Search Console's ranking data now misleads on AI Overviews

Google Search Console will tell a Holcim or an Adecco that their page ranks position 2 inside an AI Overview. It will not tell them that the citation sits at the bottom of a collapsed accordion of eleven other sources, one click away from invisible. Per Search Engine Journal, this is not a bug. It is how Search Console's "block flattening" method works, and it means the position metric millions of SEO teams rely on is now, for AI Overviews, close to fiction.

Block flattening takes the nested structure of an AI Overview, the main summary, the expandable "show more" links, the sub-panels of source citations, and irons it into a single flat ranking order for reporting purposes. A citation buried three folds deep gets counted the same way as one sitting in the visible summary. Search Engine Journal's reporting makes the mechanism plain: position 1 in Search Console can mean genuine top billing, or it can mean first among sources nobody scrolls to reach. The metric cannot tell you which.

Why a good rank now means less than it did

For a decade, position tracking was a reasonable proxy for outcomes. Rank 1 to 3 meant clicks; rank 8 meant obscurity. That correlation held because the container was linear: ten blue links, read top to bottom. AI Overviews are not linear. They are collapsible, personalised, and query-dependent in their layout, so the same brand can occupy a flattering position number while sitting inside a fold that most users never open. The number survives; the meaning it used to carry does not.

This matters more for institutional and enterprise brands than for a consumer retailer chasing transactional queries. A bank's compliance-adjacent content, an industrial group's technical documentation, a multilateral's policy explainer: these are exactly the source types AI Overviews like to cite, because Google's summarisation layer favours authoritative, structured, low-ambiguity content. Recall firms like CGAP or ISO have this experience already: audits show "citations" that never translate into referral traffic. Now there is a mechanistic explanation for the gap, and it is the reporting tool itself, not the content.

The practical failure mode is a false negative in reverse. Comms and marketing teams building dashboards for leadership will see a stable or improving position metric and conclude AI visibility is healthy. Budget gets allocated on that story. Meanwhile actual referral traffic and conversions from AI Overview citations stay flat, because the citation was never in a place a human would find it. The dashboard says one thing; the analytics platform, if anyone bothers to cross-check, says another. Most teams do not cross-check, because Search Console has trained a generation of SEOs to treat position as the load-bearing number.

Search Engine Journal's recommendation, buried in the practitioner detail, is the correct one: stop reading position in isolation and correlate it against Search Console's own click and impression data for the same query set, then reconcile against server-side analytics for actual sessions. If position holds steady but clicks and conversions from those queries are falling or flat, the citation is decorative. That is a real, actionable diagnostic, and it is more work than glancing at a rank number, which is precisely why most teams will keep glancing at the rank number.

There is a governance implication here that goes beyond marketing ops. Any organisation reporting AI search visibility to a board, a donor, or a regulator, on the strength of Search Console rank data alone, is reporting a number that Google itself has structurally decoupled from the outcome it used to represent. For sectors where reputational and disclosure accuracy carry weight, financial services and multilaterals chief among them, that is not a rounding error. It is a metric that will need a footnote, and eventually, a replacement.