Research · Original data
Two-thirds of the UK's busiest websites track you before you consent
We took the 100 most-visited websites in the UK, loaded each one in a real browser, and recorded what happened before and after we answered the cookie banner. Almost all of them show a banner. Most of them have already started tracking by the time it appears.
Published
The key findings
Of the 100 sites, 22 served a bot-protection challenge instead of their real page, so we left them out rather than score a security screen. Across the remaining 78:
- 67% set analytics or advertising cookies before any consent was given (52 of 78). PECR Regulation 6 requires consent before such a cookie is stored, not after.
- 44% sent data to analytics or advertising services before the visitor answered the banner (34 of 78). Loading a tag manager alone does not count, and neither do Google Consent Mode pings sent with consent denied. These are real tracking requests.
- 57% still held analytics or advertising cookies after we clicked Reject (28 of the 49 sites where we could make a genuine refusal). The visitor said no; the tracking cookies stayed.
- Only 4% had no consent banner at all (3 of 78). The banner is almost universal. Whether it does anything is another matter.
- 59% failed automated WCAG 2.1 AA accessibility checks (46 of 78).
- The average compliance score was 48 out of 100. 56% of sites scored below 50; only 5% scored 80 or above.
Banner theatre at national scale
The pattern is consistent across retail, news, travel and tech: a consent banner from a well-known consent platform, styled properly, with clear Accept and Reject buttons, and behind it analytics and advertising cookies that were written the moment the page loaded. The banner asks a question the site has already answered on the visitor's behalf.
The most common offenders are not exotic ad-tech. They are standard analytics stacks: Adobe Analytics, Google Analytics, Tealium, Snowplow, Chartbeat, Parse.ly and Optimizely identifiers, set on first load. Tag rules that should wait for consent fire on page view instead.
Reject that doesn't
Where we could click a genuine refusal (Reject all, Decline, Essential only and their variants), 57% of sites still held analytics or advertising cookies afterwards. Counting every non-essential cookie, including preference and A/B-testing cookies, the figure is 92%. Most of these cookies were set before the banner was answered, and nothing removes them when the visitor refuses. Rejecting stops new tags firing on well-configured sites; it rarely cleans up what already ran.
The public-sector sites in the sample were the exception. GOV.UK and the NHS website set nothing non-essential before consent and nothing after Reject, which shows that a large, complex site can do this properly.
The other half of compliance nobody scans for
Well over half the sites failed WCAG AA automated checks. The failures are mostly the boring, fixable kind: images without alt text, links distinguishable only by colour, and icon buttons with no accessible name. The Equality Act 2010 applies to these sites as it does to a sole trader's, and none of these fixes needs a specialist.
Methodology, honestly stated
Sample. The 100 most-visited websites in the UK by traffic (SE Ranking's UK top 100, topped up from Ahrefs' UK ranking), compiled on 19 October 2026. We excluded adult sites, infrastructure and login endpoints, ranking anomalies with no plausible UK audience, and duplicates of a site already listed (bbc.com alongside bbc.co.uk, for example).
Scan. Each site was loaded once in a real Chromium browser from a UK connection using our UK compliance scanner. It records every cookie and network request before any interaction, finds the consent banner (visually, so shadow-DOM banners count), clicks a genuine refusal where one exists, and records cookies and requests again. AI was used only for the core detection steps: finding the banner on screen, reading button labels our rules don't recognise, and classifying cookies our rule files don't cover. No AI judgement went into any finding. Accessibility results come from axe-core against WCAG 2.1 AA.
What counts. "Analytics or advertising cookie" means a cookie matched by our published vendor rules, or classified as analytics or advertising where no rule exists. Preference, A/B-testing and unclassifiable cookies are left out of the headline figures. Cookies vendors document as security, fraud-prevention, load-balancing or session cookies are treated as essential.
Caveats. 22 sites blocked our scanner with bot protection and are not in the figures. Sites that block automated browsers may differ from those that don't. The reject figure has a smaller base (49) because it needs a banner with a refusal we could click. We publish percentages, not a league table of names. An automated scan is evidence, not a legal ruling, and one visit is a snapshot of one day.
Earlier version. Until 19 October this page reported figures from every scan run through our public tool, which mixed small business sites with test domains and our own. We replaced that self-selected sample with this fixed, reproducible one. While running it we found and fixed several classification errors in our scanner, among them a rule that treated any cookie with "id" in its name as tracking. All 78 sites were scanned or rescanned with the corrected rules.
If you run a UK website
Three checks cover most of what we found. Does anything load before the banner is answered? Open your site in a private window with the network tab open; analytics requests before your first click are the breach the ICO's cookie sweeps look for. Does reject actually work? Click it, then look at your cookie jar, including the cookies that were there before you clicked. And can a screen reader use your consent banner at all? The free scan answers all three in about two minutes, and the cookie banner checker tests the banner on its own. If the banner is the problem, the cookie consent tools comparison lists the alternatives on the same facts.
Journalists and researchers: you're welcome to cite these figures with a link to this page. For methodology questions or a breakdown we haven't published, get in touch.