Google has introduced the Search Generative Experience (SGE), powered by AI and LLM technology, exclusively to participants of the Search Labs program. SGE aims to address longer, multi-step searches that cannot be adequately answered by a single website. Instead of the traditional display of blue links,
Search will generate an AI-powered “snapshot” by gathering information and tips from various sources, and then providing a concise summary to the user. The snapshot will be presented below the search bar with a distinct color background to distinguish it from the usual light/dark theme.
Google emphasizes that Generative AI is still in its experimental phase, but the responses generated by its refined large language models (LLMs), including MUM and PaLM2, strive to deliver high-quality and objective results without a personalized bias.
In instances where information is lacking or Google has limited confidence in its response, SGE will not generate a snapshot. On desktop web, websites that contributed to the answer will be showcased on the right-hand side, while on mobile devices, they will be presented in a carousel format. To provide transparency, users can expand the snapshot and examine the corroborating information on a sentence-by-sentence basis through the provided links. Additionally, at the bottom of the AI-powered snapshot, users will find suggested follow-up questions and the option to ask their own, initiating a conversational mode with a chatbot user interface designed for follow-up inquiries.
SGE caters to complex queries that necessitate insights from multiple sources. By employing advanced AI models, Google aims to furnish objective, high-quality responses devoid of any personal bias. The inclusion of follow-up questions and the interactive conversational mode with a chatbot UI enhances the search experience. Nonetheless, Google underscores that Generative AI remains in an experimental stage, and SGE will refrain from generating snapshots when information is insufficient or Google’s confidence in the response is low.
