How GoodBarber built a RAG chatbot that respects paywalls without breaking search

GoodBarber has developed a retrieval-augmented generation (RAG) chatbot for its apps that answers user questions from published content such as articles and events. The key challenge was preventing the chatbot from leaking members-only content to non-subscribers, since standard RAG pipelines do not distinguish restricted passages from free ones during retrieval. Rather than excluding premium content from search or relying on prompt-level filters — both of which proved inadequate — the team applies access control at context-assembly time. Restricted content participates in retrieval, but non-subscribers receive only a pre-indexed free excerpt of that document along with a link to the full article. This approach keeps search quality intact while turning the chatbot into a subscription funnel rather than a paywall bypass.