ChatGPT can now search the web for current answers, a shift that changes how the chatbot handles questions about fast-moving information. OpenAI has launched ChatGPT search for paying users, with plans to expand access later.
The feature brings web results directly into the chatbot rather than keeping search as a separate prototype. It also places OpenAI more directly in the same arena as Google, Microsoft, Perplexity and other companies building AI-powered search assistants.
What ChatGPT search changes
Until this launch, ChatGPT mostly answered from its training data. For GPT-4o, that information is current up to October 2023, and web search had been limited.
With ChatGPT search, the system can look online when a question needs newer information. OpenAI says the chatbot will automatically decide when web information is useful, especially for subjects such as sports, stocks and news of the day. Users can also start a web search manually.
The result is meant to feel less like switching tools and more like continuing a conversation. A user can ask in natural language, receive an AI-generated response, and see sources and links for further reading. The tool can also return richer multimedia results when the query calls for them.
Adam Fry, OpenAI’s product lead for search, told MIT Technology Review that the aim is to make ChatGPT a smarter assistant by expanding what it can access from the web. That access matters because a chatbot limited to older training data cannot reliably answer many questions about current events, markets or live cultural moments.
Who gets it now, and what comes later
ChatGPT search is available now to paying customers. OpenAI intends to make it available for free later, including for people who are logged out.
The company had already shown a standalone prototype of web search in July. The new release folds those capabilities into ChatGPT itself, which OpenAI describes as bringing the best of the SearchGPT experience into the chatbot.
OpenAI also plans to connect search with other ChatGPT features. Voice and Canvas, the company’s interactive platform for coding and writing, are both expected to be combined with search in the future. Those integrations are not part of the initial launch.
That roadmap matters because search is not just another answer format. It is also a building block for more capable AI agents, systems that can take more complex actions in the real world. Chirag Shah, a professor at the University of Washington who specializes in online search, told MIT Technology Review that OpenAI is trying to create a new market for more powerful and interactive AI agents.
Why this is a search fight, but not only a search fight
OpenAI is entering a crowded field. Google has AI Overviews, Microsoft has pursued AI search through Bing, Perplexity has built a natural-language search interface, and Meta is reportedly developing its own AI search engine.
The approaches are not identical. ChatGPT search and Perplexity let users interact conversationally and then receive generated answers with links. Google’s AI Overviews provide a short AI-generated summary at the top of the website, alongside a traditional list of indexed links.
That difference shapes the user experience. Traditional search often asks people to inspect links, compare pages and assemble an answer themselves. AI search tries to produce the answer first, then point outward to supporting sources.
Still, Shah said it is unlikely that these tools will easily erode Google’s dominance. The source article notes Google’s 90% market share in online search, and Shah pointed to Microsoft’s high-profile Bing effort as an example of how difficult it is to shift search behavior at scale.
For OpenAI, the bigger prize may be habitual use. If people rely on ChatGPT for current answers, follow-up questions, planning and work tasks, search becomes part of a broader assistant rather than a destination on its own.
Personalization and publisher questions
ChatGPT search can use the context of the current conversation. That means a user can ask follow-up questions on the same topic and go deeper without starting from zero each time.
ChatGPT already has a separate memory feature that can remember things about users for later chats. Search does not yet have that long-term memory capability. Fry said it should gain it in the “next couple of quarters.”
If that happens, OpenAI says search results could become more personalized. Fry gave examples such as remembering “I’m a vegetarian” or understanding a travel detail like “I’m going to New York in the next few days.”
OpenAI says it used partnerships with news organizations to develop ChatGPT’s web search. Those partners include Reuters, the Atlantic, Le Monde, the Financial Times, Axel Springer, Condé Nast and Time. Results are not limited to those publishers; they can include information from other online sources that do not block OpenAI’s search crawler.
That design creates both opportunity and risk. Suzan Verberne, a professor of natural-language processing at Leiden University, said it is positive that ChatGPT can retrieve information from reputable online sources and support follow-up questions. But she also warned that AI language models can still produce incorrect or invented answers.
MIT Technology Review reported one example from its own testing: when asked for vacation destination ideas, ChatGPT suggested “luxury European destinations” that included Japan, Dubai, the Caribbean islands, Bali, the Seychelles and Thailand. The answer cited an article from the Times, a British newspaper, that listed those places as luxury holiday options along with European destinations.
Verberne also warned that misinformation from the internet could enter ChatGPT’s responses if sources are not filtered well enough. That is a central challenge for AI search: access to the web can make answers fresher, but freshness does not automatically mean correctness.
The open question for the web
AI search also changes the economics of attention online. Benjamin Brooks, a fellow at Harvard University’s Berkman Klein Center and a former public policy lead for Stability AI, argued in an op-ed published by MIT Technology Review that chatbot-based search could reduce the visits creators need.
His concern is direct: if users receive enough information inside a chatbot, they may not click through to the publishers, writers and websites that produced the underlying material. That would make citations and links important, but possibly not sufficient.
For users, ChatGPT search promises convenience: current answers, conversational follow-ups and sources in one place. For OpenAI, it is a move toward a more capable assistant. For the wider web, it raises a harder question: how much of the internet should be experienced through an AI answer before the original source is visited?