https://www.techtarget.com/searchenterpriseai/news/366624201/Google-expands-Gemini-model-Search-as-AI-rivals-encroach
As generative AI rivals start to erode its longtime control of search, Google on Tuesday introduced new capabilities across its software stack and expanded its Gemini models, including adding more AI features to Google Search.
"We are in the stage of AI where decades of research are becoming reality," said Sundar Pichai, Google's CEO, during the keynote presentation at Google I/O, the tech giant's annual developer conference.
Chief among the updates Google showcased at the conference is a revamp of its AI search service in a move that appears motivated by the early success of competitors such as OpenAI and Perplexity in making inroads into Google's longtime dominance in search.
After testing AI Mode in Search earlier this year, Google said it is now rolling it out in the U.S., which means that a new AI Mode tab will now appear in Search and in the search bar in the Google app.
Google is also bringing a custom version of its Gemini 2.5 model into Search for AI Overviews in the U.S. Users can also get deeper and more thorough responses with AI Mode with Deep Search.
In addition, Google will bring the capabilities of its multimodal agent Project Astra into Search. This means that users can talk back and forth with Search about what they see in real time with their camera.
Google's search-generative AI hybrid AI Overviews is also now available in over 200 countries and territories, the cloud provider said.
The revamp of Google Search comes after speculation that the rise of Perplexity's and OpenAI's model lineups, led by ChatGPT, could overthrow the search leader. However, that is clearly not the case, at least not yet, according to Gartner analyst Chirag Dekate.
"What the evidence clearly shows is that we are all seeing greater reliance on search, greater reliance on AI-infused search and AI-infused search experiences that Google is now enabling," he said.
For enterprises, especially those that want to use SEO, they would need to transform their strategies, Dekate said. This means that some enterprises might need to partner with Google to determine what a marketing strategy and SEO strategy mean in the era of AI Overviews.
"There is an implicit notion of trust, there is an implicit notion of trustworthy summaries, and that also means that this enables enterprises new avenues of marketing and search optimization," he said.
However, it is unlikely that the SEO model will be disrupted by this, but rather that Google will change its advertising model, said Alan Pelz-Sharpe, founder of Deep Analysis.
"That is really Google's business, selling advertising," Pelz-Sharpe said. "That's their income stream, so they may well have a model."
But Google still needs to be a little clearer about its goal with AI Mode, said Mark Beccue, an analyst at Enterprise Strategy Group, now part of Omdia. While Google said Gemini is still in the early days of AI Overviews, it's unclear what its long-term strategy for Search is, he said.
"Where is Search going? Are we going toward a personal assistant? How are they envisioning it?" he added.
Nevertheless, the fast pace of innovation in search has considerable potential, and products that emerge will be useful in research and for enterprises and consumers, Pelz-Sharpe said. But doubts remain.
"The big questions are still going to be, how accurate is it? Is it hallucinating?" he said. "The deeper you go down the rabbit hole of exploring and conversing, are you veering further away from accuracy or not? It's early days."
What is clear about Google is its vision of Gemini as a personal assistant and making it a "world model" that can plan and imagine new experiences just like the human brain does. The vendor plans to transform the Gemini app into a universal AI assistant.
Google's idea of a world model differs from that of Nvidia, which introduced a world model last year. While Nvidia thinks about applying world models to physical AI and robotics, Google sees it as a personal assistant, Dekate said.
"It is about customizing this assistant to your operating context or my operating context," he said. "These two operating contexts might intersect, but they have completely different operating principles and constraints."
Besides creating a personal assistant, Google also introduced new image and video generation models, Veo 3 and Imagen 4. Veo 3 can now generate videos with audio. In addition, Flow is a new AI filmmaking tool that enables users to create cinematic clip scenes and stories with Veo 3.
Google also revealed that Jules, its agent-based coding agent, is now in public beta.
As for Gemini, Google introduced new capabilities such as native audio generation and computer use into Gemini 2.5, the latest version of the chief rival to OpenAI's GPT models. It also introduced an enhanced reasoning mode in Gemini 2.5 Pro called Deep Think, which Google said can perform complex math and coding.
Intending to turn the Gemini app into a universal AI assistant, Google is bringing some Project Astra capabilities -- such as video understanding, screen sharing and memory -- into Gemini Live, Search and the Live API for developers. Gemini Live is a feature within the Gemini app with which users can have natural voice conversations with the model.
Google also introduced new capabilities in Workspace, such as real-time speech translation in Meet and smart replies in Gmail.
With all these new capabilities, there's a markedly different Google than before, according to Dekate.
"It is AI-native to its core," he said. From its consumer offerings to its enterprise products and even Search, AI technology is in every part of what the cloud provider offers, Dekate said.
He added that enterprises exploring Google's tech stack are assured that the models they build and deploy with Gemini have been tested and validated on a scale, especially with Google's Android phones.
"They are leveraging proven technologies as they build their own," Dekate said. "Enterprises that are using unproven or open models, in many cases, are experiencing the models for the first time themselves."
One technology that Google has proved works is Gemini's multimodal capabilities, particularly with Workspace and Search.
"Many of these multimodal capabilities in Workspace and beyond can enable a transformative learning experience ... for enterprises," Dekate said.
Moreover, with the new speech translation in Google Meet, Google revealed that spoken words can be translated into the listener's preferred language.
"That is a mind-blowing idea if it works," Beccue said. "If this works, that is a game-changing kind of work tool. That puts them in a place to take on Zoom, Teams and collaboration tools."
Enterprises looking to access all of Google's AI models and features can do so through Google AI Ultra. The new AI subscription plan is now available for $249.99 per user, per month.
Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence software and systems.
20 May 2025