Report
The same query, ten countries
How little the result sets have in common once you cross a border.
Measured 2026-W36 · 2026-08-31 · google.com, gl=us
2026-W36
Overlap with the United States
| Country | Parameter | AI Overview | Overlap |
|---|---|---|---|
| France | gl=fr | 53.0% | 50.2% |
| India | gl=in | 53.0% | 50.2% |
| United Kingdom | gl=gb | 53.0% | 50.4% |
| Germany | gl=de | 53.0% | 50.5% |
| Spain | gl=es | 53.0% | 50.5% |
| Canada | gl=ca | 53.0% | 50.8% |
| Australia | gl=au | 53.0% | 51.0% |
| Brazil | gl=br | 53.0% | 51.1% |
| Japan | gl=jp | 53.0% | 51.2% |
Overlap is a Jaccard coefficient over result domains. A rank tracker configured for one country is not measuring a slightly different page — it is measuring a substantially different one.
Method
The measurement behind it
A fixed corpus of 4,185 queries, run once a week from the United States through one endpoint, stored and never re-fetched. 82,620 observations across 12 weeks so far. The method is written out in full on how we measure.
curl --request GET \
--url 'https://serpens.p.rapidapi.com/api/v1/search_ai_overview?q=best%20search%20apis&gl=us&hl=en&format=json' \
--header 'x-rapidapi-host: serpens.p.rapidapi.com' \
--header 'x-rapidapi-key: YOUR_KEY'Reproduce any figure here
Free tier, no card. Same endpoint, same parameters — the corpus is the only thing you would need to supply.