The short version
- The Self-Citation Index tracks one number every week: what share of AI citations about a brand point to the brand's own website, versus a third party.
- Current reading: 5.6% (95% CI 5.4 to 5.7%), across 60,490 citations and a panel of 20 brands.
- That is consistent with an earlier one-time study on a separate sample: a brand's own site is a small, stable minority of what AI cites about it.
- The index is a live tracker, not a one-off report. We publish the reading, the sample size and the confidence interval every time it updates, the same reporting-integrity rule as everywhere else in llemmy.
We published a one-time study in June 2026 on 37,547 citations that found brand-owned domains made up only about 5% of what AI engines cite. The obvious follow-up question was whether that was a fluke of one sample, or a real, stable pattern. So we turned it into something we track every week instead of something we measured once: the Self-Citation Index.
What it measures
For every AI answer about a tracked brand that carries at least one citation, we check where each cited source points. A citation counts as owned when its host matches the brand's own domain (or a subdomain of it). Everything else, reviews, press, forums, competitor comparisons, directories, counts as third party. The index is owned citations divided by total citations, across a panel of monitored brands with enough volume to be meaningful (at least 30 citations in the window).
The current reading
Self-citation 5.6% · n=60,490 · 95% CI 5.4-5.7%
That is a tight interval on a large sample, which is itself informative: this is not sampling noise settling on a low number by chance, it is a stable reading. It also lines up closely with the June study's roughly 5% on an entirely separate sample of 37,547 citations, gathered a different way. A full month of model updates in between did not move it. When a completely different sample keeps landing in the same place, that is the signal to trust the number, not just the method.
Why this happens
It is tempting to read a low self-citation rate as a problem to fix with better on-page SEO. That is not quite what it says. AI engines are doing something closer to editorial corroboration than indexing: when they answer a question about a brand, they lean on sources that talk about the brand, not just sources from it. A brand's own website is a claim. A review site, a forum thread, a comparison article, or press coverage is a third party backing the claim up, or not. Engines built to sound credible reach for the second kind more often.
This is the same finding behind one of the best-practices entries we publish in the app: earned media and third-party corroboration move citations more than on-page changes do. A brand's own content still matters, it is what gets referenced and repeated, but it is rarely the citation itself.
What we will keep doing, and what we will not claim
We will keep publishing this reading every week, with the sample size and the confidence interval attached, the same way every number in llemmy works. We will not name individual brands in the panel here: the point of this page is the aggregate pattern, not a public scoreboard of who is at what percentage. And a single index number does not tell you whether your specific brand is above or below it, only your own tracked data can answer that, which is exactly what the Citations view in the app is for.
Related reading: What AI engines actually cite, the original 37,547-citation study this index follows up on, and how llemmy measures for the full methodology and its limits.