Abstract: Digital advertising often conflates ad loading with user attention, leading to "fraud impressions" that bias machine learning models used in Real-Time Bidding (RTB). This paper presents an empirical study using a custom tracking system to determine the optimal thresholds for ad viewability. By comparing 11 measurement rules against human recall, the researchers identify a "75/2" standard—75% pixel visibility for at least 2 continuous seconds—as the most effective predictor of whether an ad was actually seen.
Introduction
In the display advertising ecosystem, advertisers are typically charged per impression. However, a significant gap exists between an ad being loaded by a browser and being seen by a user. Ads may remain at the bottom of a page, be hidden behind other windows, or be scrolled past too quickly to register. These "unseen" ads are categorized as fraud impressions, which not only waste budget but also degrade the accuracy of click-through rate (CTR) estimators by introducing biased data into machine learning models. This study aims to establish a measurement standard that aligns technical metrics with human memory and attention.
Methodology
The researchers conducted a controlled user study with 20 participants (average age 23.2) who browsed five custom webpages containing various articles and 60 standardized ad creatives (300x250 pixels). To monitor viewability, they developed a custom tracking system using jQuery to measure two primary variables for every ad on the page: the percentage of pixels visible within the viewport and the duration of that exposure.
The study evaluated 11 different measurement rules—combinations of pixel thresholds and time durations—against a "ground truth" established through participant recall tests. After browsing, participants identified which ads they remembered from a list of 60 (30 seen, 30 unseen) and rated their confidence in those memories.
Key Insights
- The 75/2 Rule: The study found that 75% pixel visibility for at least 2 seconds (the "75/2" rule) was the optimal threshold for predicting recall, outperforming the then-standard 50% for 1 second.
- Time vs. Pixels: While both factors are relevant, exposure time proved to be a significantly stronger predictor of a user's confidence in their recall than the raw percentage of pixels displayed.
- Attention Spillover: Counter-intuitively, users who were more engaged with the article content were also more likely to remember the ads. This suggests that high-quality content increases overall visual engagement rather than distracting from advertisements.
- Binary Recall Thresholds: The data showed that 97.3% of remembered ads reached 100% pixel visibility at some point. Users rarely remember ads that are only partially displayed.
Results
The researchers utilized the F1 score (the harmonic mean of precision and recall) to evaluate the performance of the 11 measurement rules. Stricter thresholds significantly reduced the "noise" of unviewed impressions.
As shown in the results, Measurement 6 (75% pixels for 2–4s) and Measurement 9 (75% pixels for 4+s) yielded the highest average F1 scores. User confidence was also notably higher for ads viewed for more than 4 seconds compared to those viewed for 2 seconds or less, confirming that duration is a critical component of viewability.
Conclusion
This study demonstrates that current industry standards for ad viewability may be too lenient, leading to inaccurate performance data. The proposed 75/2 rule provides a more human-centric metric for digital advertising. While the study is limited by a small sample size and a single ad format, it provides a foundation for future research into diverse ad types (such as video or mobile interstitials) and the long-term ROI impact of implementing stricter viewability standards in live commercial platforms.
Summary updated