A device ID for web apps is what you fall back on when the cookie is gone: cleared by the user, blocked by the browser, or never written because the visit started in a private window. Cookies are storage the visitor controls, so any identity that depends on them ends when the jar is emptied. A signal-based device identifier works differently: it is a value computed at request time and matched against earlier visits, not a file the browser keeps for you.
That difference decides what your rules can survive. A stable identifier keeps a paywall meter, trial limits and abuse caps working across cookie clearing and incognito. Applied carelessly, the same stability becomes tracking without notice or a false accusation against a shared laptop. This post covers how the identifier holds, how to measure what you actually get, and where to stop.
What a device ID for web is, and what it is not
A device ID is derived from properties the browser exposes at request time: rendering behavior, platform and GPU details, installed fonts, screen geometry, timezone, language, and network path. Those signals are hashed into an identifier and compared with identities seen before; when enough independent signals line up, your server gets a visitorId back. It is not a cookie and not browser storage, so there is nothing on the client to clear. It is also not a user ID: it describes a device-and-browser pair, not a person. Our device fingerprinting overview walks through the signal set.
Three identifiers, three jobs
Keep them distinct in your schema, or analytics and risk rules argue with each other:
visitorId— device and browser. Survives cookie clearing; changes when the browser or hardware identity genuinely changes.userId— the person, set at login or signup. The only identifier to trust for authorization.accountId— the tenant. The scope for team limits such as seats or shared quota.
Why a visitorId survives cookie clearing and incognito
Private browsing exists to delete local state when the window closes, and it does that well. It does not change the GPU, the fonts installed on the machine, the timezone, or how the operating system renders a canvas, so a signal-based identifier still has something to match on the next visit from the same machine.
What tends to hold: platform and OS family, GPU and WebGL renderer strings, canvas and audio rendering quirks, font set, screen geometry, pixel ratio, timezone and language.
What moves: the user agent after a browser update, sometimes much of the signal set after a major release; the IP address after a mobile handover or VPN switch; window size and extensions that inject scripts.
Because of that drift, an honest implementation returns a match with evidence rather than a certainty. Two visits either agree across enough independent signals or they do not, and you should be able to read which signals agreed and which were missing. Tampering signals matter for the same reason: when a client feeds false values on purpose, the identifier should degrade visibly instead of confidently returning a wrong match.
Where a stable device ID should not be used
Stability is not permission. Write these boundaries into policy:
- Not as proof of identity. For login and recovery, a device is a reason to step up — captcha, email verification, passkey — never a credential. A shared family laptop should not open someone else’s account.
- Not as a permanent ban. Device identities drift and shared devices exist. Prefer a scored step-up with an appeal path over a lifetime block.
- Not across sites, and not for advertising. The identifier belongs to your own first-party context; following people around the web with it is a different product on a different legal footing.
- Not to deanonymize visitors. In Prynt Reveal, a company is named only when a visit comes from that company’s own network, and people are linked only from first-party data with a consent record.
A device identifier can be personal data under regimes such as the GDPR, and notice and consent requirements vary by jurisdiction and by use: a question for your counsel. Practically, publish the notice, honor opt-outs, set a retention limit you can defend, and scope the identifier to identification and abuse decisions.
Measure first, then turn the ID into a decision
Measure what your traffic gives you
Log visitorId beside a first-party userId you control, then compute three things: match rate on known-returning devices, false merges (two people on one visitorId — corporate laptop images, VMs, kiosks), and churn clustered around browser release dates. Two weeks of that beats any comparison page; our accuracy page explains how we report match quality.
Decide with reasons, not a blacklist
Map the identifier plus request signals to a decision — allow, step-up, block — and return a reason code your support team can explain. Typical uses: one trial per device, OTP caps per device and network, and a paywall meter keyed on device and account so incognito does not reset the count. Prynt Control answers those decisions and never blocks on its own: your code decides, and a 14-day mirror period returns decisions while nothing is enforced. When device risk feeds login, the account takeover guide shows the step-up patterns.
FAQ
Does a visitorId survive a browser update?
Sometimes. A major release can shift enough signals to look like a new device, so treat a mismatch as a reason to step up rather than a reason to block an account.
Is a device ID the same as a user ID?
No. A device ID describes a browser on a machine and is not tied to a person. A user ID is created at login and is the only identifier strong enough for authorization; the device ID only informs how much you trust that session.
Does incognito reset a paywall meter built on device ID?
Not if the meter counts device as well as cookie. Prynt Control meters per device and per account, so a private window does not hand out a fresh set of free reads.
Create a free Prynt Identify account at https://pryntid.com, log visitorId for a week, and decide what to enforce with data instead of assumption. Integration steps are in the docs.
Try it free
Prynt is device intelligence with a free plan: 5,000 identifications a month, visitor IDs with confidence and bot detection — no credit card. Upgrade to Pro for VPN, proxy, tampering and behavioral signals.