MOUNTAIN VIEW, CALIFORNIA – AUGUST 13: The Google logo is displayed in front of company headquarters … More during the Made By Google event on August 13, 2024 in Mountain View, California. Google announced new Pixel phones, watches and AI features at the event. (Photo by Justin Sullivan\/Getty Images)<\/p>\n
Getty Images<\/p>\n
A year after adding AI search responses to the top of its SERP pages in the form of something called \u201cAI Overviews\u201d, Google is going a step further.<\/p>\n
This week, the company is unveiling its \u201cAI mode,\u201d where instead of a list of search results, what the user get is Google\u2018s own take on a model-generated summary of whatever was put into the search box.<\/p>\n
Media reporting shows<\/a> that Google began testing this feature in March, after AI Overviews has been active on the site for over a year. AI Overviews was said to have 1.5 billion users a year ago, so it\u2019s likely that AI Mode will have quite a few people taking the model\u2019s responses, rather than clicking through hyperlinks.<\/p>\n However, like AI Overviews, AI Mode can make mistakes.<\/p>\n Google provides an explanation page showing why this is the case, and asking users to give feedback if they see something that\u2019s not right.<\/p>\n \u201cAs with any early-stage AI product, AI Mode doesn\u2019t always get it right,\u201d explain spokespersons in this piece of user guidance. \u201cFor example, in some cases, AI Mode may misinterpret web content, or miss context, as can happen with any automated system in Search. When this does happen, we encourage you to give feedback<\/a> through thumbs up \/ thumbs down so we can continuously improve.\u201d<\/p>\n There is precedent for getting some pretty bone-headed answers from the model: our own MIT Technology Review previously covered Google AI Overviews encouraging users to put glue on pizza, and eat rocks.<\/p>\n Demos of the New Search<\/strong><\/p>\n A demo video from Google shows people asking various in-depth questions about a popsicle bridge, homebuying and other issues, and the new search engine feature replying accordingly, with cogent, multi-paragraph answers.<\/p>\n However, Googling the kinds of phrases that we traditionally used in older days brings some vague results in AI Mode, as when I supplied the phrase \u201cbig dress,\u201d which generated the following: \u201cYou seem to be interested in the term big dress, which can have several meanings,\u201d and information on everything from voluminous ballgowns to the Guinness book of records.<\/p>\n Going Further<\/strong><\/p>\n After reading several comments on Reddit about the AI Overviews feature, which Redditors have had more time to digest, I asked Google\u2019s AI Mode the following question:<\/p>\n \u201cWhy are people freaking out about Google\u2019s AI Mode? Is it a problem with the muscle memory of the interface?\u201d<\/p>\n In responding, Google put the words \u201cfreaking out\u201d in quotation marks, which seems like a pretty human thing to do, and cited the following concerns:<\/p>\n \u00b7 Accuracy and misinformation (we covered this above)<\/p>\n \u00b7 Impact on content creators on the web<\/p>\n \u00b7 Erosion of trust and user experience<\/p>\n That second one is critically important because of big changes to ad revenue models on the Internet.<\/p>\n Turning over the Apple Cart<\/strong><\/p>\n Basically, using AI results leads to fewer clicks on the SERP results which harms websites that need ad revenue. Zero click searches where users just take the AI feedback don\u2019t bring in web traffic to other parts of the Internet.<\/p>\n Then there are those lawsuits going on about intellectual property of publishers.<\/p>\n As for muscle memory, it\u2019s evident that millions who have become used to entering a keyword and getting a dependable set of links will be looking around themselves wildly when they can\u2019t find them.<\/p>\n