually looks. Pages are full of links; links that belong to the claim are rare. What sits on top of that gap is subtler still: a link that resolves, returns a healthy status, and still cannot verify the number attached to it.
About one in five of those verified external links is dead, gated, or broken when you re-check it through a real browser. Of the links that resolve and carry a readable date, more than a third point to data at least two years old. The sourcing reflex spread across the industry. The verification gap stayed where it was.
What claim verification catches that a link check cannot is exactly this gap: a working hyperlink to a page that never stated what your sentence claims.
What Sourcing Looks Like Across 46 SaaS Blogs
The dataset is 5,034 claims extracted from 961 posts across 46 SaaS domains, the same corpus behind the companion attribution study. Strip the first-party claims, the numbers a company reports about itself, and 3,299 third-party statistics remain. These are the ones that depend on someone else's research, and the ones that depend on blog source verification to hold up.
The breakdown:
- 1,006 claims (30%) carry an external link in the claim's own paragraph. A reader can click through to someone else's page.
- 752 claims (23%) name a source but give no verified external link. Some cite a named authority with no URL at all. Many link only to the publisher's own pages, which sends the reader back to the blog rather than out to the source.
- 1,541 claims (47%) cite nothing at all. A number sits in the prose as if it were common knowledge.
The verified external link is the minority. Seven in ten borrowed claims send a reader nowhere they can check: named with no URL, linked only to the publisher's own pages, or unsourced entirely. The reflex to attribute has spread faster than the discipline to make those attributions checkable.
Part of that gap is mechanical, and it is the reason this audit counts links the way it does. A link only counts here when it sits in the claim's own paragraph and survives tracking-link filters. When we credited any external link near the claim, the way most audits and most readers' first impressions do, roughly a third of the seemingly linked citations turned out to be borrowing a nearby link, a navigation item, a footer, or a neighboring paragraph's citation that never pointed at the claim's source.
The 70% that remain unverifiable still matter. A sentence reading "a named research firm reports that 75% of B2B organizations will shift to a composable architecture by 2027" gives a reader a brand and a number, not a report title, a year, or a way to check it. That claim is orphaned data from the moment it ships. But the larger story has moved past the missing link, to what happens when you follow the link that exists.
Following the Link Fails About 1 in 5 Times
We re-checked every verified third-party source URL in the dataset through a real browser, the same rendered-fetch path our scanner uses: a request to the link, escalated to a headless browser when a bot-wall blocked the first try, classified by what came back. Of the 1,006 verified citations, 801 resolved cleanly. The other 205 did not.
That is 20.4%, about one in five external citations, returning something other than a readable page. 20 were dead, a 404 or 410 where the page used to be. 102 were gated behind a login or paywall, a 403 the reader hits after clicking through. 83 returned a server error, a redirect loop, or a rate-limit block that a real browser still could not get past. Each one is a citation that passed the only test a standard editorial workflow applies, the link is blue and it points somewhere, and failed the test that matters, the page at the other end can be read.
This is the inversion the sourcing fix created. When the problem was absence, a link was progress. Now that linking is standard, the link itself is the exposure. A claim locked to a source that has moved, gated, or vanished, with nothing in the workflow that notices, still presents its number as fact. A reader who clicks through lands on a 404, a login wall, or a timeout, and the number loses its only footing. I have shipped citations like that myself, live the day they went in and dead a year later, with nothing set up to catch the difference.
The 2.0% genuinely-dead rate deserves its own note, because dead links are the one failure mode existing monitoring tools catch. The other 18% of unhealthy links, the gated pages and redirect loops and rate-limited servers, return a status that no broken-link checker flags as a problem. If one of these is your own citation, fixing link rot in your citations starts with telling a genuinely dead source apart from one a login wall is only hiding.
There is a methodology note worth making here, because the tool you check with decides what you find. A naive link-checker that sends a raw request with no browser flags lands closer to three in ten (28.3%): it counts every bot-wall, consent-redirect loop, and rate-limit block as a dead source. Re-checking those through a real browser, the way a reader's browser and our own scanner do, clears about eight points of them. We report the browser number because it is the one a reader actually experiences, and because a checker weaker than the pages it audits will always overcount the damage.
When the Link Resolves, the Data May Be Years Old
Reachability is the lower bar. The harder question is whether a link that resolves still points to current data.
Of the external citations that resolved and carried a readable publication or update date, 224 were less than a year old, 76 were one to two years old, and 157 were at least two years old. More than a third of the datable links point to data that predates most readers' assumptions about how current a cited statistic is.
A two-year-old source can be perfectly sound, and I do not treat age alone as a problem. A 2024 benchmark cited in a 2024 post holds up. The same benchmark still cited in a post a reader finds today, presented in the present tense, asks the reader to treat aging data as current. The link resolves. The page loads. The number it anchors has not been checked against a newer edition that may have moved it.
More than half of the linked sources carried no machine-readable date at all, which is its own finding: for most citations, even the freshness of the source cannot be determined from the outside. The reader cannot tell whether the page behind the link was written last month or five years ago.
When Your Team Cites a Statistic, How Far Do You Check It?
Before the rest of the findings, a baseline. The numbers above describe what ends up published. This asks what your team does before it does.
Most content teams treat sourcing as a formatting step: find a number, paste a link, move on. Whether that link points to the original dataset or to another post that also pasted a link rarely enters the workflow. The distinction matters because a secondary citation can go stale without the citing team ever knowing. Only the primary source change triggers a correction. Everything downstream inherits the error.
The distinction between linking a source and verifying it is where most of the residual gap lives. A citation that links to a page no one has opened since the day it was added carries the appearance of verification and none of the substance.
Sourcing Habits Vary More by Subject Than by Discipline
The 46 domains do not cluster around a shared standard. The unsourced share runs from 6% on the most rigorously attributed blogs to past 40% on the least, with a median near 21%. I have read across that range, and the gap rarely tracks how careful the writers are.
It tracks what they write about. Blogs covering their own platform data, usage metrics, internal experiments, and transparency reports verify their sources by default, because the company is the source and there is no third-party trail to lose. Blogs that synthesize industry trends borrow most of their numbers, and every borrowed number is a citation that has to be linked, checked, and kept current to stay verifiable.
Teams writing about their own products produce verifiable content as a byproduct. Teams synthesizing the wider industry produce content that looks sourced and depends on links that decay.
What Link-Level Monitoring Changes
The 70% unverifiable rate is a measurement of absence: claims that were never built to support a check, named with no link or pointing only at the publisher's own pages. The 20.4% unhealthy-link rate is a measurement of decay: claims that were built to support a check and lost it after publication. These are different problems, and only the second is fixable after the fact.
Link-level monitoring treats each citation as a dependency. When a source URL starts returning a 404, a 403, or a timeout, every claim that cites it gets flagged. When the page behind the link changes and the cited number is no longer on it, that propagates too. This is the layer beyond a broken-link checker: it tests whether the source still supports the claim that points to it, well past whether the URL returns 200. Monitoring the full citation chain extends that check past the first hop.
For the 30% of borrowed claims that carry a verified external link, this monitoring is possible the moment the link exists. For the other 70%, there is nothing to monitor, because the claim sits in the prose with no external connection to its origin. Detecting when published data goes stale requires a trail. Living content requires a known origin. Neither can operate on a citation that resolves to a dead page or was never recorded at all.
The infrastructure for monitoring exists. Across this dataset, the citation supply chain it would watch is now mostly built, and about one link in five has already started to crumble.
We ran this scan on 961 posts across 46 domains. You can run it on one of yours.
A link made the citation look verified. For about one in five, the page at the other end says otherwise.
Source-level tracking is in early access. Join the waitlist or browse the claim registry.