Generating content with AI is easy. Generating useful content reliably across millions of products is not. Most automated content fails: it becomes generic, lacks differentiation, and gradually erodes trust, impacting both SEO performance and user value. In this session, Stefano will show how he and his team built and evolved AI-assisted content workflows at idealo, including what went wrong in production. From poor product data leading to irrelevant web retrieval, to LLMs making unsupported product claims and FAQs answering questions they weren’t asked. He will walk through how these problems were addressed using better data, retrieval validation, task-specific models, automated evaluation, quality gates and human QA — and share a practical framework for deciding what to automate, what to validate and where humans still matter.