How 3 External Audit Errors Led to Defects in 74 Published Posts
Published
This article was first published in Korean by Intrix Lab on August 26, 2026.
The 3 errors reported by an external tool were a clue to gaps in our checks, rather than the end of the investigation. As of 2026-08-25, the external tool found 3 structured data errors across 100 pages on intrix.kr. After reviewing our internal checking code, we examined all 101 published posts and found 74 instances of missing image properties in Article-family structured data. The discrepancy came from treating valid JSON syntax and schema compliance as the same verdict.
Why did our internal checks report 0 of the 3 errors?
The starting point was a mismatch between the external audit and our own audit. The external tool reported a Site Health score of 94%, and all 3 errors involved structured data. It also reported 133 warnings. Our internal engine, examining the same site, did not detect those defects.
We therefore revisited the assessment criteria before looking at the results screen. Our internal checks verified JSON syntax but did not check whether each node had the properties required for its type.
How did we count the 74 published posts?
Rather than only fixing the results from the external tool's 100-page crawl, we separated the causes into two categories. In the homepage's @graph, we found 1 reference node without an @type. In Article-family structured data, we found a pattern of missing required image properties.
We then examined all published posts in the database to look beyond the crawl sample. The population was 101 published posts on intrix.kr. Of these, 74 posts were missing an image property in their Article-family structured data. Had we treated the external tool's reported error count as the total number of defects, we would have missed the same pattern outside the sample.
What did we change the checks to assess?
The change was intended to separate the verdicts so that passing a syntax check would no longer be recorded as passing a schema compliance check, rather than to produce more error messages.
How far can these results be trusted?
The measurement covered a single day, 2026-08-25. The external tool's population was 100 pages on intrix.kr, while our internal full count covered 101 of our own published posts. Because the populations and checking methods differed, the 3 errors and 74 missing image properties cannot be interpreted as a before-and-after change in the same rate.
This finding does not guarantee that other sites will have the same omission rate. Different structured data generation methods and post templates produce different types of defects. It also does not support a conclusion that external tools always undercount. What we confirmed here was that expanding a clue from a limited crawl into a full inspection for the same pattern revealed a wider range of omissions.
Structured data validity checks assess the conditions that help search engines interpret documents. Fixing these issues does not guarantee rankings, citations, or visibility. These data establish only the types of defects our own checker missed and their scope of impact as of 2026-08-25.
Where should you start checking for similar omissions?
First, treat successful JSON parsing and compliance with type-specific requirements as separate checks. When a sample crawl reveals a pattern, expand the population to all pages using the same template and count again. To check your site's current structured data and crawl coverage, you can measure actual responses with our free SEO and GEO audit.
Frequently asked questions (FAQ)
Q. If JSON parses successfully, isn't it valid structured data?
No. Successful parsing means the syntax is not broken. Whether a node has the properties required for its type must be verified through a separate schema compliance check.
Q. Can the 3 errors and 74 missing image properties be compared directly?
They do not share the same denominator. The 3 errors came from the external tool's 100-page crawl, while the 74 came from a full count of 101 published posts in the database. The former should be treated as a clue that revealed gaps in the checks.
Q. Why inspect every post instead of stopping at the sample?
When the same template is applied to multiple posts, a defect found in a sample can also recur outside the crawl coverage. In this case, we established the scope of impact by counting all published posts under the same conditions.