Introducing ChatGPT Search

Functional Positioning

ChatGPT Search is OpenAI’s natively integrated web search capability within the ChatGPT UI — distinct from the previous opt-in ‘Browse with Bing’ or third-party plugins. It is now default-enabled and low-latency, functioning as a built-in search module.

Key Technical Aspects

• Tight retrieval-generation coupling: Queries are rewritten and routed to a dedicated retriever (model name undisclosed, but explicitly differentiated from the default Bing API path); retrieved results are ranked, deduplicated, and summarized before injection into the LLM context;
• Source traceability: All generated answers include highlighted hyperlinks pointing to original HTTPS web pages (with support for cached page snapshots);
• Real-time assurance: Retrieval results carry hour-level timestamps, enabling markedly faster response to time-sensitive queries (e.g., breaking news, policy updates, product launches) compared to static-knowledge models like GPT-4-turbo (training cutoff: October 2024).

Practical Implications

• Represents a paradigm-level complement to traditional RAG: eliminates the need for users to build vector databases or manage document chunking/embedding pipelines — OpenAI encapsulates end-to-end retrieval as a service-layer capability;
• Currently available only to ChatGPT Plus and Team subscribers; no public API is offered;
• Does not replace conventional RAG: private enterprise data still requires handling via the Assistants API or Enterprise RAG solutions — this feature targets publicly accessible web information only.