Google's John Mueller acknowledged on 12 September that the AI Overviews performance report in Search Console — rolled out globally on 31 August as the primary first-party measurement tool for GEO performance — has fundamental flaws across its three most important metrics. Impressions are counted when the AI Overview block appears anywhere on a page, including when it scrolls off-screen before the user sees it or when it remains collapsed and unexpanded. Position data reflects where the Overview block appears on the SERP page, not where a specific URL appears within the Overview response — two entirely different pieces of information that require entirely different optimisation responses. URLs that appear only behind a "Show More" expansion in the Overview — a placement that significantly reduces user interaction — are invisible in reporting entirely. Mueller's assessment: "Position for these is hard to do in a way that makes it useful."

The practical consequences of these three failures compound each other. An impression count that includes unexpanded and off-screen blocks overstates reach and prevents accurate denominator calculation for any click-through rate metric. A position figure that measures SERP-level placement rather than within-answer placement produces a number that is high when AI Overviews appear at the top of the page — which they do by default — regardless of where a given URL ranks within the AI's response. And the complete exclusion of "Show More" URLs means the report systematically under-counts URLs that appear in AI responses for complex queries, which are precisely the queries where GEO effort is most concentrated. Practitioners building GEO strategies on Search Console's AI reporting are working from data that Google's own team acknowledges is unreliable on every dimension that matters.

LightSite AI proposed a server-log-based replacement framework on the same day (11 September) that addresses all three gaps. Four signals: AI bot traffic volume as measured in server access logs — a direct impression proxy that requires no cooperation from Google and cannot be inflated by off-screen or collapsed blocks; pages that receive repeated AI crawler visits as a content-priority indicator, identifying which articles and sections AI systems are returning to repeatedly rather than visiting once; human sessions arriving via AI search referrals in analytics — the conversion signal that matters most and the only one that directly connects AI visibility to business outcomes; and an AI click-through rate calculated as bot attention (crawler visits) divided by human demand (search referral volume), providing a relative efficiency metric comparable to organic CTR in traditional SEO. The critical insight in their analysis: approximately 12 per cent of pages absorbed approximately 50 per cent of all bot impressions — a concentration pattern that makes optimisation tractable. Most brands do not need to re-engineer every page for AI retrieval. They need to identify which 12 per cent of their content is already receiving disproportionate bot attention and ensure it is structured, accurate, and citation-worthy.

The combined picture: GEO measurement has no reliable first-party data from Google. The replacement signals are server-log-verifiable, provider-independent, and connect AI crawler activity directly to traffic and conversion outcomes rather than to estimated visibility share. For practitioners reporting GEO performance to leadership, the transition from Google's broken Search Console metrics to server-log-based measurement is now a credibility requirement. A visibility score built on impressions that include off-screen and unexpanded blocks is not a defensible metric when the underlying data methodology has been publicly disavowed by the tool's own creator.