Why AI Overview CTR studies report losses from 15% to 58%

Direct answer

Practical steps

1. Define the decision, baseline period, population and metric before looking at the result.
2. Record the primary source, collection date, denominator and exact meaning of every field.
3. Segment pages, queries or locations before comparing scenarios; preserve raw totals and missing data.
4. Act on a testable gap, monitor business outcomes and revisit the conclusion when platforms or coverage change.

Example

A site with 100,000 impressions and 5,000 clicks should not subtract 58% from all clicks. First estimate which impressions are exposed, then model 15.49%, 37% and 58% relative losses for that subset and compare them with observed changes.

Common mistakes

• Turning one study average into a guaranteed forecast for every website.
• Calling an association causal or publishing secret algorithm weights and review thresholds as facts.
• Optimizing a proxy while ignoring qualified visits, leads, revenue, accessibility and user trust.

Limitations

• Search features, reporting access and local results change by query, place, device, time and user context.
• Exports and observational studies can reveal patterns, not the hidden ranking system or the counterfactual result for one site.

Continue learning