[Feature Request] Support Google Search Grounding with Structured Outputs - Gemini 3

[RobSchilderr]

Description

Summary

The native Gemini API now supports combining Google Search grounding with structured outputs in a single call (Gemini 3+), but @ai-sdk/google doesn't expose this capability yet.

Native Gemini API Support

From the official Gemini documentation:

"Gemini 3 lets you combine Structured Outputs with built-in tools, including Grounding with Google Search, URL Context, Code Execution, and File Search."

This is available for gemini-3-pro-preview and gemini-3-flash-preview models. Ultimately they are now used with latest.

Current Limitation in @ai-sdk/google

When using @ai-sdk/google, you cannot use Google Search grounding (useSearchGrounding: true) together with structured outputs. This forces a workaround with two separate API calls:

1. Phase 1: Call with grounding enabled to search and gather information (no structured output)
2. Phase 2: Call without grounding to extract structured JSON from the Phase 1 results

This adds latency, doubles API costs, and introduces complexity in matching grounding sources between phases.

Desired Behavior

Allow combining grounding with structured outputs in a single call, matching what the native Gemini API supports:

const result = await generateObject({
  model: google("gemini-3-flash-preview", {
    useSearchGrounding: true,
  }),
  schema: mySchema,
  prompt: "Research this topic and return structured data",
});

// result.object contains structured data
// result.experimental_providerMetadata?.google?.groundingMetadata contains sources

References

• Gemini structured output docs: https://ai.google.dev/gemini-api/docs/structured-output
• Gemini grounding docs: https://ai.google.dev/gemini-api/docs/grounding

AI SDK Version

• ai: 6.0.6
• @ai-sdk/google: 3.0.2

Code of Conduct

• I agree to follow this project's Code of Conduct

[aayush-kapoor]

the useSearchGrounding feature is now a provider defined tool. it's already available with the AI SDK. you can use it in the following way with structured output:

import { google } from '@ai-sdk/google';
import { generateText, Output } from 'ai';
import { z } from 'zod';
import { run } from '../lib/run';

run(async () => {
  const result = await generateText({
  model: google('gemini-3-pro-preview'),
  tools: {
    google_search: google.tools.googleSearch({}), // tool for searching
  },
  output: Output.object({
    schema: z.object({
      winner: z.string(),
      score: z.string(),
      scorers: z.array(z.string()),
    }),
  }),
  prompt: 'Search for details about the latest football World Cup tournament.',
});

  console.log(result.output);
  console.log(result.sources);
});

please confirm that this solves your use-case and i'll close this issue

[RobSchilderr]

This is using the preview model. gemini-3-pro-preview'. any idea when you will support the real version? @aayush-kapoor

[lgrammel]

Identify issue type

Identified as Feature
Confidence: Medium

Rationale

The original requested capability is already supported, but the remaining request concerns support for a production/non-preview model, which is a model-support enhancement rather than a concrete failure.

Evidence

• The description requests combining Google Search grounding with structured output in one Gemini call.
• A maintainer shows that this capability is already available through the provider-defined googleSearch tool with Output.object.
• The follow-up asks when a non-preview Gemini 3 model will be supported.