Orange textured background

Personalization use case

Cookieless personalization

How to recognize a returning visitor and tailor their experience without third-party cookies: what you capture, what you store, and when consent is required.

Third-party cookies are going away and first-party storage is cleared aggressively. Device fingerprinting offers a probabilistic way to recognize returning visitors so you can resume state and remember preferences. Unlike fraud prevention, personalization needs explicit consent under GDPR, ePrivacy, and CCPA.

The job

A returning visitor who cleared their cookies, switched browsers, or browsed in private mode lands on your site and sees the default experience: generic content, an empty basket, a login prompt, a welcome tour they have already completed. From their perspective nothing was remembered. From your perspective you have no signal that this is someone who has been here before.

Device fingerprinting offers a way to close that gap without relying on third-party cookies or cross-site tracking. By capturing a stable device signal at each page load you can probabilistically recognize a returning visitor and restore context: the theme they prefer, the language they chose, the items they had in a cart, the step in an onboarding flow they reached. This page covers the full architecture for that recognition loop, and the consent obligations that come with it for non-fraud use cases.

What probabilistic recognition means

`getFingerprint` returns two hashes. The full `fingerprint` is per-browser: Chrome and Firefox on the same device produce different values because their rendering and audio pipelines differ. The `hardwareFingerprint` is derived from hardware-bound signals and is the same across browsers on one physical device; it also survives cookie clears and private browsing mode.

Neither hash is a guaranteed identifier. Two different devices can, in rare cases, produce the same hash. A single device can drift slightly after a major OS update or a hardware change. Recognition based on fingerprinting is probabilistic, not deterministic: it gives you a high-confidence signal that this is probably the same device you have seen before, not a cryptographic proof of identity. Design your personalization logic to degrade gracefully when no match is found, rather than treating every non-match as a new visitor that must be re-onboarded from scratch.

The end-to-end flow

StepWhat happens
1. ConsentGate collection behind your consent mechanism. If the visitor has not granted consent for personalization processing, do not call getFingerprint(). Store the consent decision in a first-party cookie or local storage so you do not re-prompt on every page load.
2. CaptureOnce consent is confirmed, call getFingerprint() client-side. You receive fingerprint, hardwareFingerprint, and consistency. No server is involved yet.
3. SendPOST the hashes to your backend over your existing API, alongside any session context (page, locale, referrer). Make personalization decisions server-side so you can audit and override them.
4. StorePersist a visitor record keyed by hardwareFingerprint: { hardwareFingerprint, fingerprint, preferences, resumeState, firstSeen, lastSeen }. You own this store; the library is client-only.
5. Look upOn each page load (with consent confirmed), re-capture and query for the incoming hardwareFingerprint. For tolerance to minor drift, run compareFingerprints in cross-browser mode to produce a matchScore.
6. PersonalizeOn a confident match, apply the stored context: render their preferred theme, pre-fill language, restore cart, skip completed onboarding steps. Log the match and update lastSeen.
7. Handle missesIf no match is found, serve the default experience. Do not treat a non-match as an error. Invite the visitor to set preferences explicitly so you have a first-party anchor alongside the device signal.

The recognition decision

Unlike fraud detection, the personalization decision does not need a binary allow/block outcome. A graded confidence level maps naturally to how aggressively you personalize: a high-confidence match can auto-apply saved preferences; a medium-confidence match can prompt the visitor to confirm their preferences before restoring state; a low-confidence or no-match result serves the default experience.

Use exact `hardwareFingerprint` equality as the fast path. When you have a stored prior record, pass both the incoming `fingerprint` and the stored `fingerprint` to `compareFingerprints` in `cross-browser` mode. The resulting `hardwareMatch` field tells you whether the hardware-bound signals align, which handles the case where the visitor returned on a different browser.

Personalization confidence tiers

ConditionSuggested approach
Exact hardwareFingerprint matchHigh confidence: auto-apply stored preferences and resume state silently.
No exact match, but hardwareMatch: true from cross-browser comparisonGood confidence: apply preferences, optionally surface a 'Welcome back' prompt so the visitor can confirm.
No match foundNo recognition: serve default experience. Invite the visitor to set preferences to build a first-party anchor.
High spoofLikelihood on incoming deviceTampered browser or randomized signals: do not personalize from a stored record. Serve the default experience.
npm install doorman-benny
import { getFingerprint } from 'doorman-benny';

// STEP 1-3: check consent, capture, send to backend.
async function initPersonalization(): Promise<void> {
  // Only proceed if the visitor has consented to personalization.
  const hasConsent = localStorage.getItem('personalization_consent') === 'granted';
  if (!hasConsent) {
    return; // Render default experience.
  }

  const result = await getFingerprint();

  await fetch('/api/visitor/recognize', {
    method: 'POST',
    headers: { 'Content-Type': 'application/json' },
    body: JSON.stringify({
      hardwareFingerprint: result.hardwareFingerprint,
      fingerprint: result.fingerprint,
      spoofLikelihood: result.consistency.spoofLikelihood,
    }),
  });
}

Client side. Always gate on consent first. Send both hashes to the server; make the personalization decision there.

import { compareFingerprints } from 'doorman-benny/server';

// STEP 4-6: store, look up, and decide server-side.
// `db` is your own datastore, indexed by hardwareFingerprint.
export interface VisitorRecord {
  hardwareFingerprint: string;
  fingerprint: string;
  preferences: Record<string, unknown>;
  resumeState: Record<string, unknown>;
  firstSeen: Date;
  lastSeen: Date;
}

async function recognizeVisitor(payload: {
  hardwareFingerprint: string;
  fingerprint: string;
  spoofLikelihood: 'low' | 'medium' | 'high';
}): Promise<{ confidence: 'high' | 'medium' | 'none'; record?: VisitorRecord }> {
  // Randomized signals cannot be trusted for personalization.
  if (payload.spoofLikelihood === 'high') {
    return { confidence: 'none' };
  }

  // Fast path: exact hardware match.
  const exact = await db.visitors.findByHardwareId(payload.hardwareFingerprint);
  if (exact) {
    await db.visitors.updateLastSeen(payload.hardwareFingerprint);
    return { confidence: 'high', record: exact };
  }

  // Cross-browser path: check recent records for a hardware-level match.
  const candidates = await db.visitors.findRecentCandidates({ limit: 20 });
  for (const candidate of candidates) {
    const comparison = compareFingerprints(
      payload.fingerprint,
      candidate.fingerprint,
      { mode: 'cross-browser' }
    );
    if (comparison.hardwareMatch) {
      await db.visitors.updateLastSeen(candidate.hardwareFingerprint);
      return { confidence: 'medium', record: candidate };
    }
  }

  // No match: create a new visitor record for future recognition.
  const newRecord: VisitorRecord = {
    hardwareFingerprint: payload.hardwareFingerprint,
    fingerprint: payload.fingerprint,
    preferences: {},
    resumeState: {},
    firstSeen: new Date(),
    lastSeen: new Date(),
  };
  await db.visitors.insert(newRecord);
  return { confidence: 'none' };
}

Server side. Exact match is the fast path; cross-browser comparison handles returning visitors who switched browsers. Store a record on every new visitor so future visits have something to match against.

What to personalize and what to anchor

Device fingerprinting works best as a complement to explicit first-party signals, not a replacement for them. Use it to bridge the gap between sessions when you have no login or cookie to rely on. Once the visitor logs in, switch to the authenticated identity as your primary key and use the device fingerprint as a secondary signal for cross-device consistency checks.

Good candidates for fingerprint-bridged personalization: UI theme and font-size preference, language and locale, cookie-consent decisions themselves (so the visitor is not re-prompted on every device), onboarding completion state, and in-progress multi-step flows. Avoid using the device signal alone to drive high-stakes personalization decisions (pricing, entitlements, or medical content routing) without a stronger identity anchor.

Handling false positives and drift

A device fingerprint can shift legitimately: a major OS update, a new external monitor, a hardware swap, or a browser major version. Use the graded `matchScore` from `cross-browser` mode rather than raw equality when you have a prior record, so minor drift does not result in a false non-match. When the score is moderate rather than high, surface a light confirmation prompt ('Are you the same person who set these preferences?') rather than silently applying the stored context.

A false positive (two different visitors matching to the same record) is low-probability but possible. It results in one visitor briefly seeing another's stored preferences - a nuisance, not a security breach. Mitigate it by capping the stored context to low-sensitivity items and building an explicit 'reset my preferences' path in your settings.

How this differs from the fraud use cases

Every other use case on this site is an anti-fraud job: free-trial abuse, account sharing, bot detection, multi-accounting, referral fraud, paywall enforcement. Those jobs typically qualify for the fraud-prevention and security exemption in GDPR and ePrivacy, which means they do not require consent in the same way a tracking or personalization use case does.

Cookieless personalization does not share that exemption. The processing purpose is commercial convenience for your business, not protection against attack. This distinction matters for your privacy notice, your consent implementation, and your legitimate-interest assessment if you rely on that basis instead of consent. See /laws/gdpr-fingerprinting for a breakdown of which bases apply to which use cases.

Also note: the visitor-facing disclosure and control requirements under CCPA are separate from GDPR. If you serve California residents, review your opt-out mechanism against the CCPA definition of 'sale' and 'sharing' of personal information, which may cover a device identifier sent to a personalization service. See /glossary/visitor-identification for the CCPA framing.

Frequently asked questions

Can you personalize a website without cookies?

Yes, but not with certainty. A device fingerprint is a probabilistic identifier: it gives you a high-confidence signal that this is probably the same device you have seen before, not a cryptographic proof of identity. For most personalization tasks (restoring preferences, resuming state, skipping a completed onboarding flow) this level of confidence is sufficient. For higher-stakes decisions, pair the device signal with an explicit first-party identifier such as an account login.

Is cookieless personalization GDPR compliant?

It can be, but it is not exempt from GDPR the way fraud prevention is. Personalization is a commercial purpose, so you need either freely given consent or a documented legitimate interest that is not overridden by the visitor's interests. You must disclose the use of fingerprinting in your privacy notice and give visitors a way to object. The /laws/gdpr-fingerprinting page covers the legal basis options in detail. This is engineering documentation, not legal advice.

Do I need a consent banner for fingerprint-based personalization?

In most cases, yes. The ePrivacy Directive and its national implementations require consent for accessing or storing information on a user's device for purposes other than strict necessity. Personalization is not strictly necessary. If you use a legitimate-interest basis instead, you must still inform users and offer an opt-out. Skipping the banner because you are 'not using cookies' is not a compliant approach. See /laws/consent-banners-fingerprinting for what the disclosure must cover.

How accurate is fingerprint-based visitor recognition?

Accuracy varies by device type, browser, and the stability of the visitor's hardware configuration. The hardwareFingerprint is stable across browser switches and private-browsing mode on the same physical device, and it survives cookie clears. It can shift after a major OS update or a hardware change. Treat it as a probabilistic signal and design your personalization logic to degrade gracefully on a non-match rather than treating every non-match as a first-time visitor.

What is the difference between fingerprint-based personalization and fingerprint-based fraud prevention?

The processing purpose. Fraud prevention (blocking trial abuse, detecting bot traffic, preventing account takeover) qualifies for a security and fraud-prevention exemption in GDPR and ePrivacy, which relaxes the consent requirement. Personalization is a commercial convenience purpose and does not share that exemption. The same technical signal (the hardwareFingerprint) is used in both cases; what changes is the legal basis you rely on and the disclosures you owe the visitor.

Get started

Build cookieless visitor recognition

npm install doorman-benny. Hardware-bound device identity and cross-browser matching in one package. No third-party cookies, no server round-trip for the fingerprint. Pair with your consent layer and first-party auth for a durable recognition stack.

Get startedRead the overview

Related posts

Last reviewed July 12, 2026