RankEcho reports
RankEcho publishes three evidence surfaces: a 384-observation cross-engine citation study, an aggregate one-request screening of 5,542 responding business homepages, and a Proof Ledger of before/after prompt observations. Each reports its population and limits. The readiness screening has no public row-level dataset, and its technical indicators do not establish citation eligibility, ranking effects, or causes.
State of AI Citations, Q3 2026
384 observations across four AI engines, testing whether they cite the same sources when asked the same question. The headline finding is how rarely they agree: cross-engine agreement sits near 19%, which means visibility on one engine predicts very little about visibility on another.
The study also measured answer stability across repeat runs of identical prompts, and Google AI Overviews coverage, which appeared for a minority of queries but cited generously when it appeared. The full 48-prompt battery and the underlying dataset are published alongside the report.
Website AI Readiness, July 2026
The readiness report screens 5,542 responding business homepages. In the aggregate counts, 33.7% had no detected structured data and 10.9% fell below RankEcho's initial-HTML visible-text threshold.
These are technical indicators, not observations of indexing, impressions, citations, or causal effects. The initial pool was 7,126 domains from public business listings across US regions; 1,584 that did not return usable content in one request were excluded. RankEcho has not published the row-level sample.
Live citation data
Two surfaces update as data accumulates rather than being fixed to a publication date.
The State of AI Citations page carries rolling figures across audits as the sample grows. The Proof Ledger records tracked prompts before and after fixes shipped, with the misses left visible, because a ledger that hid them would prove nothing.
How these reports are made
Same rules across all of them. Every figure comes from observations RankEcho collected rather than from other people's studies. Every report states its population and sampling method, so a reader can judge what the numbers apply to. Excluded data is reported as excluded rather than quietly dropped.
Results that did not move are published alongside results that did. A research programme that only publishes wins is marketing with a chart attached, and it would undercut the one thing these reports are for.
Dataset availability is stated per report. The citation study publishes a downloadable artifact; the Website AI Readiness page currently publishes aggregate counts and method only, not its row-level sample, so that screening is not independently reproducible from a public artifact.
Citing this research
The reports are free to cite with attribution to RankEcho and a link to the report page. Where a dataset is published, it carries its own licence terms on the report page.
For questions about method, or to request a cut of the data that is not published, the contact page reaches us directly.
Frequently asked questions
Three evidence surfaces: the State of AI Citations study covering 384 observations, an aggregate Website AI Readiness screening of 5,542 responding homepages, and the Proof Ledger. The readiness report publishes aggregate method and counts but no row-level sample.
In our Q3 2026 study of 384 observations across four engines, cross-engine agreement was roughly 19%. In practice that means being cited by one engine predicts very little about whether another cites you for the same question, so single-engine measurement understates the problem.
Yes, with attribution to RankEcho and a link to the report page. Where a dataset is published it carries its own licence terms on that page. For method questions or unpublished cuts of the data, use the contact page.
Yes. The Proof Ledger publishes tracked prompts where citations did not move alongside those where they did, and each study states what it does not show. A research programme that only reports wins is marketing with a chart attached.