Why we published the SEO data beside the answer
A chart can explain the answer. The data behind it lets someone use the answer again.
In September 2026, AI Symantix published SEO datasets with a readable page and a downloadable JSON file. The pages linked to the files. Dataset structured data described what those files contained.
That gave us a concrete question: would crawlers collect the underlying data after reaching the page? The full SEO dataset study follows that question across the portfolio. This case brings its most useful findings into the visibility lab.
How the original dataset study counted discovery
The original research used retained server logs from September 6 through October 10, 2026. It compared dataset requests with content requests and looked for page-to-file relationships in the recorded referrer field.
On September 23, the Digital Karma Federation v8.2 rollout put Dataset and DataCatalog structured data into rendered pages across the portfolio. The study compared distinct sites receiving catalog requests before and after that date.
BellyUp! produced a large amount of filtered-URL crawling during the same period. The catalog comparison excluded that family so a local discovery crawl burst would not dominate a different question.
The comparison is observational. The periods are unequal, the sites kept changing, and a crawler's referrer is a reported value. The logs let us describe what reached the server.
What the dataset requests showed
Googlebot requested the Exposure Velocity JSON file with the Lab page recorded as its referrer. GPTBot and PetalBot also requested research files with the linked research pages recorded as referrers. Those sequences support a useful publishing practice: make the reusable file easy to find from the page that explains it.
The catalog comparison found a wider footprint after September 23. The table counts sites, so repeated requests to one busy catalog do not inflate the reach figure.
The crawlers also behaved differently. In the original study, OAI-SearchBot devoted a larger share of its requests to the dataset layer than GPTBot did. Treating every AI crawler as the same audience would have hidden that difference.
The gain we can document is collection and discovery. A crawler taking the file is useful information about access to the data. An AI answer citing that file is a later event that needs its own saved answer.
The publishing practice worth repeating
The practical move is straightforward. Publish useful original data with an explanation a person can understand. Link the downloadable file from that explanation. Describe the dataset accurately in the page's structured data.
That gives a reader an answer and gives a crawler a clear connection between the finding and the underlying records. It also gives us something testable. We can count successful file requests and see which page the request names as its source.
The original article keeps the detailed file-request tables and its Google Search Console analysis. This case remains a dated record of the same study rather than presenting old counts as live numbers.
What we are watching in dataset discovery next
Recheck distinct catalog reach in equal-length retained windows and record dated AI answers for a fixed set of questions that these datasets can answer.
The full SEO dataset study and its public records
- Full study: AI Search Visibility Using SEO Datasets
- Original public research snapshot
- AI Symantix datasets
- Parallel Real SEO Life case study
This is a dated case study from our own portfolio. Public sources and the downloadable snapshot let you check the figures; the Digital Karma Data Warehouse supplies the private search and request records.