Grounding
Grounding is the practice of tying an AI model's answer to specific retrieved sources, so the response reflects real documents rather than the model's unverified internal memory.
Grounding is what keeps an AI answer tethered to real sources instead of invention. A grounded response is built from documents the system actually retrieved, so its claims can be traced back to and cited from those sources; an ungrounded response is the model guessing from memory, which is where hallucinations come from.
For AEO, grounding is the mechanism that turns your content into a citation. When an engine grounds an answer in your page, it both uses your words and names you as the source. To be reliably grounded, your claims need to be clearly stated, self-contained, and backed by evidence — the credibility and extractability work — so the engine treats your passage as the safe thing to quote and attribute.
Example. Ask Google AI Overviews a factual question and it shows a short answer with source links beneath it. Those links are the grounding — the pages the answer was built from. Being one of them is the goal of every AEO tactic.
Relevant pillars
Related terms
- RAG (Retrieval-Augmented Generation)RAG is the technique behind most AI answer engines, where the model first retrieves relevant documents from the live web or an index and then generates an answer grounded in what it found.
- CitationA citation in AI search is when an answer engine credits your page as a source for its response, usually as a linked reference, making it the surviving path to your site in a zero-click answer.
- CorroborationCorroboration is when multiple independent, reputable sources agree on a claim about you, giving AI systems the confidence to treat it as fact and repeat it in answers.