How ChatGPT Accesses the Web and Its Search Ambitions

Quick answer

ChatGPT now uses web browsing to answer questions with current information, overcoming its original knowledge cut-off. This capability, initially a plugin and now a core feature for paid users, allows the model to retrieve data from the internet. OpenAI is also reportedly developing its own dedicated search engine, aiming to provide direct AI-generated answers rather than a list of links.

From Limited Knowledge to Live Web Access

When ChatGPT first launched in November 2022, its knowledge was limited to the data it was trained on, with a specific cut-off date. This meant it could not provide information about events or developments that occurred after that date. Users often encountered responses like, 'My knowledge cut-off is early 2023, so I cannot provide you with any information past that time.' This limitation restricted its utility for current events or rapidly changing topics.

To address this, OpenAI introduced 'Browse with Bing' as a plugin for ChatGPT Plus subscribers in March 2023. This feature allowed the AI to query Microsoft's Bing search engine and incorporate real-time web results into its responses. The capability was later integrated directly into ChatGPT as a standard feature for Plus and Enterprise users, removing the need for a separate plugin activation. This change marked a significant shift, transforming ChatGPT from a static knowledge base into a tool that could access and synthesize live information from the internet.

How ChatGPT Browsing Works Today

When a user asks ChatGPT a question that requires current information, the system can now decide to perform a web search. It does this by formulating a query, sending it to a search engine like Bing, and then processing the retrieved results. Unlike a traditional search engine that presents a list of links, ChatGPT then synthesises the information from those links into a direct answer.

This process allows ChatGPT to answer questions about recent news, current prices, or newly published research. The model reads the content from multiple sources and attempts to combine it into a coherent, conversational response. This approach differs from typical search, where users manually click through various websites to find their answers. The system aims to provide a single, comprehensive answer rather than a collection of potential sources. However, it also means users rely on ChatGPT's interpretation and synthesis of the information, rather than seeing the raw data directly.

OpenAI's Reported Search Ambitions

Beyond integrating web browsing, reports suggest OpenAI is developing its own dedicated search product. Bloomberg reported in May 2024 that OpenAI has been working on a web search product, potentially as a direct competitor to established search engines. The company has not confirmed these reports. The goal would be to offer a distinct search experience, moving beyond simply answering prompts within the chat interface.

This standalone search product would reportedly provide a more traditional search interface, but with AI-powered direct answers at its core. It would aim to combine the immediacy of generative AI with the breadth of web indexing. This could mean a user's search query would result in a comprehensive, AI-generated summary or answer, possibly alongside attributed sources, rather than a page of blue links. The focus would be on delivering processed information directly to the user, rather than requiring them to click through external websites.

What This Changes for the Reader

For readers, the evolution of 'ChatGPT Search' means a shift in how they find information. Instead of typing keywords into a search bar and sifting through results, they can ask questions in natural language and receive direct, synthesised answers. This can speed up information gathering for simple queries or for topics where a concise summary is sufficient. It allows for a conversational interaction that traditional search does not offer.

However, this change also introduces new considerations. The accuracy and completeness of the AI's synthesised answer depend entirely on the quality of its retrieval and generation. Users may lose the ability to easily verify sources or cross-reference information from multiple original sites if the AI does not clearly attribute its findings. This means users must evaluate the credibility of the AI's response differently than they would a list of search results. Whether an OpenAI search product will clearly label sources and differentiate between AI-generated content and original web pages remains an open question.

The shift also raises questions about information diversity. If an AI consolidates information, it might inadvertently omit nuanced viewpoints or less prominent but relevant sources. A user relying on direct answers may miss important context or alternative perspectives that a traditional search page, with its variety of links, could provide. The development points towards a future where finding information might feel more like having a conversation, but with new demands on user discernment.