Google is preparing to bring generative AI into Search, while opening access to its LaMDA language model through an experimental chatbot called Bard. The plans surfaced in a presentation from Paris less than 24 hours after Microsoft announced an AI-powered Bing and a co-pilot for Edge.
Search expands beyond a single factual answer
Prabhakar Raghavan, Google's senior vice president in charge of Google Search, described how artificial intelligence has shaped the product since its launch nearly 25 years ago. He pointed to improved language understanding, multimodal search through Google Lens, and automatic translation into more than 100 languages with Translate.
The next step is to use generative text models within Search. Google says these models are intended to help with questions that do not have one fact-based answer. That changes the role of a search results page: instead of returning only links or a single established fact, it may also present generated text to help address a more open-ended query.
The company did not provide detailed information about how these answers will work. In the examples described in the announcement, Google's generated text did not cite sources directly in the text, unlike Microsoft's examples. That detail matters for readers trying to see where an answer came from, as well as for publishers whose work may inform search results.
Bard begins as an experiment
Google plans to roll out Bard, an experimental AI chatbot based on its LaMDA language model. The first access is planned for “trusted testers” this week. For the initial testing, Google is using a small LaMDA model to help scale the evaluation.
Bard and generative answers in Google Search are related, but the announcement presents them as separate ways to use the technology. Bard is a chatbot people can interact with, while the Search effort would place generative text into the existing search experience. The company has not supplied further detail here about how broad that Search rollout will be.
Raghavan also addressed concerns among publishers and content creators that AI-generated answers could reduce traffic to their sites. He said Google would prioritize approaches designed to send valuable traffic to a wide range of creators and support a healthy open web. The specific approaches were not disclosed, leaving a central question about how Google will balance direct answers with visits to original sources.
More AI tools and visual search are on the way
The Paris presentation also included multimodal search through Google Lens, described as “Multisearch,” and improvements to Google Maps. These announcements place the Bard chatbot within a broader set of changes to how people can search using different kinds of input and interact with information.
Google also discussed future possibilities for generative AI beyond text. Examples included creating 3D models from a few images and previewing kitchen cabinets in different colors. These were presented as potential uses, rather than capabilities already available to users.
Raghavan said the first customers would receive access to LaMDA through an API next month. Google expects other models to follow, including large image models, but did not reveal details about them. The announcement therefore points to a wider set of generative AI tools while leaving their availability and specific features open.
What the announcement leaves unanswered
Google's plans touch three groups at once: people seeking answers, publishers hoping to reach readers, and developers interested in language or image models. Bard offers an early route for testing a conversational interface, while AI in Search would put generated responses closer to the point where people look for information.
For now, the company has described its direction more clearly than its implementation. How Search will present sources, how it will direct visitors to creators, and what the future image models will do remain unspecified in the announcement. Those details will shape how useful these features are and how they affect the wider web.