Skip to content

Research

Methodology

How we scanned 97,304 EU websites - sample source, scanner architecture, infrastructure, and limitations.

Scan Dataset

  • Scan window: April 7–11, 2026
  • URLs submitted: 114,748
  • Successful scans: 97,304 (84.8%)
  • Failed scans: 17,444 (site unreachable, TLS errors, timeouts)
  • EU countries covered: 25 of the 27 EU member states (all except Cyprus and Malta)

Sample Source

  • Domain list: Tranco Top 1M (research-grade domain ranking, snapshot L7684)
  • Selection method: Filtered by 25 EU country-code TLDs, round-robin by country to prevent large-market bias
  • Country distribution: Germany 22,696 / France 8,463 / Netherlands 8,752 / Italy 8,598 / Poland 7,881 / Spain 4,735 + 19 other EU countries (full breakdown in the report)

Definitions Used in Published Figures

  • Tracking before consent: at least one request to a known third-party tracking service before any consent interaction. Tag managers (such as Google Tag Manager) and consent-platform loader scripts are not counted: 62.6% of sites. Counting them as well gives 68.0%.
  • Cookies before consent: analytics or marketing cookies set before any consent interaction: 36.7% of sites (31.1% analytics, 18.3% marketing). At least one cookie of any kind, including strictly necessary ones: 66.6%.
  • Reject test: run on the 28,891 sites with a clickable first-layer Reject button. 6,446 tests (22.3%) could not be verified (for example, the reload after Reject failed) and are reported as inconclusive. Of the 22,445 conclusive tests, 48.3% still had non-essential cookies after Reject and a reload, and 74.8% showed some tracking signal (non-essential cookies or continued requests to tracking services). On 2,341 sites the only signal was a request to Google Analytics or Google Ads, which can be a Consent Mode "cookieless ping".
  • High risk: 40.7% of sites (Medium 16.8%, Low 27.6%, Inconclusive 14.9%).
  • Per-country research pages count a site as tracking before consent when the scanner raised a pre-consent tracker, analytics cookie, marketing cookie or fingerprinting finding. That is a slightly different measure from the request-based 62.6% above. Their "Reject not honoured" share uses the Reject test definition above: conclusive tests only, counting the sites that still had non-essential cookies after Reject and a reload. Because each website is counted once there (its latest scan), the pooled figure across the 25 countries is 48.4% (10,813 of 22,353 conclusive tests) rather than 48.3%.
Corrected on 2026-10-21: scan window, country coverage and the definitions above were updated after re-measuring the published figures from the stored scan results. On the same day the per-country "Reject not honoured" share was changed to count conclusive tests only; it previously also counted tests that could not be verified and sites where only tracking requests continued.

Scanner

  • Engine: Go-based scanner using headless Chromium via Chrome DevTools Protocol (CDP)
  • Browser: Chromium (headless, sandboxed, clean profile per scan)
  • Viewport: Desktop 1920×1080
  • Scan location: Hetzner Cloud, Helsinki, Finland (EU)
  • CMP detection: 45 consent management platforms recognized via script signatures, DOM selectors, and JavaScript API probing
  • Selector version: 2026-04-06-v23

Infrastructure

  • App server: Hetzner CX-class VPS (web application, PostgreSQL, Redis)
  • Worker servers: 2 Hetzner VPS instances (15 concurrent scan slots total)
  • DNS cache: Local Unbound resolver with periodic pre-warming
  • Total runtime: ~60 hours for the full corpus

What Each Scan Measures

  1. Pre-consent snapshot: All cookies, tracking services, and third-party connections set before any user interaction
  2. Banner detection: Consent management platform identification, accept/reject button presence
  3. Accept flow: Click Accept, record post-consent cookies and trackers
  4. Reject flow: Fresh browser, click Reject, reload page, verify tracking stops
  5. Risk scoring: 0–100 composite score based on 27 distinct compliance findings

Limitations

  • Desktop only. Mobile viewport behavior may differ.
  • Single scan location. Geo-targeted consent banners may behave differently from non-EU locations.
  • Point-in-time observation. Websites change after scanning. Results represent the state during the scan window.
  • Bot detection. Some sites detect automated browsers and may alter behavior (we detect and report this when identified).
  • CMP coverage. 45 CMPs recognized; sites using unrecognized or custom consent solutions fall back to generic heuristic detection.
  • No legal interpretation. Risk scores are technical indicators, not legal compliance determinations.

Data Access

  • Aggregate statistics are published in our blog posts and research report.
  • Country reports: per-country aggregates (with their own sample, scan period and thresholds stated on each page) are listed below, with a downloadable CSV under the CC BY 4.0 licence.
  • Individual domain classifications are not published. Site operators can request their scan result by running a free scan or contacting us.
  • Correction requests: If you believe your site's classification is incorrect, run a new scan or contact [email protected].

Reproducibility

The scanner is proprietary software. The aggregate dataset is provided for verification of published statistics. The Tranco domain list used is publicly available at tranco-list.eu.

Country reports

Per-country results from our scan sample: tracking before consent, missing banners, ignored Reject clicks, trackers and consent platforms.


Read the Findings

See what we found across 97,304 EU websites.

Read the full reportSee sample report →