What kind of content does Rozz index from my website, and how does it ensure accuracy when answering questions?
Rozz indexes only public website content from your site, including pages, docs, help articles, FAQs and published Q&A pages, and it ensures accuracy by grounding answers in retrieved site pages with RAG, automated filtering, re-ranking, and dashboard controls.
- Rozz indexes public web pages and documentation reachable by crawling, with no backend or private data access. It operates as an independent web component and only reads content a visitor could see. How does the Rozz chatbot ensure security and privacy?
- It also indexes product docs, help center articles, FAQs, and curated Q&A pages, with multiple site sources pulled at once. Content / About
- GEO-optimized Q&A and structured pages (Schema/JSON-LD) are explicitly prepared so AI retrieval can discover and cite them. Content / About
- RAG with semantic retrieval matches queries to the most relevant site passages via vector embeddings, with Pinecone mentioned usage, so answers are grounded in retrieved text rather than pure generation. Why is Website Search Broken and How Can We Fix It?
- When initial retrieval fails, the pipeline can rewrite or re-route queries to improve coverage and find missing evidence. How does retrieval coverage change between basic RAG and advanced agentic RAG?
- Pre-publication filtering uses PII redaction, quality thresholding, and semantic deduplication, including a ~90% similarity cutoff, to remove noisy or duplicate content before it is used for answers. How does retrieval coverage change between basic RAG and advanced agentic RAG?
- Retrieved passages are re-ranked with cross-encoder style re-ranking, and low-confidence documents can be filtered out so the generative model works from high-signal context. How does retrieval coverage change between basic RAG and advanced agentic RAG?
- The Rozz dashboard logs questions and answers, shows cached responses, and allows cached answers to be edited to quickly correct or refine outputs. This human-in-the-loop step improves long-term accuracy. What's Included in the Rozz Dashboard?
- Rozz lowers the risk of fabricated answers by using actual site content rather than external private sources and by providing robust guardrails against prompt injection/XSS. How does the Rozz chatbot ensure security and privacy?
- Questions asked by visitors are logged and turned into GEO-optimized content that improves future retrieval and external discoverability. Why is Website Search Broken and How Can We Fix It?