--- title: "Two Engines Almost Never Cite the Same Source for the Same Question on the Same Day" description: "A citation index can say a source is 'consensus' because five engines have each cited it at some point. It cannot tell you whether two engines cite it for the same buyer question on the same day. Measured on 2,704 question-days: they almost never do." canonical: https://authoritytech.io/curated/answer-level-cross-engine-consensus-not-window-consensus-2026 last-updated: 2026-09-22 --- # Two Engines Almost Never Cite the Same Source for the Same Question on the Same Day A citation index can say a source is 'consensus' because five engines have each cited it at some point. It cannot tell you whether two engines cite it for the same buyer question on the same day. Measured on 2,704 question-days: they almost never do. Canonical URL: https://authoritytech.io/curated/answer-level-cross-engine-consensus-not-window-consensus-2026 Published: 2026-09-22 Author: Jaxon Parrott Tags: Afternoon Brief, AI Search & Discovery, Measurement Every citation index, ours included, computes consensus the same way: watch a domain over weeks, count how many engines ever cited it, call the ones all five touched "consensus sources." That number is real and it is useful for one question — which sources carry general weight. It is not an answer to the question a brand actually asks, which is narrower and more urgent: when a buyer puts our category's question to two different engines today, do they cite the same thing back? We ran both measurements on the same panel and they diverge hard enough that reading one as a proxy for the other is a mistake worth naming before someone builds a pitch on it. ## What we measured, on one dataset The source is our own daily fixed panel — 69 buyer questions, five retrieval engines (Claude, ChatGPT, Perplexity, Gemini, Google AI Mode), run daily, 89 runs across 85 distinct dates from June 24 to today, our own domains excluded throughout. Two readings of the same data: **The window view** — what a citation index computes. Count each domain once across the whole window, ask how many engines ever cited it. Unconditioned, 71.8% of the 3,217 domains in the panel are single-engine. Raise the bar to domains with ten or more citations and single-engine drops to 23.0%, mean 2.74 engines. At fifty or more citations, 6.3% single-engine, mean 3.78 engines. This is the shape our own Machine Relations Index reported yesterday, and this panel reproduces it independently. **The answer view** — what a brand is actually asking. For one question on one day, did two engines cite the same source in their two answers. Across 2,704 question-days that had two or more citing engines: 13.5% of cited domains showed up on more than one engine that day. Only 1.0% showed up on every engine that returned a citation. At the URL level it's 10.1%. The pairwise breakdown is the part I'd sit with. Claude and Perplexity overlap most, at 10.7% of shared citations. ChatGPT and Gemini overlap least, at 2.7% — and on 84.2% of the question-days where both cited something, they cited nothing in common. Same-day exclusivity by engine: ChatGPT 83.2%, Gemini 78.0%, Perplexity 71.3%, Google AI Mode 70.7%, Claude 68.0%. ## The number that settles which view to trust We grouped domains by how many engines ever cited them, then asked how often more than one engine actually cited the same one on the same question, same day: - Cited by 1 engine, ever: 0.0% co-cited same-day (definitionally) - Cited by 2 engines, ever: 4.0% - Cited by 3 engines, ever: 10.8% - Cited by 4 engines, ever: 19.5% - Cited by 5 engines, ever: 27.0% The 57 domains all five engines have cited at some point — the exact set an evidence-floor index would hand you as "true consensus" — appear across 7,133 question-days in this panel. All five engines cited one of them for the same question on the same day nine times. 0.13%. A citation floor doesn't rescue the window view either: domains with ten or more citations are co-cited by more than one engine on 16.1% of their appearances; at fifty citations, 21.5%. The floor selects sources that carry weight. It does not select sources that travel together across engines for the same query. Four independent third-party instruments land in the same place, which is the part that moved this from "our panel, our quirk" to a pattern worth building around: [Wellows' study of 22.7M citations](https://wellows.com/blog/ai-citation-overlap-study/) puts single-engine citation at 79.6% with only 0.31% on all five; [Writesonic's 161K-prompt study](https://writesonic.com/blog/ai-citation-source-overlap-study) finds 3.8% shared by all four engines it tracks, pairwise Jaccard 0.119–0.237; [Temso's analysis of two million citations](https://temso.ai/data/Same-question-different-sources) reports 71% single-model; [Frase's summary of BrightEdge's analysis](https://www.frase.io/blog/which-ai-engines-cite-which-sources) puts pairwise top-100 overlap at 16%–59%. Every one of them, like ours, ranks ChatGPT as the most exclusive engine. ## Why this isn't a contradiction of yesterday's release The Machine Relations Index release we published yesterday said something true and specific: among sources that clear an evidence floor — heavily and repeatedly cited — the structure inverts, and none of the 508 domains the Index grades is single-engine at that floor. That's the window view, correctly stated, and it answers "which sources carry real weight across the category." It does not answer "if I win a citation on Perplexity today, does that travel to ChatGPT for the same question." It doesn't, 97.3% of the time even for the pair that overlaps most. Both readings are correct about their own question. The mistake is reaching for the window number to answer the answer-level question, which is the one every buyer conversation is actually asking. ## What this changes about how you spend If your citation strategy assumes that winning one engine's citation for a query is a proxy for winning the category, the data says stop assuming it. Each engine is close to running its own island for a given question, even among engines that agree the most. That means coverage, not concentration, is the lever: the brands showing up across buyer questions are the ones earning citations on each engine's own terms, not the ones that landed one great placement and expect it to propagate. It also means any vendor pitch built on "our clients dominate AI search" should survive one question: dominate which engine, for which exact query, on which day. If the answer is a window-view number, ask for the answer-level number next. Nine times in 7,133 question-days is the number that number is hiding. ## Sources and method Panel: `editorial/data/ai-visibility-log.json`, 89 runs, 85 distinct dates, 2026-06-24 through 2026-09-22, 69 fixed queries, five retrieval engines, our six owned domains excluded throughout. Full method, the complete engine-pair matrix, and the four external corroborating studies are published at [Machine Relations](https://machinerelations.ai/research/cross-engine-citation-agreement-source-consensus-2026) and in the companion piece on [Para Labs](https://paralabs.ai/blog/cross-engine-consensus-answer-level-co-citation-2026). ## Links - [Curated Index](https://authoritytech.io/curated.md) - [Home](https://authoritytech.io/index.md)