What's My User-Agent

Because of a compatibility habit from the 1990s that nobody has been able to break. Early servers checked for Mozilla before sending a page that used frames, so other browsers began claiming the name to be served the better version, and every browser since has copied the one before it. The token now carries no information at all: Chrome, Firefox, Safari and Edge all open the same way, and Chromium strings additionally drag along AppleWebKit, KHTML, like Gecko and Safari, none of which say anything useful about the browser in front of you. What actually identifies it is the more specific token nearer the end of the line.

Chromium browsers offer it in developer tools, under the device toolbar or the network conditions panel, where you can pick a preset or type your own. Firefox exposes general.useragent.override in about:config, and Safari has a User Agent menu once the Develop menu is switched on. Extensions and command-line flags do the same thing more permanently. An override set that way changes what scripts on the page see as well as the request header, so reloading this page afterwards shows the substituted string — a quick way to confirm the override is live. Spoofing rarely fixes a site that has refused you, and an unusual string attracts more attention than the one it replaced.

Not yet, and not everywhere. Client Hints are Chromium's replacement: a set of Sec-CH-UA request headers plus navigator.userAgentData in the page, splitting what used to be one line into low-entropy values sent by default — brand list, a mobile flag, platform name — and high-entropy ones such as full version, architecture and device model that a site has to ask for. Firefox and Safari have not adopted them, so anything that wants to work everywhere still reads the old string, which is also why Chrome now freezes most of its version digits at zero. This page reads only navigator.userAgent, so what appears is the legacy line rather than the hints.

The value is read once from navigator.userAgent inside your own tab, the four rows below it are worked out by expressions in the same component, and there is no request behind any of it — no upload, no API route, nothing written to storage. It sits in component state and disappears when the tab closes. Worth separating from that: your browser sends this string as a header to every site you load, including the one serving this page, so it is not a secret you are guarding. The point is that nothing further is done with it here, and the copy button hands it to you rather than to anyone else.

On its own, almost never — a current Chrome-on-Windows string is shared by an enormous number of people, which is precisely why it carries little identifying weight. It becomes useful to a tracker only in combination: screen dimensions, timezone, language, installed fonts, canvas rendering quirks and a dozen other weak signals multiply into something close to unique. That is also why the intuitive defence backfires. A rare or self-invented string narrows the crowd you are hiding in instead of widening it, and a claim of Firefox on Linux coming from a browser that behaves like Chrome is itself a distinguishing signal.

Because those browsers say Chrome on purpose. A distinctive token invites sites to treat you as an unknown quantity and serve a degraded page, so several Chromium browsers either keep the Chrome token alone or hide their own name behind it. The chain here tests Edg/ first, then OPR/, then Firefox/, then Chrome/, so Edge and Opera are caught by the tokens they do declare, while anything else built on Chromium falls through to Chrome. For most purposes that is the useful answer anyway, since the engine and the behaviour genuinely are Chrome's.

The chain has five branches and no fallback, so anything none of them match ends up there. A build branded Chromium is deliberately excluded from the Chrome branch, and it carries no Version/ token either, so it misses Safari as well. The same happens to embedded webviews and in-app browsers that trim their strings, to command-line clients such as curl, to crawlers and bots, and to televisions and consoles. The version is left blank rather than filled with a guess, so the row reads Unknown on its own — while the raw line above stays complete, which is the part worth copying into a report.

For logging and triage, yes. For deciding what to send someone, no. A test for Mobi, Android, iPhone or iPad is the standard shape and it is wrong in both directions: a tablet asking for the desktop version of a site vanishes from it, a Windows laptop with a touchscreen never appears in it, and a narrow desktop window is not caught at all. Layout questions are better answered by viewport width in a media query, input questions by pointer and hover queries, and capability questions by testing for the feature itself. Keep the string for describing what happened after the fact.