AI Search Visibility Needs a Weekly Review Loop

The short version: Google’s dedicated Generative AI performance report is a separate Search Console view for a subset of sites. It exposes impressions and breakdowns by page, country, device, and date; it does not expose clicks, queries, or citations. Review those impressions weekly, then reconcile page-level movement with first-party site analytics and downstream actions outside the report.

The measurement contract is two-layered. Use the dedicated Search Console report only for the fields it actually exposes. Use site analytics for pageviews, clicks, and forms, and use the CRM or another verified system for qualified outcomes. A rise in one layer does not prove causation in another.

A useful weekly loop has four steps. First, record whether the dedicated report is available and annotate the review window. Second, compare current and prior page-level impression distributions, including the median and each URL rather than only the total. Third, compare the same URL cohort in site analytics without treating matching movement as attribution. Fourth, log confounders and choose a bounded next action: improve, reformat, merge, maintain, or stop.

Use Google’s report announcement for field definitions, then use How To Rank When Search Becomes a Chat to translate verified page-level signals into editorial work. A metric without a content rule only creates a prettier dashboard; the rule remains useful, original, sourceable content and crawlable contextual links.

The qualified next action depends on the evidence. If the dedicated report is unavailable, record the metric as unavailable and do not infer it from site analytics. If impression concentration rises on a page, inspect the visible answer and proof before changing it. If site actions move, verify the downstream system before claiming an outcome. The review loop exists to constrain decisions, not manufacture certainty.