Google AI Mode is set to transform Search. I'm concerned — and you should be as well.

      Google is advancing the integration of AI into its internet search functionality. Do you recall AI Overviews, which effectively summarizes content taken from various websites and displays it at the top of the Google Search page? This feature, which has had its share of inaccuracies, is now making its way into the US market, powered by the new Gemini 2.0 AI models. It no longer requires users to sign in with a Google account and is available to individuals of all ages. While this presents a level of risk, Google is also introducing a new AI Mode that applies similar treatment to the entire Search page.

      Currently in experimental stages, AI Mode transforms the conventional Google Search experience—with website links—into a conversational exchange akin to how AI chatbots deliver responses. While it offers a significant convenience, it could potentially lead to substantial inaccuracies, as evidenced by the history of AI Overviews.

      So, what is AI Mode for Google Search? The primary goal is to provide users with all the information they need—sourced from indexed websites—while eliminating the need to click through various sites and read pages to find answers. Users can pose follow-up questions in a natural language format rather than using keyword-heavy searches and can supply details that would otherwise require additional search queries.

      “It employs a ‘query fan-out’ technique, concurrently conducting multiple related searches across subtopics and different data sources to compile an easily understandable response,” the company explains. However, Google cautions that AI Mode may not always produce accurate results, even though internal testing has shown promising outcomes. In instances where the AI Mode lacks confidence in its summarized answer, it will revert to displaying a list of web search results, similar to the traditional Google Search experience.

      In its current form, the AI can present answers as a block of text or in organized tables, with plans to incorporate images and videos in the future. Currently, AI Mode is exclusively available to Google One AI Premium subscribers and will launch as an opt-in feature.

      This raises concerns for anyone relying on Google Search, particularly regarding accuracy. For instance, when I searched to confirm if we were in the year 2025, Google’s AI Overview incorrectly stated it was 2024, citing Wikipedia as its first source, which clearly indicated that the year was 2025.

      A history of associated risks surrounds this development. The concept of AI Mode for Google Search is theoretically about enhancing user convenience. However, the core technology is still facing several unresolved issues typical within the AI realm. One such issue is AI hallucinations, where the AI generates information and presents it confidently as truth.

      Google’s AI Overviews exemplify these errors, with mistakes still occurring regularly. A recent instance shared on Reddit demonstrated the AI Overview incorrectly asserting a driving rule in India. Despite such inaccuracies, the AI's output does not indicate that users should verify the information. "It’s so inaccurate and so buggy that I’m surprised it even exists,” remarked another report highlighting its blatant inaccuracies.

      AI Overviews appear as condensed snippets of information at the top of the Google Search page. Now, envision a complete page formatted as an extensive presentation, interwoven with a few links among the continuous text. Google has stated that AI Overviews will perform well in areas like “coding, advanced math, and multimodal queries.” Yet, it has previously mishandled factual information, particularly with the kinds of natural language questions that AI Mode is designed to handle.

      When asked if astronauts encountered cats on the moon, it mistakenly confirmed this as true and added that astronauts even cared for those lunar cats. Julia Feerrar, a digital literacy expert from Virginia Tech, noted that AI does not actually possess the answers to our inquiries, citing an example where Google AI Overview inaccurately identified Barack Obama as the first Muslim President.

      The repercussions of AI misinformation can be severe, especially concerning health and wellness queries. An analysis of over 30 million Search Engine Results Pages (SERPs) indicated that health-related searches are the top category where AI Overviews are featured. This is the same tool that previously recommended consuming a rock daily, adding one-eighth cup of glue to pizza, drinking urine to alleviate kidney stones, and stated that a baby elephant could fit in a human hand as of 2025.

      This is not the evolution of Search I desire. Despite Google's claims about advancements in AI models, the situation has not significantly improved. Recently, Futurism noted that AI Overviews falsely asserted that MJ Lenderman had won 14 Grammy awards. When I posed a simple question, such as “is it 2025” in the Google Search box, the AI Overview replied, “No, it is not currently the year 2025. The current year is 2024,” before adding unrelated information about the characteristics of non-leap years that had nothing to do with my question.

      I do not oppose AI; in fact, I frequently utilize tools like Gemini Deep Research and often rely on the Gemini 2.0 Flash AI model for creative