Google is introducing a new experimental feature in Search called “AI Mode,” aimed at rivaling services such as Perplexity AI and OpenAI’s ChatGPT Search. The tech giant announced on Wednesday that the new mode is designed to let users ask complex, multi-part questions and follow-ups to dive deeper into a topic directly within Google Search.
AI Mode is launching this week for Google One AI Premium subscribers and can be accessed via Search Labs, Google’s experimental division.
The feature leverages a custom version of Gemini 2.0 and is especially useful for questions requiring deeper analysis and comparisons, thanks to its advanced reasoning, thinking, and multimodal abilities.
For example, a user might ask: “What’s the difference in sleep tracking features between a smart ring, smartwatch, and tracking mat?”
AI Mode will then provide a detailed comparison of what each device offers, along with links to the articles from which it sources the information. Users could then follow up with a question like: “What happens to your heart rate during deep sleep?” to continue the exploration.
Google says that in the past, conducting such comparisons or exploring a new concept would have required multiple queries using traditional search.
With AI Mode, users not only access web content but can also pull from real-time sources like the Knowledge Graph, real-world information, and shopping data covering billions of products.
“What we’re seeing in testing is people are asking questions that are about twice the query length of traditional search, and they’re also following up and asking follow-up questions about a quarter of the time,” Robby Stein, VP of Product at Google Search, told in an interview. “And so they’re really getting at these maybe harder questions, ones that need more back and forth, and we think, it creates an expanded opportunity to do more with Google search, and that’s what we’re really excited about.”
Stein pointed out that as Google has introduced AI Overviews — a feature that shows a summary of information at the top of results — the company has heard from users that they want access to similar AI-powered answers for a broader range of searches. This led to the creation of AI Mode.
AI Mode operates using a “query fan-out” approach, issuing multiple related searches simultaneously across various data sources and combining those results into a clear, user-friendly response.
“The model has learned to really prioritize factuality and backing up what it says through information that can be verified, and that’s really important, and it pays extra attention to really sensitive areas,” Stein said. “So this might be health, as an example, and where it’s not confident, it actually might just respond with a list of web links and web URLs, because that’s most helpful in the moment. It’s going to do its best and just be most helpful given the context of the information available and how confident it can be in the reply. This does not mean it will never make mistakes. It is very likely that it will make mistakes, as with every new kind of new and cutting edge AI technology that’s released.”
Since this is an early-stage experiment, Google notes that it will keep improving the user experience and expanding its capabilities. For example, the company intends to make the interface more visual and display data from a broader variety of sources, including user-generated content. Google is training the model to determine when to add hyperlinks (e.g., for booking tickets) or when to prioritize visuals like images or videos (e.g., for how-to searches).
Google One AI Premium subscribers can use AI Mode by opting into Search Labs and typing a question into the Search bar, then tapping the “AI Mode” tab. Alternatively, they can go directly to google.com/aimode. On mobile, users can open the Google app and tap the “AI Mode” icon located beneath the Search bar on the home screen.
As part of today’s announcement, Google also revealed that it has launched Gemini 2.0 for AI Overviews in the U.S. The company says AI Overviews can now handle more complex questions, starting with coding, advanced math, and multimodal queries. Additionally, users no longer need to sign in to use AI Overviews, and the feature is now also being rolled out to teen users.
