Brave Search API

More Search API pages

API glossary

Web grounding

A B C D E F G H I J K L M N O P Q R S T U V W X Y Z

Web grounding is the practice of basing a language model’s answers on live web content retrieved at query time, so its responses reflect current, verifiable sources rather than only what it learned in training. A grounded model searches the Web for a query, pulls in the relevant pages, and generates its answer from that retrieved content—often citing the sources it used.

In short: web grounding ties a model’s answers to fresh, real web sources fetched at query time, instead of relying on its training data alone.

How web grounding works

The model is given current web content to reason over before it answers:

  1. Search: When a query needs current or factual information, the system searches the Web and gets back relevant, ranked results.
  2. Fetch content: The system retrieves content from those pages in full text, extracted passages, snippets, or rich metadata and since the model can only reason over what it is handed. More complete content generally produces better grounded answers.
  3. Inject as context: The retrieved content is placed into the model’s context window alongside the question.
  4. Generate grounded answer: The model writes its answer from that supplied content, and typically attaches source attribution so each claim can be verified.

The defining requirement is freshness and fidelity: an answer is only as good as the content retrieved, so grounding depends on getting relevant, current content into the model’s context window.

Web grounding vs. RAG and ungrounded generation

  • RAG is the general technique of retrieving documents to inform the generation of an answer; web grounding is RAG where the source is the live, public Web rather than a private corpus.
  • Model training knowledge is a fixed snapshot bounded by its knowledge cutoff; web grounding can supply information about events and changes that occurred after that cutoff.
  • Ungrounded generation answers from memory and may hallucinate; grounded generation draws on retrieved sources (which can also be cited). Ultimately, grounding reduces but does not eliminate unsupported claims.

Where web grounding matters

  • Current questions: News, prices, weather, and anything else that changes after a model has been trained need live grounding to answer correctly.
  • Trust and verification: Grounding lets a model cite real sources, which is how users and answer engines verify its claims.
  • AI answer engines and agents: Grounding is what lets an agentic assistant answer about the live Web rather than its frozen knowledge.
  • Search APIs as a common grounding layer: A search API is often the retrieval engine behind web grounding. It finds and returns web content for model grounding. Depending on the API, results may include URLs and snippets, extracted passages, or richer pre-formatted content. The Brave Search API includes an LLM Context endpoint that returns pre-extracted, relevance-ranked page content designed for grounding workflows.

Web grounding closes the gap between a model’s training data and the present moment. A search API is often the engine that does the closing.

Retrieval-augmented generation (RAG), source attribution, hallucination, knowledge cutoff, AI answer engine, agentic search, semantic search, context window, search API.