The State of AI Citations
Observed citation rates across completed RankEcho audit runs, joined with fixed-prompt proof outcomes. Runs can differ in prompt panels, engines, sample depth, dates, and repeat domains, so this is a descriptive product-data report rather than a representative industry benchmark. Data through 2026-09-17.
Descriptive citation rate by recorded industry
| Industry | Audits | Avg citation rate | Off-site share |
|---|---|---|---|
| B2B SaaS | 45 | 22% | 96% |
| Other | 27 | 33% | 93% |
Industries below the 5-run display minimum are not shown. Rows above it can still contain repeat domains and unlike audit configurations; the minimum controls display, not comparability or representativeness.
Observed movement after a fix shipped
Of 28 prompts re-tested after a fix shipped, 12 met RankEcho's reporting rule — a citation observed across at least 3 post-ship runs, with the observed post-ship rate beating that prompt's own baseline. This repeat-count rule is not a statistical-confidence threshold. Across those wins, 5 first appeared within 3 days, 2 from 4-13 days, and 5 at 14+ days. These are elapsed-time bands only; timing cannot identify whether retrieval, training, or another system behavior produced a citation. Movement is reported as correlation with controls — engines change for many reasons, and this report never claims otherwise.
Methodology in one paragraph
Every figure comes from completed RankEcho audit runs and tracked proof loops. The audit aggregate does not deduplicate domains or require one common prompt panel, engine configuration, sampling depth, or calendar window. Free single-sample audits may reuse an engine answer observed within six hours, as disclosed on the methodology page; paid sampling queries live. Industry rows are included with 5+ runs, and outcome figures require 25+ measured prompts with 10+ reportable wins. These display rules do not establish statistical sufficiency.
Aggregate counts, rates, and medians only. This page never publishes domains, prompts, or customer-level results. Cross-audit rates are descriptive, proof movement is correlational, and neither predicts an individual outcome.