Our goal is to determine the market share of general search engines around the world.

Cloudflare accelerates and protects traffic for large parts of the Internet, with a global network that spans more than 200 cities in over 100 countries. Approximately 25 million Internet properties rely on Cloudflare's network to reach their audiences. More than 10% of all websites connect through our reverse proxy, including 17% of the Fortune 1000.

Due to the size of our network we are in a unique position to measure statistically relevant market share data.

Our methodology uses the referer header. Whenever a link is clicked from a given search engine, that search engine is identified by this header. For this analysis, we only use the referer origin, and not any referer paths or query strings.

Our methodology uses the user-agent header parsed using mainly uap-core 7e634c8 in combination with a proprietary parser. In Radar page only proprietary parser is currently being used and thus the results are different.

• We use an anonymized data sample from our reverse proxy.
• We remove bot traffic that is automatically identified by Cloudflare systems.
• We remove sub-resource requests (JavaScript, fonts, etc.) using the content-type response header.

• This methodology assumes the average amount of link clicks conducted in a search engine visit is the same across search engines, as we are measuring link clicks and not searches.
• This methodology assumes that the traffic Cloudflare sees is representative of the whole in terms of link clicks from general search engines, which seems appropriate given the wide breadth of our customer base (many different types of websites use Cloudflare).
• This methodology assumes that the referer header remains intact. In some browsers, it is possible for a user to disable the referer field. We expect this to only happen in rare cases. In such cases that field will come up empty.
• This methodology cannot account for Chrome's NoState Prefetch which tries to predict which result a user will click on and prefetch it. These count as searches for Google even if the user doesn't click as predicted.
• The higher the volume of data the more reliable this methodology becomes.
• This report takes into account only the following operating systems: iOS, Android, Desktop (Windows, MacOSX, Linux, ChromeOS), Mobile (WebOS, WindowsPhone, Blackberry, SmartTV, Nintendo, Xbox, Playstation, Kindle, Android, iOS). Anything that is not identifiable as being into one of those categories is excluded.

This report take into account only the following operation systems:

• iOS
• Android
• Desktop (Windows, MacOSX, Linux, ChromeOS)
• Mobile (WebOS, WindowsPhone, Blackberry, SmartTV, Nintendo, Xbox, Playstation, Kindle, Android, iOS)

Anything that is not identifiable as being into one of those categories is excluded.

Click here to download the report data as CSV files.