What Is Search Grounding?

Search grounding is the technique of anchoring a language model's answer in live search results: the system runs a search, feeds retrieved results into the model's context, and generates a response with citations back to the sources. It converts a closed-book model into an open-book one at answer time.

The canonical implementation: Gemini's Google Search grounding

Google's Gemini API offers "Grounding with Google Search" as a developer feature. When enabled, the model decides per query whether to search, retrieves from Google's index, and returns the answer together with groundingMetadata — the source URLs, per-claim support attributions, and suggested search chips. Google prices grounded requests separately (list pricing has been $35 per 1,000 grounded queries, with a free daily allowance of 1,500 grounded requests on the free tier), which reveals something strategically interesting: retrieval is a metered cost, so engines everywhere ration when they ground.

Why grounding decisions shape brand visibility

• Grounding trigger: only some prompts fire a search. Stable knowledge gets answered from memory; volatile topics — pricing, availability, news — get grounded. Your brand's freshness-sensitive facts live almost entirely in the grounded path.
• Source selection: grounded answers inherit the search index's judgments. If you rank and extract well in Google Search, Gemini's grounded answers tend to cite you.
• Attribution surface: grounding metadata is machine-readable citation data, which is exactly what visibility platforms parse to measure citation share across prompts.

Grounding beyond Google

OpenAI grounds ChatGPT answers through Bing plus its own index; Anthropic added web search grounding to Claude in March 2025; Perplexity is grounding-first by design. The vocabulary differs — search, browsing, retrieval — but the architecture is the same three steps: decide, retrieve, cite.

Example

Ask Gemini "what does Menra's platform do" with grounding enabled and the response carries groundingMetadata pointing at the specific pages that supported each sentence. For a brand team, that metadata is the ground truth of which URLs are doing your reputational work — and which competitor pages are doing it instead. The mechanics of the underlying pipeline are covered under retrieval-augmented generation in the glossary.