September 2026 edition · measured April 11, 2026 to September 5, 2026
At least 12.9% of the sites we scan load trackers without valid consent
Across 1068 sites submitted to our scanner, 918 of which returned a usable detailed report.
These sites are not a sample of European small businesses. They are sites someone came to test — so people already worried about their own compliance. No weighting would make this sample representative, and I would rather say so here than in a footnote.
The figure that holds: what the site does, not what it displays
Two failings are measured in the browser, against tracking cookies actually being set. They depend on no presentation convention, and they are the only ones I will build a headline on.
- 12.3% — no consent banner, while trackers are being set
- 5.3% — a banner is present, but trackers fire before the choice
Combined per site rather than added up — one site can carry both — that comes to 12.9%. Read it as a floor: the scanner does not accuse when the headless browser failed to run.
Missing documents, with the caveat that belongs to them
- 46.3% — no privacy policy page detected
- 40.5% — no cookie policy page detected
The scanner detects pages, not clauses. A site that informs its visitors inside its legal notice page is counted here as having no privacy policy page, whereas GDPR Article 13 requires the information to be provided, not to have a page of its own. That is a limitation of our tool, it penalises one country's presentation conventions more than another's, and it is why I publish no country comparison in this edition. It will be fixed before the next one.
What I do not know, and would rather write down
- Our detection of a site's country gets it wrong. It is inferred from the domain, the address, the language or the registry. Checked against the country domain where there is one, it is wrong 8.5% of the time — which is why I kept only the country domain.
- The channel these sites arrive through cannot be determined. The site builder is identified for only 6% of domains. A gap between two populations may therefore reflect two different channels rather than a difference in practice.
- The scanner changed mid-period, on July 19. Every rate published here was checked stable on both sides of that switch; those that were not have been removed rather than averaged.
- The score out of 100 appears nowhere. It is a tiered verdict, not a measurement: a site with no privacy page is capped at 45 whatever its other merits. The average of a capped score does not say what you think you are reading.
Method and data
The aggregated dataset is open, and it carries its method with it — a figure quoted without its method is a figure no one can defend.
Dataset, September 2026 edition (JSON)
If you cite it, the exact framing is: “across 1068 sites submitted to WebLegal.ai's scanner, between April 11, 2026 and September 5, 2026”. Next edition due December 2026, adding consent-platform market share.