Grounding: tying AI answers to real sources
Grounding means basing an AI's answer on information supplied at question time (web search results, documents, a database) rather than on its training memory alone. A well-grounded answer can show where each claim comes from.
Data as of October 22, 2026
Key figures
| Level | Intermediate |
|---|
A simple definition
A language model learns from a huge body of text, then stops learning: it has a knowledge cutoff and knows nothing about current events or your private documents. When it answers “from memory”, it can produce a hallucination, a plausible but false statement. Grounding fixes this by giving the model reference texts just before it answers and asking it to rely on them. In practice, those sources can be web results, documents you upload, a database or a business tool; the model is then asked to stick to them and, ideally, to cite them.
A concrete example
You ask: “What are my local library’s opening hours during the school holidays?”
- Without grounding, the model may invent believable hours, or quote last year’s.
- With grounding, the assistant runs an AI-powered web search, reads the library’s official page and answers with the link. If no such page exists, a well-designed system says so instead of improvising.
Why it matters
Grounding is now a standard feature of major assistants and APIs. Google, for instance, offers grounding with Google Search in the Gemini API, presented as a way to reduce hallucinations, reach recent information and provide citations. In companies, grounding often relies on RAG, which retrieves useful passages from internal documents. To see which assistants offer web search, check our AI assistants comparison.
Common mistakes and myths
- “An AI connected to the internet can’t be wrong.” It can pick poor sources, misread them or blend two pieces of information. Citations exist precisely so you can check.
- “A link proves the claim.” Sometimes a link is cited while the page actually says something else. Open the source for anything important.
- “Grounding only means the web.” An internal knowledge base, a price list or a business API can serve as sources too.
Related terms: RAG, hallucination, semantic search, knowledge cutoff.
Related terms
- RAG (retrieval-augmented generation) · A technique that first retrieves relevant passages from your documents, then hands them to the model so its answer is based on them.
- Hallucination · An AI answer that sounds confident and plausible but is false or made up: a figure, a quote, a source or a reference.
- AI search · A search where an AI reads web pages for you and writes a summarized answer with its sources, instead of a list of links.
- Knowledge cutoff · The date after which an AI model received no more training data, so it knows nothing about later events.
- Semantic search · Search that finds content by meaning rather than exact words, by comparing numerical representations of the texts.
Try it
Full glossary
Frequently asked questions
What's the difference between grounding and RAG?
Grounding is the goal: basing the answer on sources. RAG is one technique for getting there, by retrieving passages from a document collection. Built-in web search in an assistant is another form of grounding.
Is a grounded answer always correct?
No. If the source is wrong, outdated or misread, the answer will be too. Grounding makes mistakes easier to spot, because you can open the cited source.
How can I tell whether an assistant grounds its answers?
The clearest signs are links or references next to sentences and an indication that a search is running. With no source shown, the answer most likely comes from the model's memory.
Data sources
Data as of October 22, 2026
Report an error
Does a figure, a date or a fact look wrong? Tell us: every report is checked.