What is grounding?

By the Variask team · Updated

Definition

Grounding is when an AI model bases its answer on specific sources, such as live search results or a set of documents, instead of relying only on what it learned in training.

Also called Grounded generation, Search grounding

With Variask: Variask lists the sources cited in every tracked answer and flags wrong facts about your brand with the pages they likely came from.

How grounding works

Before writing, the system retrieves relevant material, for example by searching the web, and places it in the model's context. The model is told to answer from that material and often to cite it. The result is usually more current and easier to check than an answer from memory alone.

Google offers grounding with Google Search for Gemini, and ChatGPT, Claude, Perplexity and Copilot ground answers in web results when they search.

Grounding and hallucinations

Grounding reduces made up facts but does not remove them. The model can still misread a source, mix two sources or lean on an outdated page that ranked well. That is why the pages an engine finds about you matter so much.

What it means for brands

  • When an answer is grounded, the pages the engine retrieves shape what it says about you, more than its training data.
  • Keep key facts current and consistent across your site and the third party pages that rank for your name.
  • Look at which sources are cited when AI talks about you; those are the pages to update or win.

Questions

Grounding:
common questions

They are closely related. Retrieval augmented generation is the general method; grounding describes the goal of tying the answer to retrieved sources.

No. Many answers come from the model's training alone, especially for general questions. Questions about current facts, prices or products are more likely to trigger a search.

Related terms

Keep exploring

How AI search works

Retrieval-augmented generation (RAG)

Retrieval-augmented generation (RAG) is a way of building AI answers where the system first retrieves relevant documents or web pages, then has the language model write the answer using them.

Read

How AI search works

AI hallucination

An AI hallucination is a confident statement from an AI model that is false or not supported by any source, such as a wrong price, a feature that does not exist or an invented quote.

Read

How AI search works

Knowledge cutoff

A knowledge cutoff is the date after which an AI model has no information from its training data. Anything that happened later is unknown to the model unless it searches the web or is given the facts.

Read

Measurement

AI citations

AI citations are the links an AI assistant shows next to its answer to point to the pages it used, such as the numbered sources in Perplexity or the source links in ChatGPT and Google AI Overviews.

Read

Measurement

Sources vs citations

Sources are all the pages an AI engine read or retrieved while writing an answer. Citations are the subset it shows to the user as links. A page can shape an answer without being cited.

Read
Full glossary