Anthropic introduces new Citations feature to improve Claude's answer reliability

Anthropic has announced the launch of Citations, a new API feature designed to enhance Claude by grounding its answers in source documents. This update allows Claude to provide detailed references to specific sentences and passages used in generating responses, thereby improving the verifiability and trustworthiness of its outputs.

Before Citations, developers had to rely on complex prompts to instruct Claude to include source information, which often led to inconsistent performance and required significant time for prompt engineering and testing. With the introduction of Citations, users can now add source documents directly to the context window. When querying the model, Claude automatically cites claims in its output that are derived from these sources.

Anthropic's internal evaluations indicate that Claude's built-in citation capabilities outperform most custom implementations, enhancing recall accuracy by up to 15%. The Citations feature is now available for the Claude 3.5 Sonnet and Claude 3.5 Haiku versions. It is generally accessible via the Anthropic API and Google Cloud’s Vertex AI.