Google's AI Overviews are way too chatty and personal for a search engine, and it shows

At I/O this year, Google sold the vision of a new Search, one that offers hyperpersonalized responses and gets a host of AI additives. As Google moves on to eliminate the distinction between Search and Gemini, it wants vanilla responses to be flavored by more personalized AI responses. But some of those may not be to everyone’s taste, especially when they deter users from getting the results they want. And we’re seeing fresh examples of that in action.

Google Search’s AI Overviews are once again showing their buried chatbot traits. Just like last month, when we learned that AI Overviews were getting perplexed by words such as “ignore” or “remember,” they’re repeating a similar pattern with phrases.

Redditor Timekiller_74 recently shared their experience of Google’s AI Overviews “flirting” with them. They had apparently searched for “you mean a lot to me,” expecting Google to tell them what the phrase implies, but instead got Gemini to bypass search results and tell them, “the feeling is mutual.” A typical search would show results about the phrase’s meaning and common usage — as it still does under the AI Overview. However, with the AI-generated section taking up a large chunk of the space, the useful results get overshadowed, prompting users to scroll more.

I tested searching for the phrase at my end and got similarly cheeky results. I tested similarly endearing phrases and got a blend of just responses with plain endearment, as well as some logical responses telling me about the meaning of the phrase.

In certain cases, it also suggested songs with titles matching the phrase. What I realized was that the heartening answers in AI Overviews were more likely to appear in response to generic terms, rather than to less specific sayings or idioms.

Google’s “nice” AI responses could be problematic

Ironically, none of the responses were as intimate or heartwarming when I said the same to chatbots. Gemini, ChatGPT, and Claude all responded with a tone of gratitude but showed no signs of affection, sounding more professional than friendly. Notably, while Claude and ChatGPT also emphasized that they were merely AIs not worth loving, Gemini omitted this from the Flash models’ responses, including it only when using Gemini 3.1 Pro.

Warning users about its inanimate nature also becomes imperative, as humans have increasingly shown signs of emotional connection and dependence on AI, and some of these “relationships” haven’t ended on a good note.

Last month, Google addressed the previous error and confirmed it is aware of these issues with AI Overviews. A fix for the same has yet to come, even though Google was prompt to fix another issue where AI Overviews confidently told some users it was still 2024.

With Google now eyeing a more personalized search experience, with Gemini already baked in, it becomes even more important for it to ensure such slipups are minimized. Compared to Gemini, Apple plans to take a different approach with Siri AI, limiting sycophancy, or an AI’s tendency to blatantly agree and appease you. Apple recently said that it prefers Siri to offer cold, practical responses instead of warm one that make it seem like a “companion.”