The traditional search engine, with Google as its leading representative, is losing ground to the rise of queries through artificial intelligence and chatbots, which offer more structured answers, although not always accurate. For years, Google has dominated access to online information, making its search engine an indispensable tool and thus consolidating its position in the digital ecosystem. However, the emergence of large language models (LLMs) has begun to change this landscape. In 2025, Google's market share in search engines fell to 89.6%, the first time in a decade it dropped below 90%. Although searches via AI and chatbots only represented 8% of the total in June of last year, the traffic generated by these platforms grew by 721% during 2025. A recent report from Semrush, a digital marketing firm, reveals that AI platforms now account for 45 billion monthly sessions globally, with ChatGPT capturing 89% of searches conducted through artificial intelligence. As OpenAI, led by Sam Altman, establishes itself as a leader in the AI sector, Google has responded by integrating AI features into its search engine. The common goal is clear: to replace traditional listings of web pages with structured information that streamlines the user experience. To keep up, Google launched AI Overviews in May 2024 and AI Mode in March 2025, incorporating conversational search into its engine. This strategy has paid off: in the third quarter of 2025, Alphabet reported a year-on-year revenue growth of 14.5% in search, driven by these new services. However, Google's own AI poses a challenge to its advertising model. Revenues from Google Network, which includes AdSense and AdMob, fell from $31.3 billion to $30.4 billion in 2024, accumulating an 8% decline from its historical peak two years earlier. Industry experts believe that both search models will coexist: LLMs will be used for complex queries, while simple and commercial searches will continue to rely on the traditional search engine. While bias was the main concern at Google, inaccuracy is the major challenge of AI. Research published in Frontiers in Artificial Intelligence and arXiv indicates that even advanced models like GPT-4 fail 5% to 10% of the time when answering general knowledge questions. The root of this problem lies in the probabilistic nature of LLMs, which predict the most likely word instead of verifying facts. Larger models can generate incorrect answers with high confidence, a phenomenon known as "confident nonsense." Therefore, the relevance of "world models" advocated by Yann LeCun and mentioned by Enrique Dans in Business Insider Spain increases: these models aim for AI to learn from a sensory representation of the world, not just from human data. Additionally, LLMs are designed to keep users engaged, similar to dating apps, encouraging dependency. The Open Worldwide Security Project warns that this trend can be dangerous, as more people blindly trust AI responses, which can have serious consequences in areas such as health, finance, or law. (Source: businessinsider.es)