Grounding

In Salesforce's AI context (Einstein), the technique of anchoring generative AI responses in your actual CRM data and trusted knowledge sources rather than relying solely on the language model's general training data.

In Salesforce's AI context (Einstein), Grounding is the technique of anchoring generative AI responses in actual CRM data and trusted knowledge sources rather than relying solely on the language model's general training data. When a generative AI feature like Einstein Copilot answers a question, grounding feeds the model relevant context (like the customer's record, recent interactions, knowledge articles) so the response is based on real, verified information instead of general patterns the model learned during training.

Grounding is one of the most important techniques for making generative AI useful in enterprise contexts. Without grounding, models hallucinate plausible-sounding but incorrect information, which is unacceptable for customer-facing or business-critical work. With grounding, the model's responses are anchored to actual data, reducing hallucinations and improving accuracy significantly. The Einstein Trust Layer handles grounding automatically for built-in features, fetching relevant CRM context and including it in prompts to the underlying LLM. For custom generative AI work, developers can implement grounding through Prompt Builder and similar tools.