Browser Fingerprint & Privacy Leak Inspector

Fingerprinting without cookies

Cookies require explicit storage: a tracker must write a file to your browser and read it back later. You can delete cookies, block third-party cookies, or browse in private mode. Browser fingerprinting requires none of these conditions. It works by querying APIs that every browser exposes by design, including your screen dimensions, installed fonts, GPU renderer, audio pipeline behaviour, and dozens of other signals that your browser reports to any website at any time without warning.¹

Each individual signal is weak. Knowing your screen resolution alone tells a tracker almost nothing because millions of users share 1920x1080. But combine it with your GPU renderer, your audio fingerprint, your installed fonts, your timezone, and your language list, and the intersection becomes very small. At 20 or more bits of entropy, a fingerprint is specific enough to re-identify a single browser across sites, sessions, and private windows without ever touching your storage.² That is why clearing your cookies alone does not stop tracking when your browser fingerprint stays the same.

Where the law stands

The consent rulebooks grew up around storage. A cookie banner speaks to the files a site writes to your device and reads back later, because regulators drafted these frameworks when a stored record was the normal tracking mechanism. Fingerprinting fits that model awkwardly, since it queries APIs that any page may query by design and writes nothing anywhere. Regulators have caught up regardless. The UK's Information Commissioner's Office, for example, now lists fingerprinting among the storage and access technologies its rules cover, and applies them where the technique stores or accesses information on your device.³ Because each authority draws its own lines, the treatment of fingerprinting under consent rules varies by jurisdiction and has tightened over time. You cannot assume a cookie banner covers it.

This page can settle the technical questions for you. It cannot settle the legal one, because the binding answer comes from whichever regulator governs the sites you visit, and those authorities do not speak in one voice. What you can take from here is the shape of the question: storage-era consent rules, an API-era technique, and a gap between the two that each jurisdiction fills differently. For anything with real consequences, consult the guidance your own data protection authority publishes, since it carries the weight this inspector never will. No paragraph above is legal advice, and you should not read any of them as such.

Private windows, same fingerprint

Does a private or incognito window change any of this? No, and the mechanics explain why. A private window runs the same engine, the same GPU, the same fonts, and the same screen as a normal window of the same browser, so every signal this inspector reads comes back identical. What private mode actually clears is the storage layer: cookies, history, and cache vanish when the window closes. That is exactly the layer fingerprinting does not need. The two simply do not overlap, which is why the paragraphs above describe fingerprinting as tracking that never needs to touch your storage.

The comparison tip in the entropy section below leans on that carry-over deliberately. When you run this inspector in a private window, the entropy score lands close to the normal-window score, because the signals underneath never changed. The gap worth chasing lives somewhere else: a hardened browser that randomizes or suppresses the high-weight signals produces a visibly lower number, while privacy mode on an ordinary browser barely moves it. Run both comparisons once each and the difference between the two numbers tells you where your protection actually comes from, which is the browser's configuration rather than the window's mode.

Check What Your Browser Fingerprint Reveals

The inspector shows the same signals a tracking script reads, and each one on its own looks harmless: a screen size, a time zone, a list of fonts. The guide below explains how those small values combine into an identifier, how a fingerprint differs from cookies and local storage, what it can and cannot reveal about you, how ad tech and fraud prevention use it and what the entropy numbers in published research actually mean.

Your browser silently shares: User-Agent, screen resolution, installed fonts, timezone, language, WebGL renderer, canvas rendering, audio fingerprint, installed plugins, Do Not Track setting, and more.

A website combines dozens of signals like User-Agent "Mozilla/5.0...", screen "2560x1440", timezone "America/New_York", and WebGL renderer "ANGLE (NVIDIA...)". Together these create a fingerprint unique enough to identify you among millions.

Site A and Site B both embed the same ad network. The ad network reads your fingerprint on both sites and links your visits even though you never logged in or accepted cookies.

The ad network generates a stable fingerprint hash from your browser signals. On Site A it records hash "a3f8c1...". When you visit Site B, the same hash appears, linking your activity across both sites without cookies.

Canvas & WebGL explained

Canvas and WebGL fingerprinting are two of the most widely used tracking techniques because they tap into rendering pipelines that vary subtly between otherwise identical browsers. By drawing a fixed image and hashing the pixels, or by reading the GPU renderer string directly from the graphics driver, a tracker can derive a stable identifier that survives cookie clears and private browsing sessions.

How the canvas hash is built

The canvas fingerprint draws a specific test image (text at two font sizes, a filled rectangle, and an emoji) onto an offscreen canvas and then hashes the raw pixel output. Different operating systems, GPU drivers, and font renderers produce slightly different pixel values for the same drawing commands. The result is a short hex hash that stays stable across page loads on the same machine but differs across most other configurations, which is what makes it useful for tracking without cookies.⁴

WebGL fingerprinting reads the unmasked GPU vendor and renderer strings via the WEBGL_debug_renderer_info extension. This reports the actual GPU model name, for example ANGLE (NVIDIA GeForce RTX 4070 Direct3D11 vs_5_0 ps_5_0), directly from the graphics driver. Because GPU models are specific and the driver version is often included, this string narrows the fingerprint to a very small population even before combining it with other signals.⁵

What the entropy score means

Shannon entropy, measured in bits, describes how much information a set of signals carries. One bit of entropy halves the candidate pool: a signal worth 1 bit means half of all browsers share your value, so you are narrowed to half the population. Ten bits means 1 in 1,024 share your exact combination. Twenty bits is 1 in roughly 1 million. At 22 bits the candidate pool is smaller than most cities.

The score here is an approximation based on weighted signal contributions. Canvas hash, audio fingerprint, and WebGL renderer each carry the most weight because they vary the most between devices. Signals that return unknown or blocked contribute nothing, so a browser that suppresses these signals actively lowers its entropy score.² Browsers that resist fingerprinting by reporting generic values end up blending into a larger crowd of identical users.

A concrete score makes the scale easier to read. A combined entropy reading of 24 bits sits above the 22-bit mark where the candidate pool shrinks below the population of most cities. That score alone can re-identify a single device without a cookie in sight. Dropping below 20 bits usually takes more than one fix, since a tracker recombines every remaining signal until a comparable pool re-emerges.

How to reduce your fingerprint

Reducing a browser fingerprint is less about hiding any single signal and more about blending into a large group of identical users. Privacy tools that randomise or standardise the high-weight signals while leaving the browser otherwise functional tend to be far more effective than piecemeal tweaks that leave the distinctive combinations intact.

Joining a large crowd

The most effective approach is to join a large crowd rather than hide individual signals. Brave with shields enabled randomizes values from browser and OS feature APIs, including canvas and Web Audio outputs, so the hash can change by site and session.⁶⁷ Firefox with privacy.resistFingerprinting set to true in about:config reports generic spoofed information for data commonly used for fingerprinting, such as CPU cores, timer precision, and the local timezone.⁸

Blocking a single signal in isolation, such as spoofing just the user agent string, rarely reduces trackability in practice. Trackers cross-reference all available signals and weight them by stability. A spoofed user agent combined with an unchanged canvas hash still produces a distinctive fingerprint. Effective reduction means suppressing the high-weight signals (canvas, WebGL, audio, and fonts) simultaneously rather than one at a time.

Hardware-level signals like screen resolution and hardware concurrency are harder to spoof without breaking site layout or performance. Using a standard monitor resolution such as 1920x1080 and a common CPU core count reduces but does not eliminate the contribution of those signals. Pairing a common configuration with a privacy-hardened browser is the most reliable way to shrink your fingerprint without breaking the sites you visit.

What a VPN does and does not change

A VPN changes your IP address and the rough geography that travels with it, and it changes nothing else on this page. That split deserves a slower look. The canvas hash comes from your GPU's rendering, the font list comes from the software installed on your machine, and the renderer string comes from your graphics driver. A tunnel between your browser and the internet touches none of those, because the machine itself produces every signal this inspector reads, not the route your traffic takes. Swap the route freely; the fingerprint stays put.

From a tracker's point of view, the two identifiers arrive as columns of one table. Your VPN changes exactly one column, the IP address, while the fingerprint columns keep identifying the browser, so the route change alone buys less anonymity than it appears to. The honest pairing runs both fixes at once: a hardened browser for the fingerprint side, a changed route for the IP side. Neither stands in for the other. The entropy score here reads nothing from your IP address either way; it measures machine-side signals only, so a VPN will never move this number.

Which signals trackers weight most heavily

Trackers do not treat every signal equally. Some signals vary wildly between devices and carry a lot of entropy; others are shared by millions of users and do little to narrow the candidate pool. Understanding the hierarchy helps you decide which signals to suppress first in order to get the largest reduction in uniqueness.

The highest-weight signals

Not all fingerprinting signals carry equal weight. Privacy guidance treats active fingerprinting as code that observes browser, device, and context characteristics, then recommends limiting APIs to the entropy they truly need.¹ Canvas hash and WebGL renderer sit at the top because they vary by GPU model, driver version, and operating system simultaneously, producing values that narrow the candidate pool before combining with other signals.⁴⁵ Audio fingerprint ranks next: research shows Web Audio vectors can add entropy to existing fingerprinting mechanisms, even when individual audio outputs vary across repeated attempts.⁹

Installed fonts occupy a different tier. The presence or absence of specific fonts correlates with the software installed on a system: a machine with Adobe Creative Cloud has different fonts than a stock Windows install, and those differences appear consistently in font-probing results. W3C notes that installed font lists can fingerprint a user because the default set varies by user agent, platform, and locale.¹⁰ Timezone and language list sit lower on the hierarchy but rarely appear alone; trackers combine them with the high-weight signals to resolve ambiguity when two users share the same GPU or font set.

Screen resolution and hardware concurrency carry the lowest weight of the commonly collected signals. Millions of users share 1920x1080 or 2560x1440 displays, so resolution alone distinguishes almost no one. Similarly, 8-core and 12-core CPUs are extremely common. The rows marked TRACKABLE in the inspector flag the highest-weight signals; suppressing those first gives you the greatest entropy reduction with the least impact on normal site functionality. Addressing lower-weight signals without fixing the high-weight ones leaves the fingerprint nearly as distinctive as before, because trackers discard redundant low-entropy signals once the high-entropy ones are sufficient to make a unique identification.

The signal this tool does not read: WebRTC

One whole signal class sits outside this scan: WebRTC. The inspector reads canvas, WebGL, audio, fonts, and the roughly sixty signals the sections above walk through, and it opens no WebRTC connection at any point while scanning. WebRTC is a different leak class. Browsers use it for voice and video calls, and in some configurations its peer-connection machinery can expose local IP addresses to a page, which makes it a network-layer signal rather than a rendering one. The two leak families barely resemble each other, so this tool deliberately stays on the rendering side of the line and leaves that column unread.

Whether your setup leaks through WebRTC depends on your browser's handling and, if you use one, on your VPN's WebRTC support. That check belongs to a WebRTC-specific test or to your browser's own settings rather than to this page. The score here simply excludes the column, so a LOW tier on this page never certifies a leak-free browser on its own; it certifies a machine-side fingerprint that blends into a crowd. Read the two answers side by side when your threat model cares about both, because they respond to different questions about the same machine, and only one of them concerns this tool.

The user agent's shrinking role

The user agent row is living through a deliberate decline. Browser engines are reducing the classic string to cut its tracking value: platform versions and device models freeze to fixed placeholders, and minor build numbers collapse to zeros. The distinguishing details now live in a separate client-hints mechanism that a page must explicitly request before your browser returns them, so fewer identifying details reach every site by default.¹¹ A tracker reading only the default string from a modern engine learns less with each passing year, and that shrinking payload is exactly the point of the exercise.

This row reports the classic string and only the classic string. The scan makes no client-hints request at any point, so the USER AGENT row shows exactly what your engine still discloses to every site by default. On engines that have applied the reduction, the row reads short and generic, and that is the privacy work doing its job rather than a broken result. The trade-off sits in the score: the row carries a fixed weight in the approximation the entropy section above describes, and as the strings converge across engines, that fixed weight increasingly overstates the signal's real contribution to your uniqueness. Treat the row as a snapshot of the default disclosure, not as the ceiling of what your engine can reveal on request.