Research with Google Search grounding

A model only knows what was in its training data, which is months or years old. A deck about this quarter's market needs this quarter's numbers. Google Search grounding lets Gemini search while it writes, and tells you which pages it used.

Turning search on

const searcher = model.bindTools([{ googleSearch: {} }]);
const reply = await searcher.invoke("Research this presentation topic: …");

• googleSearch is a built-in Gemini tool. Unlike the tools in Course 1, Google runs it, not your code. You don't get a function call to handle; you get an answer written from the search results.
• The notes are the reply's text. They'll be passed to generateOutline as its research option.

Where the sources are

The reply carries metadata listing the pages used, in reply.response_metadata.groundingMetadata.groundingChunks. Each chunk looks like { web: { uri, title } }. The same page often appears several times, so your function removes duplicates.