OpenAI just launched SearchGPT

For over two decades, searching the web has meant typing keywords, scanning ten blue links, and clicking back and forth until you piece together an answer.

OpenAI wants to retire that muscle memory. The company just unveiled SearchGPT, a temporary prototype designed to combine its AI models with live information from the web.

Answers first, links alongside

Instead of handing you a list of links to sift through, SearchGPT serves up direct, conversational answers sourced from real-time web data.

You can ask a question naturally, see a synthesized response, and then ask follow-up questions without resetting the prompt. The search session maintains conversational context just like a normal chat.

Every response includes named, inline attributions along with clear source links. There is also a dedicated sidebar that lays out additional web sources, letting users click through to original reporting whenever they want.

OpenAI says SearchGPT is currently open to a small group of test users and publishers. While the standalone prototype is temporary, the team plans to roll the best elements directly into ChatGPT down the road. Local info and commerce are next on the roadmap.

Playing nice with publishers

AI companies have a rocky relationship with web creators, but OpenAI is making an explicit effort to keep media outlets on board.

The company emphasized that SearchGPT's search indexing is entirely separate from training OpenAI's foundation models. Sites can choose to opt out of generative AI model training and still appear in SearchGPT search results.

Partners like The Atlantic have already jumped in. Executives there noted that if AI search becomes the main way people navigate the internet, the technology must protect journalism and attribution from day one.

Publishers will also get controls to manage how their content appears inside the interface, along with a direct feedback channel.

Why it matters

This is OpenAI taking direct aim at Google's bread and butter.

For everyday users, getting a direct, sourced summary beats scrolling through ad-heavy search results. If it works seamlessly, it solves both the hallucination problem in ChatGPT and the link-fatigue problem on traditional search engines.

We think the smartest move here is decoupling search indexing from model training. That distinction throws an olive branch to nervous media companies and lowers the immediate risk of copyright battles that could stall deployment.

Still, the fundamental tension remains unsolved. Even with inline citations and a sidebar of links, a conversational answer removes the need to click through to the original publisher. OpenAI is promising a symbiotic relationship with content creators, but whether this prototype genuinely drives traffic or just cannibalizes pageviews is the billion-dollar question.