Of the 301 live pages that AI engines cited across 316 prompts in seven verticals, exactly eight carried original data plus a stated methodology. Those eight averaged 11.3 citations each. Every other cited page averaged 3.4. The numbers come from Kevin Indig and Amanda Johnson, published on Search Engine Land on 2026-07-08, drawn from Gauge's citation dataset of 1,075 citations. The concentration is as sharp as the premium: 75 of the 90 primary-research citations went to one cluster of pages on cloud data-warehouse benchmarks, and Fivetran's warehouse benchmark, published in 2022, took 44 of them by itself.
Credibility: MEDIUM — named authors, with method and sample stated in print. Against that: the sample is small (1,075 citations), the dataset is a single vendor's proprietary corpus (Gauge), no downloadable dataset or methodology document is linked, and the authors sell growth and AI-visibility services, which leans them toward "invest in research" conclusions.
- 8 of 301 cited pages (2.7%) qualified as primary research (original source data + methodology).
- Primary-research pages averaged 11.3 citations per page vs 3.4 for all other pages (3.3x density).
- 75 of the 90 primary-research citations came from a single cluster: cloud data-warehouse benchmarks.
- Fivetran's warehouse benchmark (published 2022) alone captured 44 citations, just under half of all primary-research citations in the dataset.
Stay ahead of AI search
The changes that move your rankings and AI citations — and the exact move to make — before they cost you. Free, three times a week.
For a content team weighing an original-research budget, the density premium is real, and it gets paid to the incumbent. Fivetran's benchmark has been sitting there since 2022 and still absorbs nearly half the primary-research citations in this dataset. Age cost it nothing. The engines keep returning to the page they already trust with the question, so a second study on the same question collects the leftovers. The budget decision starts by naming who currently owns the answer in your category. Where an incumbent sits, you need a reason to be cited next to it: newer data, a segment it skips, a metric it never publishes. Where nobody owns the answer yet, the 3.3x is unclaimed.
- Run the client's tracked prompt set and write down which study or benchmark each engine cites today; fund original research only where no incumbent benchmark exists, or where yours differs on recency, segment, or metric.
- Ask the web team to add a visible methodology block (sample size, date range, method) to the stats pages you already publish; the dataset's own definition of citable primary research requires it.
- Rewrite each key finding as a one-line extractable stat (number + unit + year) near the top of the page.
Search Engine Land: Why most original data never gets cited, by Kevin Indig & Amanda Johnson, Search Engine Land, 2026-07-08 — https://searchengineland.com/why-most-original-data-never-gets-cited-481676