Citing your sources does not make your statistics accurate.
That is the standard advice in every content audit checklist, every E-E-A-T guide, every editorial policy document a content team writes and then files away. Add a citation. Name the study. Link to the report. Blog source verification is treated as handled the moment a writer types "according to."
We tested that assumption. The dataset: 575 claims extracted from 100 posts across 20 SaaS domains, the same corpus from the original scan. That scan found 34% of claims cited no source at all. This one asks a harder question: what about the other 66%? Can a reader follow the trail?
Of 316 third-party claims, 252 cannot be verified. That is 80%. The citation supply chain content teams assume exists is, for four out of five borrowed statistics, absent. That is content debt that accumulates before the post is a day old.
What "Sourced" Actually Means in SaaS Blog Posts
Strip the first-party claims and the dataset narrows to 316 third-party statistics. Numbers borrowed from someone else's research, someone else's survey, someone else's report. The numbers that depend on blog source verification to hold up.
The breakdown:
- 64 claims (20%) have both a named source and a URL. A reader can click through and check.
- 58 claims (18%) name a source without providing a link. "According to Gartner." No URL. "A McKinsey study found." No path to the study.
- 194 claims (61%) cite nothing at all. A number appears in the prose as if it were common knowledge.
Naming a source feels like citing a source. It is not the same action. A sentence that says "Gartner reports that 75% of B2B organizations will shift to a composable architecture by 2027" gives the reader a brand name and a statistic. It does not give them a report title, a publication year, or a way to verify the number. The claim is orphaned data from the moment it is published.
HubSpot's scanned posts illustrate the pattern. Across 65 total claims, 18 were classified as sourced-named. HubSpot names its sources more often than most: "According to Gartner." "A Forrester study." "Research from McKinsey." The names are there. The links are not. Authority signaling without a verification path.
These are zombie statistics: numbers that circulate with a label attached but no provenance a reader can follow. They look sourced. They read as credible. They cannot be checked, because charts are claims about reality and a claim without a trail is a claim without accountability.
Blog Source Verification Is 80% Broken
The 252 unverifiable third-party claims follow a simple pattern. 194 have no source at all. 58 have a name but no link. Combined: 80% of every borrowed statistic across 100 posts.
The obvious suspect is dead links. The data says otherwise.
Of the 99 source URLs found across all claim types, 89 are alive and returning content. 7 are restricted behind paywalls or login walls. 1 is dead. 2 timed out. Of those that resolve, 35 are fresh, 4 are stale, and 1 is aging. When a source URL exists, it almost always works. The problem is upstream of link rot. The links were never there.
That makes this a frozen liability. 194 claims that cannot be corrected because they cannot be traced. If the original study is retracted, updated, or debunked, no notification reaches the post that borrowed the number. The post does not know where the number came from. Neither does the team that published it.
Blog source verification fails at the input layer. The writing workflow does not require a source URL. The CMS does not require one. The editorial checklist does not check whether a reader can follow a number back to its origin.
The scan used the same claim-extraction pipeline that produced the original 575-claim scan. The classification added one question: given this claim and its context, can a reader reach the primary source? For 80% of third-party claims, the answer was no.
Two Companies, Same Industry, Opposite Infrastructure
The 20 domains in the dataset do not cluster around a shared standard. The differences are structural, not editorial.
GitHub publishes 91% first-party claims and 4% unsourced. OpenAI publishes 9% first-party and 88% unsourced. Same industry. Opposite infrastructure.
GitHub writes about its own platform data. Copilot usage metrics, repository growth, developer survey results. When the source is the company itself, blog source verification is built into the act of writing. There is no third-party trail to lose because there is no third party.
OpenAI's scanned posts draw heavily on external research and industry assertions. 29 of 33 claims cite no source. The content reads well. A reader cannot check a single one of those 29 numbers without running a separate search.
HubSpot occupies its own category. 55% of claims are unsourced, but 18 carry a source name without a link, the highest named-but-unlinked count in the dataset. HubSpot has a sourcing culture and names authorities consistently. It does not close the loop with a URL. The gap between naming and linking is where most of the verification failure lives.
Buffer tells the opposite story. 84% of its claims are first-party: product data, internal experiments, its own transparency reports. When you write about your own data, the sourcing problem disappears. Buffer's 8% unsourced rate is a structural outcome, not an editorial achievement. The content draws from what the company measures.
Twilio is the extreme case: 0% unsourced across 9 claims, 78% first-party. When every number traces to internal data, the verification problem does not arise.
The difference between 4% unsourced and 88% unsourced has less to do with editorial rigor than with whether the blog's subject matter generates its own data or borrows it. Teams that write about their own products and datasets produce verifiable content by default. Teams that synthesize industry trends produce content that looks sourced and cannot be checked.
97 Out of 100 and Still Unverifiable
The average freshness score across all 100 posts in the scan: 97 out of 100. 99 of the 100 posts scored between 81 and 100. Zero scored below 60.
And 34% of their claims cannot be verified.
The audit passed. The claims did not.
Freshness measures when a page was last touched: update recency, metadata accuracy, link health. It does not ask whether the statistics inside the page trace back to a primary source. A post updated yesterday with five unsourced numbers scores higher than a post from 2024 where every claim links to a named, dated study.
That is a passing score that says nothing about whether the content can withstand scrutiny. The page was touched recently. The claims inside it were not part of the audit. The 34% unsourced rate exists independently of freshness. Fresh posts carry unsourced claims at the same rate as older ones, because the sourcing gap was baked in at creation. No update cycle fixes what was never recorded.
Blog source verification and freshness scoring measure different axes. One asks when. The other asks whether. Most content audits only ask when.
The gap between these 20 domains reflects different sourcing standards, and most teams have never written theirs down.
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 options A and B is where most of the verification gap lives. Linking to a secondary citation without verifying the primary source creates a claim that looks sourced but cannot be checked.
What Source-Level Tracking Changes
The 80% figure is a measurement of absence. The citations do not fail verification. They were never built to support it.
Source-level tracking treats each citation as a dependency. When the upstream URL changes, every claim citing it gets flagged. That requires one precondition: the URL has to exist in the first place.
For the 64 claims with both a source name and a link, monitoring is already possible. For the other 252, there is nothing to monitor. The claim sits in the prose with no connection to its origin. How to detect when published data goes stale requires a trail. Living content requires a known origin. Neither can operate on a citation that was never recorded.
The infrastructure for monitoring exists. The citation supply chain it would monitor does not.
We ran this scan on 100 posts across 20 domains. You can run it on one of yours.
If 80% of the citation trail does not exist, what exactly are freshness scores measuring?
Source-level tracking is in early access. Join the waitlist or browse the claim registry.