OpenAI’s big GPT-4o-powered search engine has finally arrived, and you can sign up for its waitlist right now. The company said it plans to integrate “the best of these [search] features” directly into ChatGPT at some vague point. As it stands, this new search engine seems set to take on Google’s utter dominance in the space and offer a counterpoint to the Mountain View tech giant’s initiative to stuff AI into Google Search.
As detailed by OpenAI, SearchGPT isn’t just your run-of-the-mill search engine. In many ways, it works, like Google AI Overviews or current Bing Search responses, but with a twist. Users type in a prompt in a ChatGPT-like interface, and the AI will generate a response with several in-line links to where it sourced its information from the web. Depending on the question, such as asking for lists of music festivals in a certain location, the AI might respond with blurbs with direct links to each site. There’s a separate tab on the left-hand sidebar where you can see and click on each link provided by the page.
SearchGPT will give you a straight response with the name and link to the source in parentheses at the end of each paragraph on other queries requiring more nuance. Users can ask follow-up questions, and the AI should respond based on the context of previous prompts. ChatGPT already had many of these capabilities and a full connection to the Internet. The difference now is a new UI that emphasizes links to sites where the AI is sourcing all this information.
The new integration is closer to what you get in Google Search now. In the current version of AI Overviews, users receive a few paragraphs of AI-generated text based on your prompt, plus a few drop-down menus to link where Gemini took its information from. You can still remove these AI overviews from Google Search if you want to return to a more pure experience. For SearchGPT, the AI is intrinsic to how it functions.
Publishers and writers have been rather ticked off by how companies like Google and OpenAI have been relatively laissez-faire with the content used for training these AI models. Most recently, PerplexityAI has been taken to task by sites like Forbes for creating AI-generated articles based on original reporting without permission. The New York Times is currently suing OpenAI for training on its content. At the same time, major news networks like the Financial Times, Axel Springer, and The Associated Press have inked multi-million dollar deals with OpenAI to let the company train on their juicy content.
The problem is that so many sites depend on clicks and ad revenue to survive, something that doesn’t happen if users read AI blurbs and don’t actually click on the link. In its announcement post, OpenAI claims it is using AI to “enhance” the search experience “by highlighting high-quality content in a conversational interface with multiple opportunities for users to engage.” This is the company’s promise to offer those “prominent” links in users’ searches.
OpenAI is sending the message that it’s doing websites a favor by presenting their information differently than Google. On Thursday, Google updated its chatbot to Gemini 1.5 Flash, adding more links at the end of longer, more elaborate responses—up to four times a longer context window, according to the company. Gizmodo tried it out, but the answers it provides to some rather basic questions still aren’t as comprehensive as Gemini Advanced. Its few links to external sources are pretty standard and appear as gray boxes outside the text.
SearchGPT is still a beta product, and things will likely change. We’ll have to wait and see whether it will give original reporters and writers more credit with all the promised links. Though if SearchGPT offers a small blurb with the (hopefully correct) answer right up front, why would most bother to click on the link, even if there’s a bounty of them?
