Microsoft Corp. today finally unveiled its response to Google LLC’s experimental AI-powered search experience, with the limited preview launch of a new feature called Bing Generative Search.

The company said the new feature combines the foundation of Bing’s search results with the power of large and small language models, known as LLMs and SLMs. In doing so, it can understand the user’s search query and review millions of sources of information on the web to provide instant summaries and useful information related to the topic in question.

According to Microsoft, Bing Generative Search “dynamically matches content and generates search results in a new, AI-generated layout to fulfill the intent of the user’s query more effectively.”

The launch was announced by Microsoft’s Head of Engineering and Product for Copilot and Bing, Jordi Ribas, who explained that it represents a “meaningful step forward” in the evolution of AI-driven search.

Bing Generative Search is a direct response to the AI Overviews feature in Google Search that was launched in the U.S. in May. Google said at the time that AI Overviews was all about providing quick answers for users who don’t have the time to read through dozens of articles to get the information they need. It attempts to summarize the information the user is looking for, sourcing what it deems to be the most relevant insights from multiple websites.

Microsoft’s offering does much the same. In his blog post, Ribas offered a couple of examples of how it works. For instance, if a user types “What is a spaghetti western?”, Bing will surface information about the film genre’s history and origin, and highlight some of the best-known films in that category. It will also show a list of links and sources that it obtained the information from.

Just as with Google’s AI Overviews, users can dismiss the generative AI responses and summaries entirely and instead see a traditional results page.

No doubt Microsoft has taken extra care to try to avoid some of the controversy that followed the debut of Google’s AI Overviews, which made headlines for all of the wrong reasons in the days after it came online.

The new service was documented as generating a series of nonsensical, inaccurate and sometimes even potentially dangerous responses to user’s queries. In one instance, AI Overviews told a user that asked how to prevent the cheese sliding off a homemade pizza to “mix non-toxic glue with the cheese.” In another response, when someone asked about the best way to clean a washing machine, AI Overviews spat out a recipe for deadly mustard gas, along with instructions on how to go about making it.

Google isn’t the only AI search provider that has had facepalm moments, though. For instance, an alternative service called Arc Search made headlines for telling a news reporter with confidence that cut-off toes will eventually grow back, while Genspark happily recommended a list of weapons that can be used to kill someone effectively.

Another risk that Microsoft will want to sidestep is the lack of attribution, which befell a rival search engine called Perplexity. According to Forbes, Perplexity plagiarized articles from sites such as Bloomberg, CNBC and Forbes itself in a series of responses, without providing any credit or attribution links.

Microsoft will also want to be careful to ensure that its new search experience doesn’t put smaller websites out of business. One of the big fears around generative AI search is that they may steal traffic from the websites they obtain their information from. A study by Raptive Inc. earlier this year revealed that Google’s AI Overviews feature could hurt about 25% of publisher’s traffic, because of the way its summaries eliminate the need for users to click through to the actual source articles.

Ribas insisted that won’t be the case with Bing Generative Search, saying the company is looking closely at how generative AI search affects website traffic. He insisted the new experience has been able to maintain the number of clicks to websites. However, he didn’t provide any statistics or studies to back up his claims, instead citing “early data” that the company hasn’t yet published.

Image: Microsoft