Ask how much of a web page is advertising in Germany versus the United States and you will get confident answers from people quoting nothing. The honest position is that the number has never been published. There is no dataset anywhere giving ad bytes as a share of total page bytes, broken down by country.
What does exist is worth laying out properly, because three separate measurements circle the question from different sides, and because Chrome shipped the missing instrument five days ago.
Key takeaways
- No published dataset gives ad bytes as a share of page bytes by country. Treat any such figure you are shown as invented until someone names a crawl.
- Median mobile page weight ranges from 3,816 KB in South Korea to 2,024 KB in Nigeria, a 1.9 times spread, but this measures the sites a country’s users visit, not what a page sends to that country.
- Chrome added Ad Count, Ad Density, Ad Weight Network and Ad Weight CPU to CrUX on 15 September 2026. Country splits in CrUX are BigQuery-only, and these metrics are not in BigQuery yet.
- The best available country comparison is advertising and tracking domains per site by visitor location: 11.7 per site for a Californian against 4.2 for a German, on identical sites.
- Much of that gap is consent state rather than a permanent market difference, and saying so is part of reporting it honestly.
What we can measure today: total page weight by market
HTTP Archive publishes median page weight by country every month, and the spread is large. These are medians from the July 2026 crawl, pulled from the public reporting API.
| Market | Mobile median | Desktop median | Mobile JavaScript |
|---|---|---|---|
| South Korea | 3,816 KB | 4,401 KB | 822 KB |
| Japan | 2,970 KB | 3,329 KB | 645 KB |
| United States | 2,899 KB | 3,164 KB | 1,080 KB |
| United Kingdom | 2,726 KB | 3,070 KB | 961 KB |
| All countries | 2,507 KB | 2,958 KB | 719 KB |
| France | 2,447 KB | 2,731 KB | 721 KB |
| India | 2,367 KB | 2,777 KB | 673 KB |
| Brazil | 2,242 KB | 2,633 KB | 884 KB |
| Germany | 2,230 KB | 2,596 KB | 530 KB |
| Nigeria | 2,024 KB | 2,500 KB | 692 KB |
JavaScript is where markets separate most sharply. A median United States page ships roughly twice the JavaScript of a median German one, and JavaScript is where advertising lives.
One caveat decides whether this table means anything to you. HTTP Archive’s geo dimension identifies sites popular with users in that country. The crawl itself runs from Google Cloud locations in the United States, as the Web Almanac methodology states plainly. So the correct reading is “the web that Korean users typically load is heavier”, not “the same page sends more bytes to Korea”. Those are different claims and only the first is supported.
The table also has no advertising category. Total, JavaScript and images are published. Ads are not.
Chrome just built the missing instrument
On 15 September 2026 the Chrome team added four experimental ad metrics to the Chrome User Experience Report:
| Metric | Definition |
|---|---|
| Ad Count | The average number of ads in the viewport |
| Ad Density | The average fraction of the viewport area occupied by ads |
| Ad Weight: Network | The resources consumed by ads measured in bytes |
| Ad Weight: CPU | The resources consumed by ads measured in milliseconds |
The methodology is specific. Chrome samples the viewport once per second and takes a snapshot, and an ad counts as visible if any of its pixels are inside the viewport at that sample point. Overlapping ads are unioned for density but counted individually for Ad Count. Chrome reports the 75th percentile. Only origins listing at least one authorised seller in their ads.txt are eligible, so sites with no file, or a placeholder record, are excluded.
This is field data from real Chrome users, which is exactly what the question needs. But the country dimension in CrUX is BigQuery-only, and Google’s documentation states that country dimensions are not available in the CrUX APIs. The ad metrics are not in BigQuery yet. So the instrument that will finally answer “how much of a page is advertising, country by country” exists, is five days old, and cannot be split by country today. When those metrics reach the BigQuery country tables, the answer becomes a query.
The closest available country comparison
Until then, the defensible proxy is counting advertising and tracking domains, and here the measurements are genuinely good because they load the same sites from different countries.
A crawl running from ten countries between 5 and 24 January 2025 visited an identical set of globally popular sites from each. Average tracker connections per site came to 11.7 from California and 11.2 from Australia, against 5.3 from Spain and 4.2 from Germany. The authors report EU vantage points seeing 50.5% fewer tracker connections. Advertising dominates the mix: 55,079 of 84,170 observed connections, or 65.4%, were advertising, roughly 2.8 times the analytics volume.
An April 2023 crawl of 2,004 sites measured the same thing across five cities:
| Vantage point | Tracker domains per site | Third-party domains per site |
|---|---|---|
| New York City | 23.4 | 31.6 |
| San Francisco | 21.3 | 29.3 |
| Frankfurt | 15.6 | 23.2 |
| London | 14.3 | 21.3 |
| Amsterdam | 14.3 | 21.4 |
Same sites, same week. A visitor in New York met 63% more tracker domains per site than one in Amsterdam. The mobile crawls moved in parallel, 21.8 in New York against 15.2 in Frankfurt.
A third study, presented at MADWeb 2026, crawled 743 popular websites from eight vantage points with browsers and consent states varied systematically. It found GDPR-regulated locations exhibit about 80% fewer third-party advertising and tracking domains before consent, and keep 89% to 91% of those requests within the EEA or adequacy countries.
Three independent designs, one consistent direction.
The honest caveat
A large part of that EU gap is consent state, not a permanent property of the market. The same ten-country study measured it: on a sample of sites, simply not interacting with the cookie banner rather than accepting reduced tracker connections by 48.5% in Germany but only 21.1% in California. A crawler that never clicks anything is measuring the pre-consent web, and in Europe the pre-consent web is deliberately quiet.
That is not a reason to discard the numbers. It is a reason to publish the consent state alongside them, and it explains why a European page can look lightweight in a crawl and feel heavy to a reader who clicked accept. The related patterns in consent interfaces themselves are covered in consent walls by country.
Measuring your own markets
If you need ad load for specific markets rather than the global picture, this is measurable today with ordinary tooling. The design that survives scrutiny looks like this.
- Fix the site list before you start. One list, run identically from every market. A locally popular list per country answers a different question and the two get confused constantly.
- Collect from the market itself. Not a VPN endpoint in a datacentre that a publisher’s ad stack treats differently from a consumer connection.
- Record the consent state. Pre-consent and post-consent are separate observations of the same page. Never average them.
- Classify by request, not by eye. Count requests and bytes against a maintained ad and tracker domain list, and keep the list version in your results.
- Repeat. A single load is an anecdote. Header bidding makes consecutive loads of the same page differ.
- Record the vantage point next to every row, for the reasons argued in the case for a vantage-point standard.
Shifter’s residential gateway takes the target country in the username, so the same harness runs the same list from several markets without any other change:
curl -x customer-USERNAME-country-de-sid-9f3a2b7c-ttl-600:PASSWORD@p.shifter.io:443 \
https://example.com/Country codes are ISO 3166-1 alpha-2. A session id holds the same exit IP across a page load and its subresources, which matters because an ad stack that sees each request from a different address behaves differently from one serving a real visitor. Details are in the geo-targeting and sessions documentation.
The same measurement discipline underpins ad verification work, and the creative side of it is covered in catching cloaked and geo-targeted ads and competitive campaign intelligence.
The bottom line
Two things are true at once. Ad load clearly varies by market: three independent crawls agree that a visitor in New York receives substantially more advertising and tracking infrastructure than one in Amsterdam or Berlin, on the same websites in the same week. And nobody has yet published the number everyone quotes, the share of a page that is advertising, country by country.
Chrome has now built the measurement. It reaches the country tables eventually, and on that day the question stops being a matter of opinion. Until then, the defensible move is to measure your own markets with a fixed list, a real vantage point and a recorded consent state, and to say which of those you did.
Sources and references
- HTTP Archive, Tech Report page weight by geography. Median page weight, JavaScript and images by country, July 2026 crawl, retrieved via the public reporting API at
cdn.httparchive.org/v1/page-weight. - HTTP Archive, Web Almanac 2025 methodology. Test agents run from Google Cloud Platform locations based in the USA, and site selection from the Chrome UX Report.
- Chrome for Developers, Ad metrics in CrUX, 15 September 2026, and ad metrics methodology. Metric definitions, one-second viewport sampling, 75th percentile reporting and ads.txt eligibility.
- Chrome for Developers, CrUX methodology dimensions. Country inferred from IP address, and not available in the CrUX APIs.
- Yu, Yin and Zimmeck, Global Web, Local Privacy? An International Review of Web Tracking. Crawl 5 to 24 January 2025 from ten countries. Tracker connections per site, category shares and the consent-state experiment.
- Prasad, Vora, Lim, Hoang and Pasquier, RegTrack: Uncovering Global Disparities in Third-party Advertising and Tracking, MADWeb 2026. 743 websites from eight vantage points across browsers and consent states.
- Targeted and Troublesome: Tracking and Advertising on Children’s Websites. April 2023 crawls of 2,004 websites from five cities, desktop and mobile.
- Shifter, Residential Proxies geo-targeting documentation and sessions documentation.