Unveiling the Integration of Citations within Anthropic API

Introducing Citations, a fresh API addition allowing Claude to anchor its responses on authentic source materials.

, and Administrator

2025 September 24 . 8:16 PM

2 min read

Unveiling Referencing in the Anthropic API

Unveiling the Integration of Citations within Anthropic API

In an exciting development for the AI industry, Amazon's Bedrock platform has introduced the Citations feature in its Claude models. This new addition aims to address a specific need in AI applications: verifying the sources behind AI-generated responses.

Citations is an API feature that allows Claude to ground its answers in source documents, providing a more reliable and trustworthy response. The feature was developed by Anthropic, the company behind Claude models, although specific individual developers are not named in the available sources.

Previously, developers relied on complex prompts to instruct Claude to include source information. However, with Citations, this process is streamlined. The API chunks the source documents into sentences and passes them, along with user-provided context, to the model with the user's query.

Citations can be found in the new Claude 3.5 Sonnet and Claude 3.5 Haiku models. It's also available on the Anthropic API and Google Cloud's Vertex AI. Users can explore the documentation to start using Citations.

One of the key benefits of Citations is its ability to reduce hallucinations, a common issue in AI-generated responses. Cited text will reference source documents to minimize hallucinations. For instance, in customer support, Citations can create support systems that can answer complex queries by referencing multiple product manuals, FAQs, and support tickets, always citing the exact source of information.

In document summarization, Citations can generate concise summaries of long documents with each key point linked back to its original source. This feature is particularly useful for financial firms, as demonstrated by Endex, which uses Claude to power an Autonomous Agent. With the implementation of Citations, Endex saw a reduction in source hallucinations and formatting issues, as well as a 20% increase in references per response.

Moreover, internal evaluations show that Claude's built-in citation capabilities outperform most custom implementations, increasing recall accuracy by up to 15%. This is evident in complex Q&A, where Citations can provide detailed answers to user queries across a large corpus of documents, with each response element traced back to specific sections of relevant texts.

Thomson Reuters, which uses Claude to power their AI platform, CoCounsel, has also found the Citations functionality to be beneficial for citing and linking to primary sources.

Users can provide their own chunks for the source documents or let the API handle the process. Either way, Claude generates a response that includes precise citations based on the provided chunks and context for any claims derived from the source material.

In conclusion, the Citations feature in Claude AI represents a significant step forward in the AI industry. By ensuring the accuracy and veracity of AI-generated responses, it enhances trust and reliability in AI applications, particularly in fields like customer support, document summarization, and complex Q&A.