Deutsch
Retrieval & grounding

Grounding

Also known as: AI grounding, grounded generation

Grounding in AI means tying a language model’s answer to external, checkable sources instead of relying on training data alone. Before or while answering, the model retrieves information, for example from a web search or a company’s documents, and can cite those sources. The word has other meanings in psychology and electrical engineering; this entry covers the AI sense.

At a glance

Type
technique for AI language models
Goals according to Google
fewer hallucinations, real-time information, citations
Values
sources such as Google Search, Google Maps, own data, custom search APIs
Used in
Gemini API and Google Cloud (examples)
Docs
Gemini API: Grounding with Google Search

Last checked:

How does grounding work?

Google documents the process for “Grounding with Google Search” in the Gemini API. The model first checks whether a search would improve the answer. If so, it writes one or more search queries, runs them, processes the results and composes a response. The API returns the answer together with citation annotations that link individual passages to the sources used.

Google lists three goals: fewer hallucinations, access to information more recent than the training data, and citations that let users verify claims. Besides web search, Google Cloud offers grounding on Google Maps, on a company’s own data and on Elasticsearch, among others. Technically, grounding with retrieved documents is a form of retrieval-augmented generation (RAG).

Grounding makes answers easier to check. It does not make them correct by default. If a source is outdated or wrong, the model may repeat the error, and, as an assessment, a cited link does not guarantee that the page actually supports the sentence next to it.

Retrieval-augmented generation: retrieve first, then answerA question triggers a search in a search index or vector database. The best passages are added to the prompt together with the question. The language model writes the answer from them and points to the sources of the passages. The answer can only cite what was retrieved first.1 QUESTION2 RETRIEVAL3 CONTEXT4 ANSWERQuestionfrom a useror an appRetrievalsearch orvector search findsmatching passagesKnowledge sourcesearch index, web,vector databasePrompt with contextquestion + passages[1] passage · URL A[2] passage · URL B[3] passage · URL CLLMAnswertext withsource links[1][3]Citations point to retrieved passagesSchematic, based on Lewis et al. (2020) and Google’s grounding docs.
In retrieval-augmented generation, the system first retrieves matching passages and passes them to the language model as context. Google calls the link between model output and verifiable sources grounding. The answer can only cite what was retrieved first (an inference from the method, not a Google statement). Sources: Lewis et al. (2020): Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (arXiv 2005.11401); Google Cloud: Grounding overview (Vertex AI); Google Cloud: Grounding with Google Search (Vertex AI).

What does grounding mean for SEO and GEO?

When an AI system grounds its answer in web search, web pages become the sources of that answer. A page can only be used if the retrieval step reaches it:

  • Gemini and Google’s AI features ground on Google Search, so indexing in Google matters.
  • Claude fetches pages on behalf of users with the Claude-User agent; Perplexity uses Perplexity-User for the same purpose.
  • In Search Console, Google added a setting in 2026 that excludes a site’s content from appearing as links or grounding in Search generative AI features.

The basics are the same as in classic search. Pages need to be reachable, indexable and clear about what they say. Facts that change, such as prices, opening hours or specifications, should be current on the page, because a grounded answer passes them on as written.

Example

Someone asks an AI assistant: “What are the museum’s opening hours on the public holiday?” Without grounding, the model answers from old training data and may be wrong. With grounding, it searches, finds the museum’s current page, states the hours and links the page. If the page still shows last year’s schedule, the answer repeats exactly that.

FAQ

What does grounding mean in the context of AI? Connecting a model’s output to specific, verifiable information outside the model, usually retrieved at answer time.

What is the difference between grounding and RAG? RAG describes the architecture (retrieve, add to the prompt, generate). Grounding describes the goal of anchoring the answer in sources. Google Cloud lists RAG among its grounding approaches.

What is a grounding page? Some practitioners use the term for a factual reference page that describes a company or brand clearly for AI systems. It is not a standard defined by AI providers.

More on the searchVIU blog

Sources

  1. Grounding with Google Search – Gemini API Docsai.google.dev
  2. Grounding overview – Gemini Enterprise Agent Platform, Google Cloud Docsdocs.cloud.google.com
  3. Search generative AI features control – Search Console Helpsupport.google.com
  4. Does Anthropic crawl data from the web? (Claude-User) – Claude Help Centersupport.claude.com
  5. Perplexity Crawlers (Perplexity-User) – Perplexity Docsdocs.perplexity.ai

Last updated: