Cross-platform citation concordance

Why concordance matters

Suppose AI search engines mostly agreed on what to cite for a given question. Then "AI visibility" could reasonably be treated as one single target. The available evidence suggests the opposite. Engines appear to disagree substantially.

That would mean a strategy optimized for one engine's citation pattern transfers poorly to another's. Quantifying exactly how much disagreement exists, and whether it varies by query type, turns a widely repeated impression into a measured fact.

What is known so far

This figure is Partial . It's a real, named vendor finding, but it comes from a corpus that isn't independently reproducible. It's the single best available estimate today. It's also the reason a first-party, reproducible version is worth building.

Defining the metric

Cross-Platform Concordance is measured as a Jaccard index: intersection over union, of cited domain sets between two engines, for the same query. That gets averaged across the full query set. A score of 1.0 would mean two engines always cite an identical domain set. A score of 0 would mean no overlap, ever.

A worked calculation example

Here's the formula applied to an invented, illustrative single query. Say the question is "best project management software for small teams." ChatGPT cites four domains for it: A, B, C, D. Perplexity cites five domains for the same question: B, C, E, F, G.

The intersection, domains cited by both, is {B, C}: a count of 2. The union, all distinct domains cited by either, is {A, B, C, D, E, F, G}: a count of 7. The Jaccard index for this one query is 2 ÷ 7, or roughly 0.29. Averaging that same calculation across the full 1,000-query set, not just one example, produces the overall concordance figure this study reports for each engine pair.

Study design

This falls directly out of the Citation Index's core collection. Every engine already runs against the same published query set. So concordance needs no separate data collection, only a different analysis of data already being gathered.

Pre-registered hypotheses

The eventual output: a concordance matrix

With seven engines tracked, the final output isn't a single number. It's a 7×7 matrix of pairwise concordance scores. One cell per engine pair, readable at a glance for which pairs behave most alike, and which behave least alike.

A matrix format also makes it easy to test hypothesis CC3 directly. That means comparing the AI-Overviews/AI-Mode cell against every cell involving a non-Google engine. A single blended average would hide exactly this kind of structural pattern instead of revealing it.

What this means for strategy, if confirmed

Low concordance would mean a brand serious about AI-search visibility likely needs a platform-specific strategy, rather than one blended approach. That directly reinforces the case made in the measurement standard for reporting citation metrics per-engine, rather than as one composite score.

A practical read on low concordance

Say the final numbers land near the vendor-reported 11%. What should a site owner actually do with that? First, stop optimizing for a single "AI visibility" score. It hides more than it shows once agreement between engines is this low.

Second, track citations per engine, not as one blended total. A page cited heavily by Perplexity but ignored by ChatGPT is a genuinely different result from a page ignored by both. A single combined number can't tell those two situations apart. Third, expect that a tactic proven to work on one engine may need separate testing on another, rather than assuming it transfers automatically.

Limitations

• Concordance measures agreement, not quality — two engines could agree on a poor source, or disagree while both citing good ones.
• The result depends heavily on the query set — a different set weighted toward more subjective or more factual queries could shift the overall figure meaningfully.