Every time you watch a YouTube ad on your phone, you're probably paying more than you think — not just with your time, but with your mobile data. And if you're in a developing country, that cost can be genuinely significant. Our paper, published at WWW '24 (ACM Web Conference 2024), is the first independent empirical study of YouTube video ad data costs from the user's perspective. Prior work focused entirely on the platform or advertiser side. We wanted to know: what does the user actually pay?
What We Did
We built a distributed scraper using Selenium and PyTube, parallelizing collection across 20 machines over the course of a year. The result: a dataset of 17,600 YouTube videos and 46,600 ads spanning 8 countries — including the first large-scale corpus of YouTube data with per-video buffer metrics.
Then we analyzed where data actually goes during ad delivery. Not just what plays, but what gets buffered and thrown away.
The Findings
The headline result: video ads contain multiple latent and avoidable sources of data wastage that users never see but always pay for.
The biggest driver? Main-video resolution. At 720p, latent buffer loss is approximately 3× higher than at 360p — roughly 10.1 MB vs 3.4 MB per session. That buffered data is fetched, then discarded. You pay for it either way.
This isn't a minor rounding error. When you map these losses against the UN Broadband Commission's "1 for 2" affordability benchmark (1 GB should cost no more than 2% of monthly income), the burden on users in low-income countries is stark. A waste that's invisible in the US can represent real money for a user in Pakistan, Nigeria, or Indonesia.
Why It Matters
The web measurement community has spent a lot of energy studying performance from the network's perspective. We think the user's wallet deserves the same attention — especially as mobile internet remains the only internet for billions of people.
Our dataset and code are publicly available if you want to dig in further.
Read the paper.
Top comments (0)