What the estimates leave unresolved

This is an observational study. It cannot establish causation, prove why a page ranks, or reverse-engineer or reveal Google's algorithm.

Selection and sampling

Eligibility thresholds favor content with recurring impressions and enough observed days. Sparse and newly published content is less represented, which may make the results look more stable than a full portfolio.

Unmeasured confounding

Demand, competitor changes, SERP composition, seasonality, and measurement changes can affect both the recorded signals and subsequent position. The bias direction may differ across clients and months.

Proxy outcome

Average position combines query, device, geography, and impression mixes. A numerical change need not correspond to the same change in qualified traffic or business value.

Temporal drift

May and June scores differ materially. The coefficient-order correlation of 0.624 suggests partial continuity, while the June R² near zero shows that continuity did not produce dependable absolute estimates.

Sparse clicks

Many content windows record no clicks. This compresses the CTR reference curve at lower positions and limits separation among low-exposure items.