AllSERP: Exhaustive Per-Element Enrichment of the Versatile AdSERP Dataset

K. Andrew Edmonds has released AllSERP, a major enrichment of the AdSERP dataset that dramatically expands its utility for information retrieval research. AdSERP originally shipped 2,776 full-page screenshots with 150 Hz eye tracking, mouse telemetry, scroll signals, and pupil data collected against real Google SERPs before AI Overviews — but its bounding boxes only covered ad surfaces, accounting for just 15.5% of attributable clicks. AllSERP fills that gap with pixel-accurate bounding boxes for organic results, widgets, and inter-result gaps, using computer vision anchored to screenshots and an HTML parser to classify 13 element types.

The new enrichment achieves 91.7% click attribution across the entire corpus, flagging the remaining unattributed clicks at the trial level. The Phase C ad-vs-non-ad partition shows zero disagreements across 38,250 classifications with the original ad rectangles, confirming internal consistency. Edmonds ships the full pipeline, per-trial JSONs, a corpus CSV, and a browser-based replay viewer — all reproducible from the AdSERP Zenodo volume. Researchers can now perform per-element click, fixation, regression, and above-fold analyses that were previously impossible, unlocking deeper understanding of how users interact with search results beyond just ad clicks.