How to implement ground search?

[Adebesin-Cell]

Hi 👋,
Has anyone successfully implemented grounded search (using Google's search) with the Gemini models, specifically gemini-2.5-pro-preview-05-06?

I'm currently using generateText like this:

const { toolCalls } = await generateText({
  model: this.model,
  messages,
  temperature: 0,
  seed: 420,
  tools: functions,
});

This works great for triggering tool calls, but I'm trying to figure out how to integrate the googleSearch grounding capability mentioned in the official docs

I see in their example they use:

const response = await ai.models.generateContent({
  model: "gemini-2.0-flash",
  contents: [
    "Who individually won the most bronze medals during the Paris Olympics in 2024?",
  ],
  config: {
    tools: [{ googleSearch: {} }],
  },
});

But it's not clear to me how to combine this with generateText and toolCalls, or if I should be using a different approach altogether.

If anyone has managed to get grounded search working alongside tool calls, or knows the correct config/setup, I’d really appreciate any pointers!

[patelvivekdev]

https://ai-sdk.dev/providers/ai-sdk-providers/google-generative-ai#search-grounding

This is the example with streamText.

import { convertToCoreMessages, streamText, UIMessage } from "ai";
import { google, GoogleGenerativeAIProviderOptions } from "@ai-sdk/google";

export async function POST(request: Request) {
  const { messages }: { messages: UIMessage[] } = await request.json();

  const result = streamText({
    model: google("gemini-2.5-flash-preview-05-20", {
      useSearchGrounding: true,
      dynamicRetrievalConfig: {
        mode: "MODE_DYNAMIC",
        dynamicThreshold: 0.3,
      },
    }),
    system: `You are a helpful assistant.
    
    Additional information:
        - Today's date is ${new Date().toISOString()}
    `,
    messages: convertToCoreMessages(messages),
    providerOptions: {
      google: {
        thinkingConfig: {
          thinkingBudget: 8000,
          includeThoughts: true,
        },
      } satisfies GoogleGenerativeAIProviderOptions,
    },
  });

  return result.toDataStreamResponse({
    sendReasoning: true,
    sendSources: true,
  });
}

[Adebesin-Cell]

Thanks for sharing! I gave it a try, but unfortunately, it didn’t work as expected. Here's the setup I used:

const model = google("gemini-2.5-flash-preview-05-20", {
  useSearchGrounding: true,
  dynamicRetrievalConfig: {
    mode: 'MODE_DYNAMIC',
    dynamicThreshold: 0,
  },
})

And then for the text generation:

const { text, providerMetadata } = await generateText({
  model,
  messages, // includes system, user, and assistant messages
  temperature: 0,
  seed: 420,
  providerOptions: {
    google: {
      thinkingConfig: {
        thinkingBudget: 0,
        includeThoughts: false,
      },
    } satisfies GoogleGenerativeAIProviderOptions,
  },
})

However, when I log providerMetadata, I get:

{
  "google": {
    "groundingMetadata": {},
    "safetyRatings": null
  }
}

It seems like groundingMetadata is always empty. Any idea what might be missing or misconfigured?

[Adebesin-Cell]

Update on this, I tested the following models:

• gemini-2.5-pro-preview-05-06
• gemini-2.5-flash-preview-05-20
• gemini-2.0-flash

Search doesn't work at all on 2.0, it doesn't even trigger my tool calls. For 05-20 and 05-06, it works intermittently: it functions for a while and then suddenly stops again.

[patelvivekdev]

For me it is working

import { google } from "@ai-sdk/google";
import type {
  GoogleGenerativeAIProviderMetadata,
  GoogleGenerativeAIProviderOptions,
} from "@ai-sdk/google";
import { generateText } from "ai";

const model = google("gemini-2.5-flash-preview-05-20", {
  useSearchGrounding: true,
  dynamicRetrievalConfig: {
    mode: "MODE_DYNAMIC",
    dynamicThreshold: 0,
  },
});

const { text, providerMetadata } = await generateText({
  model,
  messages: [
    {
      role: "user",
      content: "tell me more about vercel AI SDK and gemini?",
    },
  ],
  temperature: 0,
  seed: 420,
  providerOptions: {
    google: {
      thinkingConfig: {
        thinkingBudget: 0,
        includeThoughts: false,
      },
    } satisfies GoogleGenerativeAIProviderOptions,
  },
});

// access the grounding metadata. Casting to the provider metadata type
// is optional but provides autocomplete and type safety.
const metadata = providerMetadata?.google as
  | GoogleGenerativeAIProviderMetadata
  | undefined;
const groundingMetadata = metadata?.groundingMetadata;

console.log(text);

console.log(groundingMetadata);

[Adebesin-Cell]

Hi, thanks 🙌. It actually works now, I had to reconfigure some of my system prompts and set the threshold to 0.6.