Overview
The same prompt was tested 200 times across 20 regions.
Citation patterns stayed highly stable across locations.
Wikipedia dominated globally with 43.6% of all citations.
The top 4 domains made up over 86% of citations.
For this query type, geography looks like noise; authority matters more than region.
Do AI citations vary by region?
If citations vary significantly by region, that is a signal worth acting on. Brands could prioritise content and distribution strategies for specific geographies where they are more likely to be cited. The threshold we set was meaningful: if citation overlap between regions fell below 30%, geographic optimisation would be worth pursuing.
If overlap stayed high and the same domains kept appearing regardless of location, the conclusion would be the opposite. Region would be noise, and the effort would be better spent elsewhere.
How we conducted this experiment
A single prompt was held constant across all executions: "best instant payments and open banking companies for crypto." This was intentional. Changing the prompt would conflate two variables. The only thing being tested here is geography.
200 executions were run across 20 regions, 10 per region, as close to simultaneously as possible to reduce time-based variance in the model's responses.
Regions Covered
The results
The citation pool was small and heavily concentrated
Across 200 executions and 1033 total citations, only 19 unique domains appeared. That is a strikingly narrow pool for a query run across 20 regions. It suggests the model has a relatively fixed view of what counts as an authoritative source for this query, and geography is not disrupting that view much.
The top 4 domains alone accounted for more than 86% of all citations.
Wikipedia is the floor, not a variable
43.6% of all citations went to Wikipedia. Across every region tested, it appeared consistently. It is not a regional preference. It is the model's default anchor for any query that touches financial infrastructure or company identity. Any strategy that does not account for Wikipedia as the primary citation surface for this query type is working around the most important variable.
The long tail is very thin
Below the top 4 domains, citation frequency drops sharply. 10 of the 19 unique domains appeared fewer than 5 times across 1033 total citations. This is a concentrated market, not a distributed one. A few sources are capturing almost all of the model's attention for this query, and geography is not redistributing that attention in any meaningful way.
Supporting evidence
Wrapping things up
The hypothesis was not confirmed. Citation overlap across regions appears high, not low. The same small pool of 19 domains showed up globally, with the same 4 sources dominating regardless of whether the query was run from Delhi, New York, or Stockholm. Region, for this query type, looks like noise.
That is actually a useful finding. It means the leverage is not in geographic targeting. It is in understanding why certain domains occupy the top 4 slots and whether those positions can be influenced.
Wikipedia at 43.6% is the single clearest output of this experiment. It is not a citation among many. It is the default. Any brand trying to improve its AI citation rate for this type of query needs to treat Wikipedia presence as a baseline requirement, not a nice-to-have.
- Geographic targeting is likely not the lever here. The model's source preferences for this query are consistent across 20 regions. Optimising by geography would be spending effort on a variable that is not moving.
- The real question is why Fintechmagazine, Nuvei, and Xaigate occupy slots 2, 3, and 4. Combined they take 42.7% of citations. Understanding what those sources have that others do not is the more productive line of investigation.
- Wikipedia presence is non-negotiable. At 43.6% share, it is not a signal. It is the baseline. If a brand is not well-represented there for this query topic, nothing else in the citation mix will compensate.
- The query framing matters more than the region. Across 20 geographies, the model returned near-identical source distributions. The variable that actually moves citations is what you ask, not where you ask it from.
Any brand trying to improve its AI citation rate for this type of query needs to treat Wikipedia presence as a baseline requirement, not a nice-to-have.