Knowledge-base grounding
Definition
Knowledge-base grounding is the technique of constraining an AI model's answers to content retrieved from a company's own documentation, so responses cite what the business actually knows instead of what the model guesses.
A language model answering from its training data will produce something plausible about your refund policy whether or not it has ever seen your refund policy. Grounding fixes this by splitting the job in two: a retrieval step finds the relevant passages in your own knowledge base, and the model is instructed to answer from those passages. The pattern is often called retrieval-augmented generation (RAG).
Why it matters
Ungrounded AI in customer-facing roles invents policy. A bot that confidently states a warranty period you do not offer creates a commitment someone has to unwind. Grounding narrows the model’s raw material to content you wrote and can correct — which also means the quality of grounded answers is bounded by the quality and freshness of the knowledge base behind them.
How grounding works in Oneop
Oneop’s knowledge base is a working content system: create, categorize and search articles, mark them public or private, with per-plan caps from 10 articles on Free to 2,000 on Business. That KB content grounds AI answers in four places — the chatbot’s KB-answer block, the support copilot, AI reply drafts, and the opt-in AI autoresponder — and every one of those calls is metered against the workspace’s AI credit balance.
Two honest limits are worth knowing, because they are the limits of most grounding systems and Oneop states them plainly. First, retrieval quality depends on configuration: with a vector-search sidecar or embedding key configured, retrieval is semantic; in a stock deployment without them, it falls back to keyword matching. Second, the retrieval index refreshes on a manual sync, not continuously — after a batch of article edits, you re-sync so the AI sees the current text. “Always up to date automatically” is a claim Oneop deliberately does not make.
Grounding also composes with Oneop’s guardrails: retrieved content and user input alike pass the prompt-injection screen before the model call — see prompt injection.
Related terms: AI chatbot, AI governance. Product page: AI chatbot.
Related terms
Start on the free plan and see the real thing
No credit card. Free plan, 2 seats.
Sign-up opens when the app launches. We will email you once — no marketing list.