USA Today Enters Its Gen AI Era With a Chatbot (www.wired.com)

Gannett, publisher of USA Today and 220 other titles, today launched DeeperDive, a chatbot-style search-and-answer tool that converses with readers, summarizes reporting, and recommends articles across its network. Replacing the traditional search box, DeeperDive offers suggested prompts (e.g., “How does Trump’s Fed policy affect the economy?”), returns short, sourced answers, and links to relevant pieces from USA Today’s newsroom. Built by Taboola using fine-tuned open-source models, the product is designed to only draw from “real journalism” (not opinion) and requires sentence-level citations; it will refuse to generate an answer if two sources conflict. The rollout signals how legacy publishers are adopting generative AI to stem traffic losses caused by features like Google’s AI Overview and to better monetize reader intent. Technically, DeeperDive leans on Taboola’s data footprint (claims of 600M daily readers across ~11,000 publishers) and content-licensing deals (including Amazon and Perplexity) while actively blocking scrapers. For the AI/ML community, the project highlights practical production patterns: fine-tuning open-source LMs, grounding outputs with retrieval and citation, conflict-detection logic, and potential next steps toward agentic commerce features—illustrating both an operational approach to minimizing hallucinations and the evolving business incentives shaping deployed generative systems.