Trying to Implement Citation in Spring-ai along with vertexAI

[ery3k]

I am creating a Chatbot in spring-ai using VertexAI and i want to add citation to the response generated, but I am having probelms getting the cited sources from the ChatResponse object. I have set the .googleSearchRetreival(TRUE) and i am able to get answers to recent events, but when I look through the MetaData for citations, its being populated with the DefaultChatGenerationMetaData which does not contain the citation sources (ie. groundingMetadata)

Any help regarding this would be helpful

[MILLERMARRU]

Spring AI's default DefaultChatGenerationMetadata doesn't surface VertexAI grounding data, but you can access it through ChatResponseMetadata using the raw metadata map that Spring AI populates from the VertexAI response.

Accessing grounding metadata

import org.springframework.ai.chat.model.ChatResponse;
import org.springframework.ai.chat.metadata.ChatResponseMetadata;
import org.springframework.ai.vertexai.gemini.VertexAiGeminiChatOptions;

@Service
public class CitationService {

    private final ChatClient chatClient;

    public CitationService(ChatClient.Builder builder) {
        this.chatClient = builder
            .defaultOptions(VertexAiGeminiChatOptions.builder()
                .model("gemini-2.0-flash-001")
                .googleSearchRetrieval(true)  // enable grounding
                .build())
            .build();
    }

    public void askWithCitations(String question) {
        ChatResponse response = chatClient.prompt()
            .user(question)
            .call()
            .chatResponse();

        // Grounding metadata lives in ChatResponseMetadata under "groundingMetadata"
        ChatResponseMetadata metadata = response.getMetadata();

        // The raw VertexAI grounding metadata object
        Object groundingRaw = metadata.get("groundingMetadata");

        if (groundingRaw != null) {
            // Cast to the VertexAI Gemini type
            // com.google.cloud.vertexai.api.GroundingMetadata
            var groundingMetadata = (com.google.cloud.vertexai.api.GroundingMetadata) groundingRaw;

            groundingMetadata.getGroundingChunksList().forEach(chunk -> {
                if (chunk.hasWeb()) {
                    System.out.println("Source: " + chunk.getWeb().getUri());
                    System.out.println("Title:  " + chunk.getWeb().getTitle());
                }
            });

            // Grounding supports also gives you the text segments and confidence scores
            groundingMetadata.getGroundingSupports().forEach(support -> {
                System.out.println("Claim: " + support.getSegment().getText());
                System.out.println("Confidence: " + support.getConfidenceScoresList());
            });
        }
    }
}

Alternative: access from generation-level metadata

The grounding data can also appear per Generation in some Spring AI versions:

response.getResults().forEach(generation -> {
    var meta = generation.getMetadata();
    // Try the raw map key
    Object grounding = meta.get("groundingMetadata");
    if (grounding != null) {
        // process same as above
    }
});

Why DefaultChatGenerationMetadata doesn't have it

Spring AI uses a generic metadata interface to stay provider-agnostic. VertexAI-specific fields like groundingMetadata, searchEntryPoint, and webSearchQueries are stored in the metadata map under string keys — they're accessible but not typed. The cast to com.google.cloud.vertexai.api.GroundingMetadata (from the google-cloud-vertexai dependency) gives you the full typed API.

Required dependency

Make sure you have the VertexAI API dependency on the classpath (it usually comes transitively with the Spring AI starter):

<dependency>
    <groupId>com.google.cloud</groupId>
    <artifactId>google-cloud-vertexai</artifactId>
</dependency>