Process blocked responses

When interacting with Gemini on Vertex AI, your requests (prompts) or the generated responses might be blocked by content filters. This document provides guidance on how to identify and process these blocked responses.

When is content blocked?

There are two scenarios for content filters blocking content:

  • The input prompt is blocked.
  • The model's response is blocked.

The sections that follow describe how to process each scenario.

The input prompt is blocked

If your input prompt is blocked by content filters before it is sent to the model, the API response exhibits the following characteristics:

  • The promptFeedback field is populated. This field contains information about why the prompt is blocked.
  • The candidates field isn't set.

The model's response is blocked

If the model generates a response but that response is blocked by content filters, the API response exhibits the following characteristics:

You can determine why the content was blocked by inspecting these fields in the API response.

What's next