Google made Gemini 3.5 Flash the default model behind AI Mode in Search worldwide, the company confirmed at its I/O developer conference on Tuesday. The swap matters because AI Mode, Google’s conversational search experience, now passes more than one billion monthly users, which makes the model change the single largest shift in how Google answers queries since the surface launched.
Gemini 3.5 Flash is Google’s lowest-latency frontier-class model. The company says it outperforms the older Gemini 3.1 Pro across its internal benchmark suite, with the widest margin on complex coding tasks, and that it generates output tokens roughly four times faster than competing frontier models. For a search surface, latency is not a vanity metric. A faster model lets Google run more reasoning steps inside the same response window, which expands what AI Mode can attempt before a user loses patience.
The user number frames the stakes. Google reported at the keynote that AI Mode queries have more than doubled every quarter since launch. AI Overviews, the generative answer block that appears above traditional results, now reaches more than 2.5 billion monthly users. AI Mode at one billion is the smaller surface, but it is the one that removes the blue-link results page entirely for an engaged session. That is the surface where publisher referral traffic is most at risk.
The practical consequence for search teams is a moving target. When Google changes the default model, it changes the retrieval and synthesis behavior that decides which pages get cited in an answer. A site that earned consistent AI Mode citations under the previous model has no guarantee of the same placement under Gemini 3.5 Flash. Citation patterns are a function of the model, not a fixed property of a page.
Google did not publish a citation-behavior comparison between the old and new defaults. That absence is the story for practitioners. Without a before-and-after on which domains get cited and how often, teams cannot tell whether the model change is neutral for their visibility or quietly redistributes it. The only reliable read is first-party measurement.
There is a second-order effect worth tracking. A faster model encourages Google to run multi-step agentic queries inside Search, where the system reasons across several sources before answering. More reasoning steps usually means more sources consulted per query, which can widen the citation pool. It can also concentrate citations on the few domains the model treats as most authoritative for a topic. Both outcomes are plausible, and they point in opposite directions for a mid-sized publisher.
Google’s framing positions the model swap as a quality and speed upgrade with no downside for the open web. The keynote did not include any independent measurement of how the change affects outbound clicks. Publishers should treat the claim as a company position, not a verified result.
Search and content teams should start a citation log now: for a fixed set of priority queries, record which AI Mode answers cite the site and which cite competitors, then re-check the same queries in thirty days. That baseline is the only way to separate a real visibility shift from normal SERP noise once Gemini 3.5 Flash is fully live.
ATTRIBUTION: Reported by Google on its company blog and confirmed at the Google I/O 2026 keynote on May 19, 2026.
