Grounding is the practice of anchoring an AI model's responses in verified information, such as your product catalogue, price list, allergen sheet or returns policy. A grounded assistant answers from those sources rather than from general knowledge or guesses.
Common methods include retrieval-augmented generation, where relevant passages are fetched and added to the prompt, connecting the model to live systems through an API for stock or booking data, and instructions that require the model to quote or cite its source. Some tools also check each answer against the sources afterwards.
In hospitality and retail, grounding is what makes a customer assistant trustworthy. A grounded bot can say a dish is available at this venue today because it checked the live menu, and can say "I can't find that, let me pass you to a team member" when the source is silent.
Grounding is only as good as the sources. Out-of-date pages, conflicting documents and missing policies still produce wrong answers, just with confident references attached. Test with questions your sources do not cover, and check that the assistant declines rather than invents.
Grounding is the most practical protection against hallucination and the foundation of an AI assistant you would let speak for your brand.