More links in AI Overviews, fewer that leave Google
More citations in Google's AI Overviews don't mean more visitors reaching your site, and Search Console can't yet tell the difference.
Key takeaways
- Google AI Overviews now show more links, per Search Engine Journal, but a growing share route to AI Mode, not external websites.
- Search Console measures AI Mode and website clicks differently, making citation counts a weak proxy for actual referral traffic.
- Institutions that depend on readers reaching primary source documents, not summaries, lose the most when citations don't convert to visits.
- Brands reporting rising AI Overview citations to leadership may be reporting appearances, not reach.
- Measurement built around citation volume needs replacing with tracking of actual traffic from AI surfaces.
Google's AI Overviews now carry more links than before. Search Engine Journal reports that a growing share of them do not go to websites at all: some route users to AI Mode, Google's conversational search layer, rather than off Google's own property. Search Console, Google's own measurement tool, counts these two destinations differently, which means the headline number of "links shown" has quietly become a poor proxy for "traffic sent."
This is a bookkeeping change with real teeth. For years, the working assumption among SEO teams was that AI Overviews, whatever their faults, still functioned as a directory: cite a source, link a source, send a click. Publishers grumbled about summarisation cannibalising the click, but the link was at least a link to somewhere external. What Search Engine Journal's reporting makes clear is that Google has built a second exit ramp inside the answer box, one that keeps the user inside Google's own conversational interface rather than depositing them on a brand's site. More links, less web.
Why the link count was always the wrong metric
Marketing teams have spent two years asking "are we cited in AI Overviews?" as though citation were the finish line. It never was. The finish line, for a bank's product page or a UN agency's policy brief, is a human reading the actual source, not a fragment of it laundered through Google's summary layer. A citation that routes to AI Mode rather than to holcim.com or undrr.org is a citation that never reaches the institution's own words, own caveats, own framing. It is visibility without contact.
That distinction matters more for regulated and reputational sectors than for e-commerce. A retailer losing a click to an AI Mode follow-up still probably converts the intent somewhere in Google's ecosystem, and a paid ad might catch the same user later. A multilateral organisation or a financial regulator has no such fallback. Its value lies precisely in the reader engaging with the primary document: the loss data, the methodology footnote, the actual text of a standard. If Google's own interface intercepts that reader before they reach the source, the institution has been cited into invisibility. It shows up in nobody's traffic logs and nobody's brand-recall study.
The measurement gap will outlast the feature
Search Console's inability to cleanly separate "sent to your website" from "sent to AI Mode" is not a rounding error. It means the primary dashboard that enterprise SEO and comms teams use to justify AI-search investment is currently blind to the exact mechanism Search Engine Journal describes. Any brand reporting "AI Overview citations up" to a CMO this quarter is very possibly reporting a vanity metric: more appearances, not more reach. Boards asking whether search visibility investment is paying off deserve a truer number than the one Google currently exposes.
The mechanism also changes the incentive structure for what gets written. Content optimised purely to earn an AI Overview citation, thin, extractable, FAQ-shaped, is now competing to be summarised into a system that may never release the reader onto the open web at all. That is a bad trade for institutions whose authority depends on depth: a central bank's working paper, an industrial group's technical standard, a philanthropic foundation's evaluation report. These are documents built to be read in full, with caveats intact, not reduced to a two-line answer card that satisfies the query and closes the loop inside Google.
None of this argues for abandoning AI Overview optimisation. It argues for treating citation as an intermediate metric, not an outcome, and for building measurement that tracks actual referral traffic from AI surfaces rather than trusting Google's own Search Console to distinguish "sent to you" from "sent to itself." Brands that keep score using the old metric will look successful right up until someone asks how many of those visitors ever landed on their site. For sectors selling trust rather than products, that gap between citation and contact is the whole business.