> ## Documentation Index
> Fetch the complete documentation index at: https://docs.rippit.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Gemini

> Use agent search to ask Google's Gemini models a set of questions with Google Search on, and collect every answer, citation, and search in a Rippit table.

With Gemini selected in agent search, Rippit sends each of your questions to a Google Gemini model with Google Search grounding turned on, and saves the answer, the sources it cited, and the searches it ran. For how research tables work overall, see [How research tables work](https://docs.rippit.com/research/overview.md).

## Setup

None. Gemini models are built into Rippit. When you turn on agent search, pick one or more Gemini models from the **Search models** list.

## How it works

1. **Write the brief.** It sets the topic, intent, and audience for the questions.
2. **Rippit writes the questions.** 30 per input by default, grouped by theme. You can read and edit them before running.
3. **Each question goes to Gemini as-is**, with Google Search on and no extra instructions from Rippit.
4. **Every answer is saved as a row**, and each question is repeated (twice by default) to show how consistent the answers are.

Select up to three models in one table to compare Gemini with [ChatGPT](https://docs.rippit.com/research/chatgpt.md) or [Claude](https://docs.rippit.com/research/claude.md) on the same questions.

## What each row holds

One row per question, model, and repeat: the question and its theme, Gemini's answer, the links it cited, the searches it ran, its grounding details (which parts of the answer each source supports), a full trace, and the status of the call. Failed calls are kept as rows too.

Gemini cites sources through Google redirect links. Rippit keeps those links as Google returns them.

## Good to know

* These are API calls with search on, not a recording of the Gemini app or Google's AI Overviews. Answers can differ from what someone sees there.
* A sample runs up to three questions on all selected models, once each.
* Each new run is saved as its own study. The results show the latest, and older studies stay available in **Agent search**.

## Good fits

* Seeing how Google's AI describes your product and your competitors.
* Comparing which sources Gemini grounds its answers in against ChatGPT and Claude.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.