# Google placed AI Mode in Search Labs as a query-fan-out experiment for complex questions Google introduced AI Mode in Search Labs on March 5, 2025, using a custom Gemini 2.0 version and query fan-out for a limited opt-in experiment. Canonical: https://rangoon.ai/insights/archive-google-search-ai-mode-march-2025/ Author: Rangoon Editorial (https://rangoon.ai/insights/editorial/) Published: 2026-10-24 Updated: 2026-10-24 Event date: 2025-03-05 Tags: [Google](https://rangoon.ai/insights/tags/google/), [Google Search](https://rangoon.ai/insights/tags/google-search/), [AI Mode](https://rangoon.ai/insights/tags/ai-mode/), [search](https://rangoon.ai/insights/tags/search/) Source note: Historical reporting: Google introduced AI Mode as an early Search Labs experiment on March 5, 2025. The company described a custom Gemini 2.0 version, follow-up questions, web links and a query-fan-out technique across subtopics and sources; initial invitations were for Google One AI Premium subscribers. A June Google design post confirms the March Labs phase was followed by later U.S. rollout work. The March experiment is not treated as general availability, a ranking guarantee or a claim that AI answers were always correct. Image: Query fan-out with source review. Original Rangoon explanatory diagram: Source boundary: dated primary release reporting; original editorial analysis. No vendor interface, photograph, measurement result, or endorsement implied. ## Key takeaways - AI Mode began as a limited opt-in Search Labs experiment, separate from AI Overviews and later public rollout stages. - Query fan-out described a retrieval approach that expands a question into related searches; it did not guarantee accuracy or exhaustiveness. - Links and fallback web results remained part of the launch design because early AI answers could be unhelpful or wrong. ## The March launch was a limited Search Labs experiment [source 1](https://blog.google/products-and-platforms/products/search/ai-mode-search/) On March 5, 2025, Google introduced AI Mode as an early experiment in Search Labs. Google said the feature was built for complex, multi-part questions and follow-ups, using a custom version of Gemini 2.0. It began by inviting Google One AI Premium subscribers to try it. That access state is central to the event: an invitation-based Labs test is not a general Search launch, a promise of availability to every user, or a statement that an AI Mode answer would replace standard results for all queries. The same post expanded AI Overviews to more people and upgraded some U.S. AI Overviews with Gemini 2.0. AI Overviews and AI Mode were related but separate product surfaces. AI Overviews could appear within ordinary Search results, while AI Mode was an opt-in experience aimed at more exploratory queries. Treating them as one feature erases both the Labs boundary and the difference between a result-page summary and a dedicated follow-up conversation. ## Query fan-out was a retrieval strategy, not a truth guarantee [source 1](https://blog.google/products-and-platforms/products/search/ai-mode-search/) Google described AI Mode as using query fan-out: issuing several related searches across subtopics and data sources, then bringing the results together into a response. That can widen the set of evidence a search experience considers when a question has multiple constraints. It can also create new synthesis work. A system still needs to decide which sources are relevant, reconcile disagreements and present links in a way that lets a person inspect where an assertion came from. Google said AI Mode could draw from web content, the Knowledge Graph, real-world information and shopping data. Access to several systems does not make every piece of information equally current, authoritative or suitable for a user’s purpose. A product listing can change, a web page can be incomplete and a Knowledge Graph relationship can lack the context a decision needs. Query fan-out describes breadth of retrieval, not a substitute for checking original sources when an answer carries material consequences. ## Google documented early-product uncertainty alongside the experiment [source 1](https://blog.google/products-and-platforms/products/search/ai-mode-search/) [source 2](https://blog.google/products-and-platforms/products/search/ai-mode-development/) Google wrote that it would aim to show AI-powered responses when helpful, but would show web search results when it lacked confidence in helpfulness or quality. It also said early AI responses could be wrong or appear to express a persona or opinion. Those qualifications are part of the launch rather than a side note. They explain why links, follow-up questions and ordinary results remained meaningful elements of the product instead of evidence that a generated summary had settled a topic. The June design post later described how internal teams and trusted testers informed the March Labs stage before subsequent U.S. rollout work. That later account is context for the product sequence, not proof that the March experiment had already reached general availability. It also reinforces the value of separating an interface experiment from a mature search contract: the first can collect feedback and change rapidly, while users and publishers need dated records to understand what was available and how it behaved at a particular time. ## AI Mode changed query interaction without guaranteeing publisher or user outcomes [source 1](https://blog.google/products-and-platforms/products/search/ai-mode-search/) [source 2](https://blog.google/products-and-platforms/products/search/ai-mode-development/) AI Mode made it easier to phrase a question in natural language, ask follow-ups and receive a synthesized response with web links. That can alter how people discover information, but it does not promise traffic to every source, ranking for every page or a correct decision by every searcher. Search systems choose and present information under changing quality and relevance systems. A publisher still needs clear, accurate pages and a user still needs to evaluate the source behind a consequential claim. This record is limited to March 5. It does not read later tabs, multimodal features, agentic functions, market expansion or product interface changes back into the initial Labs experiment. The launch supports a precise historical claim: Google tested an AI-first search mode that expanded retrieval through related queries and presented follow-up interaction, while explicitly retaining quality limitations and the need to fall back to ordinary web results. ## Sources 1. [Google Blog: Expanding AI Overviews and introducing AI Mode](https://blog.google/products-and-platforms/products/search/ai-mode-search/) 2. [Google Blog: A closer look inside AI Mode](https://blog.google/products-and-platforms/products/search/ai-mode-development/) ## Related reading - [Google’s Gemini Robotics On-Device put a vision-language-action model on the robot](https://rangoon.ai/insights/archive-google-gemini-robotics-on-device-june-2025/) - [Google put Guided Learning into Gemini with LearnLM-informed study tools](https://rangoon.ai/insights/archive-google-gemini-guided-learning-august-2025/) - [Google launched Gemini Canvas for iterative documents, code and prototype previews](https://rangoon.ai/insights/archive-google-gemini-canvas-march-2025/) ## More information - [Documentation index](https://rangoon.ai/llms.txt) - [AI access and policies](https://rangoon.ai/ai/) - [Source repository](https://github.com/hypler-dev/rangoon)