Findings · 2026-11-15 · By VSNARY | Emmanuel Orta
Top 1,000 websites: every outside tracker we logged
We opened the homepages of the 1,000 most visited websites in a real browser and logged every outside domain that loaded, set a cookie, read canvas pixels or listened to the keyboard.
We opened the homepages of the 1,000 most visited websites (Tranco list Y83YG) in a real browser and recorded every outside domain that loaded, which ones set cookies, which read canvas pixels and which attached keyboard listeners. 685 homepages could be measured. Google Tag Manager loaded on 38.5% of them, DoubleClick set a cookie on 110 sites, and 1,450 different outside domains appeared in total.
What we measured #
One visit per site, to the homepage, on 15 November 2026 between 03:50 and 20:59 UTC, in a headless Chrome browser on our own measuring network. No consent banner was clicked, nothing was typed, and no account was signed in. For every script and request from a domain other than the site itself, we recorded four things:
| What we recorded | What it means | What it does not mean |
|---|---|---|
| Loaded | A request went to that domain while the page loaded | That it collected anything |
| Set a cookie | That domain set a cookie in the browser from outside the site | That the cookie identifies a person |
| Read canvas pixels | A script from that domain read pixels back from a canvas (toDataURL / getImageData) — a known fingerprinting method | That fingerprinting happened; charts and image editors read canvases too |
| Keyboard listener | A script from that domain registered a listener for key presses or text input | That it recorded what was typed. Consent banners listen for the Escape key; search boxes listen for typing |
The last column matters. "Listens for keys" is a capability, not proof of keylogging, and we report it as exactly that.
The sample, and the 306 sites we could not measure #
The frame is the top 1,000 registrable domains of one Tranco list, frozen on 2 October 2026 so every run measures the same sites. Nine were left out by our publish filter (adult and gambling). That leaves 991.
| Sites | |
|---|---|
| In the frame | 991 |
| Homepage measured | 685 |
| No result | 306 |
The list includes infrastructure domains — CDNs, API hosts, ad servers — that serve no homepage at all, and those count as unreachable. Others refused or timed out on a visit from a data centre. This run did not record a reason per site, so we do not split the 306 further. Treat every percentage below as a share of the 685, not of the web.
The outside domains seen most often #
| Domain | Company | Kind | On sites | Share | Set a cookie on | Keyboard listener on |
|---|---|---|---|---|---|---|
| googletagmanager.com | tag manager | 264 | 38.5% | 0 | 65 | |
| google.com | reCAPTCHA / maps | 225 | 32.8% | 39 | 23 | |
| doubleclick.net | ads | 174 | 25.4% | 110 | 0 | |
| gstatic.com | static files | 145 | 21.2% | 0 | 49 | |
| google-analytics.com | analytics | 122 | 17.8% | 0 | 8 | |
| googleapis.com | files / fonts | 116 | 16.9% | 0 | 18 | |
| onetrust.com | OneTrust | consent tool | 102 | 14.9% | 0 | 6 |
| cookielaw.org | OneTrust | consent tool | 100 | 14.6% | 0 | 81 |
| bing.com | Microsoft | ads | 83 | 12.1% | 66 | 0 |
| facebook.net | Meta | pixel | 74 | 10.8% | 0 | 0 |
| linkedin.com | pixel | 71 | 10.4% | 70 | 0 | |
| cloudflare.com | Cloudflare | CDN | 52 | 7.6% | 1 | 18 |
| clarity.ms | Microsoft Clarity | session recording | 52 | 7.6% | 34 | 48 |
| reddit.com | pixel | 46 | 6.7% | 40 | 0 | |
| tiktok.com | TikTok | pixel | 25 | 3.6% | 25 | 4 |
Six of the ten most common outside domains belong to Google.
Three things the table shows #
1. The cookie is usually set by a different domain from the one that loads the script. googletagmanager.com loaded on 264 sites and set a cookie on none; doubleclick.net set one on 110 of its 174. Reading only the domain in a site's script tags undercounts who drops cookies.
2. The consent tool listens to the keyboard more often than the trackers do. cookielaw.org, which serves OneTrust's banner, registered a keyboard listener on 81 of the 100 sites it loaded on — more than any ad domain. That is expected behaviour for an accessible banner (the Escape key closes it), and it is why a keyboard listener on its own is not evidence of anything.
3. Session recording is the category that pairs keyboard listeners with cookies. Microsoft Clarity loaded on 52 sites, set a cookie on 34 and listened to keyboard input on 48. Session-recording tools replay what visitors do on a page; they offer settings to mask typed text, and a homepage visit cannot see how a site has set them.
What this does not cover #
- One page per site: the homepage. Checkout, sign-in and article pages load different scripts.
- No consent was given. On sites that wait for consent, fewer trackers load in our visit than for a visitor who clicks "Accept".
- A visit from a data centre can be treated differently from a visit at home.
- 1,346 of the 1,450 outside domains are not yet assigned to a company in our map. They are counted, not named. Rows above are only those we could attribute.
Check your own site #
The same check runs on any site in a CrawlCheck report (section 51, "Tracking and privacy"). It lists every outside company your homepage loads, which set cookies before any consent choice, and which read canvas pixels or attach keyboard listeners: crawlcheck.io/
Corrected 15 November 2026: the first version of this page said 690 sites measured and 300 not. A recount from each site's own stored result gives 685 and 306, which add up to the 991 in the frame; the run's running total lost some updates and counted some sites twice. Every figure on this page is from the recount.
Published by CrawlCheck, whose reports include this tracking check as section 51.
CrawlCheck’s dataset when this page loaded: 570,959 sites measured.
Share this finding:LinkedInXEmailFollow Emmanuel on LinkedIn
Questions this post answers
Does a keyboard listener mean a site records what I type?
No. It means a script asked to be told about key presses. Consent banners, search boxes and forms do this for ordinary reasons. Recording is a separate step we did not observe.
Why only 685 of 1,000 sites?
306 returned no homepage to our browser: infrastructure domains with no homepage, and sites that refused or timed out. 9 were excluded as adult or gambling.
Which list did you use?
Tranco list Y83YG (top 1,000 registrable domains, created 1 October 2026, combining CrUX, Farsight, Majestic, Cloudflare Radar and Umbrella).
Did you accept cookie banners?
No. Every visit declined nothing and accepted nothing; what loads before any choice is what we measured.
Why the same 991 sites?
The frame is frozen so a later run measures the same sites and any change is a change on those sites, not a different sample.
Related findings
The main product
Found this on your own site? We fix it for $100.
Scan free to see where you stand. The fix is one site, every finding implemented and re-measured, with a sealed before and after.
Free browser extension
Check this on any site you open, without leaving the page.
CrawlCheck for Chrome grades every site you visit A–F and shows, from your own browser, what search engines and AI assistants are actually served. Free, no account.
Write for the library
Measured something yourself? Submit an article.
Send your finding, the data and the method. A person reads every submission and replies by email; accepted articles are published under your name with a link to your study. Free to submit. Articles are accepted on merit; no one can pay to have one published.
How it works · a reply within about a week
Spotted a mistake or a bug on this page? Report an error. Confirmed errors are fixed and logged on the corrections page.
Comments
Comments are read before they appear. Nothing is published automatically, and no account is needed. An AI agent with a published signing key can post without the queue: it signs the request and the signature is checked against its own domain (how). Those appear at once, marked verified.
Writing about this? Facts, live figures and marks — every number on that page is dated and traceable to a scan.