Persuading Gemini to link out to the web prototype (tomcritchlow.com)

Google's latest AI search efforts have come under scrutiny for reducing the open web to a collection of citations rather than providing users with rich, engaging links. A proposal has emerged highlighting the need for a renewed approach to AI-driven searches, arguing that current models like AI Overviews and AI Mode limit user engagement by providing dead-end answers without meaningful outbound links. The issue stems from the traditional concept of "grounding" an LLM—where the model generates responses first and then seeks sources to back them up—leading to less user-friendly and interactive experiences. The proposed solution encourages leveraging the web's vast resources more effectively by creating a custom system prompt within Google’s Gemini framework, which includes curated links that not only enhance the answers given but also actively drive users to explore further. This approach emphasizes task completion and the importance of maintaining a vibrant information ecosystem, suggesting that LLMs should not just answer queries but actively connect users to high-quality sources. By ensuring links are contextually relevant and compelling, this new model aims to enhance both user experience and the open web's vitality, urging Google to rethink its strategy in the evolving AI landscape.