PrivacyProbe

PrivacyProbe Research · Edition 01

State of Website Privacy 2026

An independent field measurement of cookie behaviour across 1,000 measurable, high-traffic web origins.

Measurement window: 6–12 November 2026
Source cohort: September 2026 Chrome UX Report
Anonymous · deterministic · no AI classification

PRIVACYPROBE

Executive summary

A clean initial observation was the exception, not the norm.

PrivacyProbe observed each homepage in an isolated browser without interacting with a consent prompt. Of 1,000 measurable origins, 58.3% triggered a fail verdict under the research rules, 31.1% required review, and 10.6% produced a clean classified observation.

Interpretation: these verdicts are technical early-warning indicators, not legal findings. A fail identifies known issue cookies or high-risk unclassified cookie names during the observation. A pass does not establish compliance.

58.3%

Flagged for potential privacy risk

583 of 1,000 sites.

10.6%

Clean classified observation

Only 106 sites returned PASS.

92.2%

Set at least one cookie

Before any consent interaction.

49.1%

Analytics or marketing detected

Recognised cookie categories across 491 sites.

83.2%

Contained unclassified cookies

A material ownership and governance gap.

65.9%

Set a cookie lasting over one year

659 sites had at least one.

Verdict distribution

Share of the 1,000-site cohort

58.3%31.1%10.6%
Fail · 583Review · 311Pass · 106

01 · Core findings

Five signals for privacy and digital leaders

Known issues appeared on almost half of sites.

489 sites contained at least one cookie classified as an issue. A further 94 failed solely because unclassified names matched high-risk analytics, advertising or tracking patterns.

Unknown inventory is the dominant governance problem.

5,953 of 8,220 observed cookies—72.4%—were unclassified. Unknown does not mean harmful, but it does mean purpose, ownership and consent requirements cannot be confidently established from the inventory.

Analytics and marketing are embedded in the initial experience.

491 sites showed a recognised analytics or marketing cookie, and 140 showed both categories. The observation occurred without accepting a consent prompt.

Cookie lifetimes are routinely long.

88.2% of observed cookies were persistent rather than session-only. At least one cookie lasting longer than a year appeared on 659 sites.

Visibility remains incomplete even at scale.

The original CrUX top-1,000 bucket yielded 840 measurable sites. The final research cohort required four recovered origins and 156 source-ordered replacements from the next popularity bucket.

02 · Cookie load

The median site set six cookies; the long tail reached 55.

Across the cohort, PrivacyProbe observed 8,220 cookies. Most sites sat in the low-to-middle single digits, while 103 sites set 20 or more cookies during the short initial observation.

0Minimum
325th percentile
6Median
1075th percentile
2090th percentile
55Maximum

Sites by observed cookie count

Number of sites in each band

No cookies78
1–4338
5–9295
10–19186
20–49101
50+2

03 · Classification

The web has an inventory problem before it has a verdict problem.

Unknown cookies represented nearly three quarters of the observed inventory. This does not establish misuse; it shows that common automated inventories cannot readily connect most cookie names to a reliable owner and purpose.

Cookie classification across 8,220 observations

Unclassified72.4%
Issue21.0%
OK5.4%
Warning1.3%

What leaders should ask

  • Can every cookie be mapped to an owner and business purpose?
  • Is the consent category documented?
  • Is the expiry proportionate to that purpose?
  • Does the deployment inventory match the CMP configuration?

What “unknown” does not mean

It is not automatically a violation, tracker or security defect. It is an evidence gap. At this prevalence, that gap becomes an operational risk because teams cannot consistently explain what their sites place in a browser.

04 · Technology signals

A small group of platforms accounts for a large recognised footprint.

Vendor references below are based on recognised cookie signatures, not script inspection or proof of a vendor relationship. Counts show the number of sites where at least one matching cookie was observed.

Recognised signalSitesCohort
Google Analytics40040.0%
Google Ads13513.5%
Cloudflare bot protection10310.3%
Meta Pixel838.3%
OneTrust consent cookie505.0%
Yandex Metrica464.6%
Microsoft Clarity313.1%
Microsoft Ads282.8%
Adobe ECID272.7%
TikTok Pixel272.7%
Cookiebot consent cookie171.7%
A recognised consent cookie is not evidence that a banner worked correctly; equally, its absence is not evidence that no consent system exists. This study observed cookie outcomes, not banner configuration.

05 · Persistence & controls

Persistent identifiers dominate the observed inventory.

Of 8,220 cookies, 7,247 had an expiry and 973 were session-only. More than 5,000 were set to persist beyond 30 days.

88.2%

Persistent cookies

7,247 of 8,220 observations.

61.0%

Persisted beyond 30 days

5,016 observed cookies.

30.3%

Persisted beyond one year

2,494 observed cookies.

Attribute observations

62.3% of cookies lacked the Secure attribute, 63.1% had no explicit SameSite value, and 86.0% were not HttpOnly. These attributes serve different technical purposes; absence is not automatically a privacy or security failure, particularly for cookies that must be read by client-side code.

Site-level exposure

801 sites had at least one cookie without Secure. 819 had at least one cookie with no explicit SameSite value. These figures indicate a large review surface, not 801 or 819 confirmed defects.

Review priority

Start with long-lived analytics and marketing identifiers, then assess whether purpose, consent state, retention and attributes are aligned. Attribute checks are most useful when interpreted alongside purpose.

06 · Industry implications

Privacy assurance needs to become an operating discipline.

Privacy & legal teams

Move from periodic policy review to evidence-backed inventories. The unknown-cookie rate shows why questionnaires alone cannot maintain an accurate record of live behaviour.

Marketing & ad operations

Recognised analytics or marketing cookies appeared on 49.1% of sites before any consent interaction. Validate tag firing and consent state after every campaign or container change.

Engineering & security

Treat cookie purpose, owner, expiry and attributes as deployable configuration. Add external regression checks to releases rather than relying only on CMP settings.

Agencies & CMP providers

Demonstrate outcomes, not installation. A banner or consent cookie is not proof that optional activity is gated as intended.

Boards & risk owners

The clearest management metric is explainability: what proportion of live cookies can the organisation identify, justify and assign to an accountable owner?

Researchers & regulators

Automated measurement itself has a visibility boundary. Accessibility filtering should be disclosed because blocked sites may differ systematically from measurable sites.

07 · Action agenda

Five practical moves for the next 90 days

01

Resolve the unknown inventory.

Assign every recurring cookie a purpose, owner, consent category and retention period. Escalate anything that remains unexplained.

02

Test before interaction.

Observe the browser before any banner choice. This is where default tag-manager and plugin behaviour is most likely to create unintended exposure.

03

Reduce persistence by design.

Challenge multi-year expiry defaults and retain identifiers only as long as the documented purpose requires.

04

Make privacy regression part of delivery.

Re-test after releases, CMP updates, tag changes and campaign launches. Compare evidence over time rather than treating a scan as a certificate.

05

Report evidence, not reassurance.

Give risk owners a small set of operational measures: known inventory rate, optional activity before choice, unresolved changes and time to remediation.

Methodology

How this research was produced

Source
September 2026 global Chrome UX Report list, frozen by checksum. CrUX provides popularity buckets, not an exact ordinal ranking within the top-1,000 bucket.
Cohort
840 measurable origins from the top-1,000 bucket, four recovered top-bucket origins, and the earliest 156 successful replacements from the next popularity bucket. The final cohort is therefore accessibility-filtered.
Observation
One anonymous, unauthenticated homepage load in a fresh isolated browser context. No consent choice, login or page interaction was performed.
Timing
Measurements ran from 6 to 12 November 2026. 840 observations waited 2.5 seconds after load; 160 recovery and replacement observations waited 5 seconds.
HTTP outcome
993 observations returned status 200, six returned 202 and one returned 203.
Classification
Deterministic PrivacyProbe rules; AI was disabled. FAIL means a known issue cookie or high-risk unclassified name was observed. REVIEW means unclassified cookies remained without a fail trigger. PASS means no issue or unclassified cookie was observed.
Anonymity
No site names, domains, URLs or cookie values are published. Vendor references are aggregate cookie-signature observations.
Evidence retained
Cookie names, attributes, classifications, navigation status and timings. The compact research dataset does not retain raw script, request, localStorage or sessionStorage inventories.

Limitations: This is a single initial-load observation, not a multi-state consent audit or legal assessment. Sites may vary by location, device, personalisation, time and later interaction. A zero-cookie observation does not rule out non-cookie tracking. Domain relationship estimates are not used as headline findings. Percentages may not sum to 100 due to rounding.

PrivacyProbe Research

Privacy maturity starts with being able to explain what the browser actually does.

This research points to a practical industry priority: build inventories that remain accurate after deployment, verify consent outcomes independently, and treat unknown behaviour as work to resolve—not evidence to ignore.