Home › Meta Model API lesson 2: Search grounding
Meta Model API lesson 2: Search grounding
Search grounding lets the model search the live web and answer with inline citations, with no search index or retrieval pipeline of your own.
Step by step
- Use client.responses.create (search grounding only works on the Responses API, not Chat Completions).
- Add tools=[{"type": "web_search"}].
- The model decides whether a search is needed and runs it on Meta's side.
- Read the url_citation annotations: each points at the text it supports and the source URL.
- Turn the citations into footnotes or links for your users.
Try it: examples
- Grounded question
What changed in the latest Python release? Answer in 3 bullets with sources.
Good to know
- Turn it on when accuracy about current facts matters; leave it off for pure writing tasks.
- Meta's cookbook recipe takes about 10 minutes and uses muse-spark-1.3.
- Show sources so people can check the claims.
Sources (official pages, read Nov 8, 2026)
Written from Meta's official documentation and product pages; we don't run hands-on tests. Features and limits vary by plan and country.