Grounding: Grounding means tying an AI model's answer to specific, verifiable sources, such as documents you provide, a database, or live search results, instead of relying on what the model remembers from training.
An ungrounded answer comes from the model's internal knowledge: patterns it absorbed during training, with no source attached. A grounded answer is built from material the model was given at the moment of the request. The model reads your contract, your policy manual, or the search results and answers from those, often with citations pointing back to the exact passage.
Grounding is done several ways. You can paste the source directly into the prompt. Retrieval-augmented generation (RAG) searches a document library and feeds the relevant chunks to the model automatically. Web search tools fetch live pages. Google's Gemini offers grounding with Google Search as a named feature, and tools like NotebookLM are built entirely around answering from the sources you upload.
Grounding reduces hallucination but does not eliminate it. The model can misread a source, combine two passages incorrectly, or cite a page that does not say what it claims. The habit that makes grounding work is checking the citation: click through, find the sentence, and confirm it supports the answer. Grounded output is easier to verify, and that is the real value.
Example at work
A benefits administrator uploads the company's 80-page plan document and asks whether a new hire qualifies for coverage during a probationary period. The assistant answers "yes, after 30 days" and cites section 4.2. She opens section 4.2, confirms the language, and quotes it in her reply to the employee.
Why it matters
Most workplace questions are about your documents, your data, and current facts, none of which live in a model's training. Learning to ground your prompts, by attaching the source and asking for citations, is the single biggest upgrade in accuracy you can make without any technical setup.
Related terms
- Retrieval-augmented generation (RAG)Retrieval-augmented generation (RAG) is a technique in which an AI system first searches a set of documents for relevant passages and then gives those passages to a language model to answer from. It lets an assistant use your own, current information.
- HallucinationA hallucination is when an AI model produces information that is false or fabricated but presents it confidently, such as an invented statistic, a citation to a paper that does not exist, or a made-up product feature.
- Knowledge cutoffA knowledge cutoff is the date after which a language model's training data stops. Unless it is given live information through search or documents, the model knows nothing about events, prices, or rules that changed after that date.
- Semantic searchSemantic search finds results based on the meaning of a query rather than exact keyword matches, so a search for 'unpaid invoices' also returns documents about 'overdue bills.' It uses embeddings to compare meaning.
- Context windowThe context window is the maximum amount of text, measured in tokens, that a language model can consider at one time. It includes your prompt, any pasted documents, the conversation so far, and the model's reply.