Tracking the Trackers: A Large-Scale Analysis of Embedded Web Trackers

Sebastian Schelter, Jérôme Kunegis

Abstract

We perform a large-scale analysis of third-party trackers on the World Wide Web. We extract third-party embeddings from more than 3.5 billion web pages of the CommonCrawl 2012 corpus, and aggregate those to a dataset representing more than 41 million domains. With that, we study global online tracking on two levels: (1) On a global level, we give a precise figure for the extent of tracking, and analyse which trackers (and subsequently, which companies) are used by how many websites. (2) On a country-specific level, we analyse which trackers are used by websites in different countries, and identify the countries in which websites choose significantly different trackers than in the rest of the world. We find that trackers are widespread (as expected), and that very few trackers dominate the web (Google, Facebook and Twitter), except for a few countries such as China and Russia.

Introduction

While the presence of "cookies" and "trackers" is common knowledge, quantifying their global reach is a significant technical challenge. Most previous research relied on limited samples or specific geographic regions. This study, presented at ICWSM 2016, provides a massive-scale audit of the third-party tracking ecosystem, analyzing billions of web pages to map the concentration of web surveillance and its correlation with national borders.

Methodology

The researchers utilized the Common Crawl 2012 corpus, processing 3.5 billion web pages (approximately 210 TB).

  • Extraction: A custom pipeline scanned HTML for script, iframe, link, and image tags pointing to third-party domains, including parsing JavaScript for URI patterns.
  • Aggregation: Data was grouped by "pay-level domains" (e.g., google.com) to identify the entities behind the trackers.
  • Ranking: In the absence of private traffic data, the authors used PageRank (calculated from a graph of 623 million hyperlinks) as a proxy for a website's importance and traffic.
  • Network Modeling: The relationship was modeled as a bipartite graph between websites and trackers.

The Global Landscape of Tracking

The study reveals a "Power Law of Surveillance." While thousands of niche trackers exist, the top 1% are integrated into millions of sites. The distribution coefficient (1.725) mirrors the link structure of the web itself.

A "Big Three" oligarchy—Google, Facebook, and Twitter—dominates the landscape. In 46 of the 50 countries studied, these three companies account for more than half of the "rank share" of the top ten trackers. Interestingly, there is a strong positive correlation between a country's Democracy Index and the dominance of these US-based trackers.

Comparison of tracker rank shares across different countries, highlighting the dominance of US tech giants in most regions and domestic alternatives in China and Russia.

National Outliers

The global uniformity of tracking is striking, with countries like Ireland, Germany, and France showing nearly identical tracker profiles. However, four countries—China, Russia, Iran, and Ukraine—stand out by favoring domestic trackers. For instance, Yandex is prominent in Russia, while CNZZ leads in China.

Key Results: Google's Dominance

Google's footprint is unparalleled. The domain google-analytics.com appears on nearly 25% of all domains. When weighted by PageRank, Google's "rank share" exceeds 0.50, indicating visibility into more than half of the "important" web. Furthermore, five of the top ten most dominant third-party entities are owned by Google.

Top third-party trackers ranked by their share of the web's PageRank, showing the overwhelming lead of Google-owned domains.

Limitations

The study's static analysis likely missed "transient trackers" that require JavaScript execution or user interaction. Additionally, the 2012 dataset, while massive, does not reflect the modern rise of mobile app tracking and advanced fingerprinting. Finally, while PageRank is a robust academic proxy, it may not perfectly capture actual user attention in "walled garden" ecosystems.

Summary updated

Abstract

We perform a large-scale analysis of third-party trackers on the World Wide Web. We extract third-party embeddings from more than 3.5 billion web pages of the CommonCrawl 2012 corpus, and aggregate those to a dataset representing more than 41 million domains. With that, we study global online tracking on two levels: (1) On a global level, we give a precise figure for the extent of tracking, and analyse which trackers (and subsequently, which companies) are used by how many websites. (2) On a country-specific level, we analyse which trackers are used by websites in different countries, and identify the countries in which websites choose significantly different trackers than in the rest of the world. We find that trackers are widespread (as expected), and that very few trackers dominate the web (Google, Facebook and Twitter), except for a few countries such as China and Russia.
BibTeX
@inproceedings{schelter2021tracking,
  author = {Sebastian Schelter and Jérôme Kunegis},
  title = {{Tracking the Trackers: A Large-Scale Analysis of Embedded Web Trackers}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2021},
  doi = {10.1609/icwsm.v10i1.14769},
}
Figure