Glossary
grounding
In short: Grounding is the process by which an AI model bases its answer on retrieved, current sources rather than on training data alone.
Grounding means connecting a language model's answer to external evidence at the moment the question is asked. Without grounding, a model answers from what it absorbed during training, which has a cut-off date and may be wrong or incomplete for a small business. With grounding, the system first fetches relevant documents (web pages, a search index, a database) and instructs the model to answer from them and to cite them.
Google uses the term explicitly: the Gemini API offers "grounding with Google Search", and its response includes metadata listing the sources used. OpenAI's web search tool, Anthropic's web search tool and Perplexity's Sonar API do the same thing under different names. Google AI Overviews and AI Mode are grounded by design.
For a business, grounding is the reason crawler access and page clarity matter. A grounded answer can only recommend you if your page (or a page about you) is retrievable and says the right things. It is also why answers can change week to week: the retrieved set changes.
Seoptist queries each engine through its official API with grounding or search enabled where the provider supports it, so the answers it measures reflect what a user would get with live retrieval, and it stores the grounding metadata as citations.