“Use Retrieval Augmented Generation (RAG) to enhance the reliability of model outputs by retrieving relevant and verified information from trusted external databases during response generation. This helps mitigate the risk of hallucinations.”

The Air Canada chatbot answered a policy question from model memory rather than the live policy database. Had it been RAG-grounded, the answer would have come from the actual policy document and would have been correct. The Mata case is the same pattern: a legal research tool without access to verified case databases falls back on model memory, which invents cases that sound real.

→ Connect the LLM to authoritative, domain-specific sources: policy databases for customer service, legal databases for legal tools, clinical databases for medical contexts
→ Return source references alongside every factual claim, not just the answer
→ Set fallback behavior: if retrieval returns nothing relevant, say so — do not let the model fill the gap with memory

Ask the system a question about a policy or fact that changed recently. If it returns the old answer instead of the current one from the database, RAG grounding is not working for that query type.