
10 ways websites are tracking you right now
15 min readUpdated
Websites can link visits using cookies, accounts, technical browser characteristics, and page events. Collection depends on embedded tools, settings, and permissions: a capability does not prove that a particular site records everything listed here.
Sources checked on October 31, 2026. Below are ten ways to collect and link data, their limits, and measures to reduce tracking. We studied documentation, but did not investigate the reader’s sites, devices, or traffic. Separate sections cover the use of data for manipulation and risks of centralized data storage.
Cookies and other browser identifiers
Cookies can link requests to a session, preference, or account. Not every cookie serves advertising.
- A session identifier helps maintain a session; its lifetime depends on implementation.
- Persistent cookies may be used on later visits while valid and accessible to the browser.
- localStorage and IndexedDB are separate mechanisms for storing site data.
Clearing cookies and deleting server records are different actions. Signing in again may reconnect a visit to an account; that does not prove that the old cookie was restored.
| Metric | Result | max | avg | min |
|---|---|---|---|---|
| Ping | —ms | — | — | — |
| Download | —Mbps | — | — | — |
| Upload | —Mbps | — | — | — |
Browser fingerprinting
Fingerprinting compares a combination of available browser and system characteristics. These may include:
- Browser and OS version, language, and time zone
- Screen and font characteristics
- Available browser features and settings
Matching characteristics do not establish identity: results depend on the visitor population, changes to the environment, and browser defenses. Clearing history alone does not change every technical characteristic, but reliable recognition is not guaranteed. These limits are discussed in the W3C working group guidance.
Canvas Fingerprinting
A script can draw graphics on a Canvas and read the result. Differences in the graphics stack and image processing may provide additional matching signals:
- Graphics stack
- Drivers
- Rendering and smoothing
This does not guarantee a unique identifier for every person. A browser may restrict or change the readback: WebKit describes adding noise to Canvas when Safari’s advanced fingerprinting protection is enabled. This example concerns the documented feature, not every browser or release.
Audio Fingerprinting
Web Audio can compute a sound signal and analyze the processing result. With OfflineAudioContext, data can be generated without playing sound through speakers; this is not a recording of a voice through a microphone.
Results may be affected by:
- Hardware environment
- Software environment, including drivers
- Signal processing
Browser defenses also change the result: WebKit describes adding noise when audio data is read. Being undetectable, stable, or impossible to block is therefore not a property of every such script.
Behavioral fingerprinting
A page script may collect events:
- Pointer movement
- Scrolling
- Intervals between input events
- Action sequences
These events may support behavioral analysis or risk assessment. Observing events and reliably establishing identity are different results. We have not measured the accuracy of such matching in this article. Clearing local data does not delete records already sent, but the profile should not be described as permanently unchangeable. Matching models are discussed in the W3C guidance.
How fingerprinting differs from cookies
Compare specific mechanisms:
- Private browsing does not universally guarantee the absence of fingerprinting; browser defenses differ.
- Deleting cookies does not necessarily change system characteristics.
- Changing IP addresses does not necessarily change other characteristics.
- Detection and mitigation depend on the browser and collection technique.
Without comparable data, fingerprinting cannot be ranked as always more dangerous than cookies or described as standard across all ad networks. The W3C guidance considers different threat models, rather than one ranking of methods.
Logging every request and click
A server may record request information. For example, the Nginx log format is configured by its operator. Distinguish:
- The connection address and an estimate of its region
- Available browser and device headers
- Time and requested resource; a click within a page does not necessarily create a separate record without event collection
- Referral information within the data the browser sends
IP-based location is an estimate, not a person’s exact address. Check the contents and retention period with the particular system.
For example, Clarity specifies 30 days for session recordings, with favorites and randomly selected samples retained for up to nine months. This is one product’s policy, not the retention period of every site.
Pixels and third-party trackers
A page may load another provider’s resources:
- Advertising pixels
- Analytics and heatmaps
- Chat, form, and social service widgets
If a request takes place, the recipient receives the parameters and network information transmitted in that connection. Not every widget is an advertising tracker, and not every request succeeds: implementation, consent, and blocking matter. Linking data across sites requires available identifiers or other matching signals; an image’s presence alone does not establish it.
For example, Google’s policy describes receiving information from sites and apps that use its services. This describes one provider, not identical transmission by every widget.
Behavioral tracking on the page
If an event collector is installed, it may record pointer movement, scrolling, clicks, and event times. Time between events does not directly measure where a person looked or what they thought.
For example, Microsoft describes Clarity recordings as a replay of page data and actions, not a video of the entire screen. The product supports content masking; that does not guarantee every collector hides all sensitive data.
Heatmaps and experiments show selected interaction metrics. They do not automatically establish an individual visitor’s intentions.
Forms and hidden fields
A form’s hidden field may contain an identifier, campaign parameter, or security token. It is not visible in the normal page view, but is not itself a way to obtain location.
Page code may add event times, language, and other information from available sources. Location through the browser API requires the relevant permission. A hidden value does not mean every site collects all these data.
A form may transmit more than its visible text fields show. Assess this from the implementation and site policy, not from the mere presence of a form.
Location and network data
Distinguish the sources:
- An IP address can support a country or region estimate; city accuracy varies and does not establish a person’s exact address.
- Connection type information depends on supported APIs; not every site can automatically identify a corporate network.
- Speed and latency require measurement of a particular exchange; they are not a complete description of the connection.
- Provider and ASN information concerns the network address, not proof of identity.
- IPv6 alone is not a leak.
- WebRTC may expose addresses depending on routes and browser policy; this does not happen identically in every configuration.
IP location limits are described by MaxMind, and limited support for connection information by MDN’s Network Information API documentation. Address exposure through WebRTC is discussed in RFC8828.
The Geolocation API requests permission. Accuracy depends on available sources and the device; neighborhood or building accuracy is not guaranteed.
Logins via accounts and social networks
With an external login, a site receives an identifier and permitted information within that login method’s scope. For Google, email/profile information depends on requested permissions and available fields; a name and photo are not guaranteed. Apple describes an option to share an email address that hides the main address.
Signing in links the visit to that site’s account. It does not automatically give every other site the main account’s password or full history. Removing a login link and deleting previously received data are different actions.
Notifications and off-site interactions
A push subscription may allow messages after a tab closes. It does not mean continuous observation of every action on the device.
A sender may record visits through tagged links and available notification events. Loading an image in an email does not prove that the message was read: Apple Mail Privacy Protection may load remote content in the background, regardless of interaction with the email.
Such signals may inform later messages, but the exact reaction profile and observation scope need to be established for the particular product.
Cross-device analytics
A shared account allows a service to link actions performed under it on different devices. Matching by technical, network, and timing characteristics is a different mechanism: matches may wrongly combine different people or fail to recognize one person.
Similar characteristics and one network address do not by themselves prove common ownership. Whether an action affects advertising on another device depends on the system and settings; it is not an inevitable result of every visit. One provider’s sources and settings are described in Google’s policy.
Why cookie banners do not solve the core problem
Rejecting non-essential cookies limits operations for which the site correctly respects that choice. It does not prohibit every technical request or delete server records.
The label “anonymous analytics” does not determine the data’s legal status. Consent, exemptions, and processing requirements depend on territory, purpose, and implementation. For example, CNIL describes a limited audience measurement exemption in France, subject to conditions; it is not permission for any site to track across sites without consent.
Consent banners as tools for “shaping expectations”
Check how clearly consent and refusal are presented:
- Is refusal hidden behind a less visible element?
- Do fragmented settings make the choice hard to understand?
- Does refusal require extra actions compared with consent?
In its report, the FTC includes interfaces that obstruct privacy choices among dark patterns. A banner’s appearance does not itself show which data are sent after a choice. A widget or fingerprinting capability does not establish ongoing transmission without checking the implementation.
Realistic view: what actually changes when you refuse cookies
With a correct implementation, a choice may limit:
- Non-essential cookies
- Associated matching
- Data transmission
The size of the change depends on the system. Separately account for:
- Necessary requests
- Server records
- Account sign-in
Their continued presence does not mean refusal is entirely useless or that every remaining collection operation is lawful.
For example, Clarity describes reduced functionality without cookie consent, including the absence of a continuous user journey and full session replays. These are one product’s conditions, not identical behavior by every banner. A button’s wording alone cannot measure remaining tracking.
Where the data goes: your ad profile
Advertising systems may use data within the available sources and settings:
- Search queries
- Site visits
- Content interactions
- Location information
- Devices and apps
- Purchases
One company does not automatically receive all six categories about everyone. For example, Google describes its own sources and settings in its policy. That does not prove a unified profile available to every site or a record of every financial transaction.
Cross-platform data linking
Accounts, shared identifiers, and site/app integrations may help link data. Distinguish the presence of those signals from permission or the technical ability to pass information to another party.
A login, the same device, or an integration does not mean all platforms automatically merge search, purchases, and offline life into one history. For a particular operator, check sources, recipients, and processing settings.
How advertising estimates interests
An advertising system may estimate interests and select content, format, and timing from available signals. This is an estimate, not knowledge of a person’s desire or psychological state.
Ad selection may influence choice, but personalization alone does not prove manipulation, precise prediction before a search, or detection of maximum vulnerability. Such conclusions require a separate study.
How the data is used against you: dark patterns
Dark patterns are interface techniques that obstruct a free and understandable choice. Not every recommendation or autoplay feature is a dark pattern.
- Feeds and autoplay: check whether you can stop and control the stream.
- Rewards and streaks: assess participation terms; equivalence to a slot machine has not been established without research.
- Urgency: distinguish a real deadline from a false or resetting timer.
- Social proof: check the basis of stock or popularity claims.
- Choice architecture: assess how clearly price, alternatives, and refusal are presented.
Data may help tune an interface, but a technique does not prove a hidden profile of every visitor or an identical effect on everyone.
Examples of data-driven manipulation in everyday services
Assess the particular interface in each area:
- Social networks: recommendation rules and feed controls; amplification of conflict has not been established for every platform.
- Shopping and booking: the full price, deadlines, and terms. A price difference alone does not prove use of your fingerprint.
- Games: purchase confirmation and virtual currency terms; do not assume difficulty is adjusted to willingness to pay without evidence.
- Subscriptions: clarity of renewal, the next charge, and cancellation.
- News and video: recommendation controls; an inevitable shift toward extreme content is not established here.
Such types of problems are discussed in the FTC’s overview of dark patterns. These signs help raise questions about a service’s practices, but do not measure the effect on an individual’s attention or self-control.
For example, in March 2023, the FTC finalized an order requiring Epic Games to pay $245 million for consumer refunds in a case about unwanted Fortnite purchases. The FTC linked the problem, in particular, to confusing button design and charges without appropriate consent.
This is a limited historical case about interface design and payment. It does not prove collection of all the data described here, an effect on every player, or the prevalence of such practices in 2026.
What you can do to reduce tracking
Choose measures for your task; results depend on the browser and site.
- Use your browser’s supported tracking defenses; do not install an unknown extension for a promise of complete protection.
- Separate profiles help separate local data; a shared login may link activity again.
- Before an external login, read which data are requested; a separate login alone does not make you anonymous.
- Check whether site permissions are needed; revoking access does not delete information already sent.
- Assess WebRTC against the required route and browser policy. Do not disable it automatically: that may break calls and other features. Consult your configuration’s documentation; do not change mandatory protection on a work device without the administrator.
- For a connection routed through a VPN, the site sees the exit address. A VPN does not remove cookies, account sign-in, or browser characteristics, replace permission settings, or guarantee an end to tracking. Traffic excluded from the tunnel may take another route.
- Limit third-party scripts using browser controls, considering possible broken features and extension risks.
- Keep only necessary notifications; disabling notifications does not delete the account or previous records.
- If a service offers a chronological feed, assess it; it does not guarantee an end to data collection.
- Check the basis of urgency and popularity claims before paying.
General capabilities are described by MDN’s privacy guide, VPN limits by EFF, and WebRTC address handling by RFC8828.
An example from Chrome’s computer help, if you use this browser:
- Open the three-dot menu → Settings.
- Open Privacy and security → Site settings.
- Select the permission category you need and check which sites have access. Change only unnecessary access, considering your network and device requirements.
This path comes from documentation, not an interface test. Mobile versions and other browsers may differ. Checking one permission is not an audit of every tracker, and revoking it does not delete data already sent.
Start with unnecessary permissions, non-essential trackers, and understanding which account you are signed into. For the site you need, check purposes, recipients, and retention in its policy.
Combining these measures does not make you invisible. Without checking the implementation, we cannot promise how much collection or profile accuracy will decrease. Do not publish history, cookies, tokens, or raw logs for such a check.
Author: Adam Jensen — Cybersecurity specialist, VPNON