Measure traffic changes around AI Overviews

What changed, and where?

Start with a defined set of landing pages and a date window. Record clicks, impressions, click-through rate and average position from the ordinary Search Console performance report. Compare equivalent weekdays and, where relevant, the same seasonal period. Keep the site, country, device and search-type filters consistent.

List other changes around the decline: a migration, indexing issue, altered title, stock availability, consent configuration, campaign ending or shift in demand. These are investigation candidates. The presence of an AI Overview does not establish which factor caused a traffic loss. Google’s Search performance documentation explains the ordinary report’s metrics and why results can vary with search conditions.

Where can you see Google AI visibility now?

Open the Search Console generative AI performance report for the same property. Its current documentation includes impressions for AI Overviews and AI Mode, with pages, countries, dates and devices. Google states that the report rolled out worldwide by 31 August 2026. Use the current report documentation to check access, aggregation and data limitations.

This report measures links shown in those Google features. Its documented dimensions do not provide a matching AI query, click or CTR breakdown. Avoid dividing ordinary search clicks by generative AI impressions and labeling the result “AI CTR.” Likewise, a Gemini chat or API answer is a different observation from a Google Search AI Overview.

Create an investigation table with separate columns for ordinary search clicks, ordinary impressions, generative AI impressions, acquired sessions and business outcomes. Keep original report exports and filters. Page totals and property totals can differ because the reports aggregate differently.

Before exporting, preserve the report’s unavailable-value markers and preliminary-data status in your notes. Google documents that values displayed as “~” or “-” become zeros in downloads. Keep those exported zeros distinguishable from measured zero impressions when comparing periods.

What would a useful diagnosis look like?

Consider an illustrative support page whose Google clicks declined while its generative AI impressions increased. That is evidence of two observed changes. It does not reveal how many people would otherwise have clicked, whether the AI feature caused the entire decline or whether those users later converted.

Inspect the page’s usefulness for the remaining visitor task. A calculator, current comparison, original dataset, booking action or detailed troubleshooting step may give readers a reason to continue beyond a summary. Choose an improvement grounded in actual user needs and available evidence. Google’s AI optimization guide keeps foundational SEO and valuable original content central; it does not prescribe special AI markup or guarantee selection.

How do you evaluate the response?

Record the page revision and its intended effect before release. Recheck accessibility and indexing, then compare the same reports on the agreed cadence. In GA4, Google AI Overviews and AI Mode belong to Organic Search; the AI Assistant channel covers recognized assistant referrers separately, according to Google’s channel definitions.

Report what improved, what stayed flat and what remains uncertain. A recovered visit, a new citation and a qualified inquiry are different outcomes. If the evidence does not identify the cause of the original loss, state that the cause remains unresolved.