After months of speculation about its search plans, OpenAI has introduced SearchGPT, a prototype search engine that could help the company compete with Google’s profitable business.
OpenAI claims that this new tool will help users find information more quickly and easily. It uses generative AI to gather links and respond to queries in a conversational style. SearchGPT may eventually be integrated with OpenAI’s popular ChatGPT chatbot. In addition to general web searches, the engine will utilize information from publishers who have agreed to share their data with OpenAI.
According to WIRED, Kayla Wood, a spokesperson for OpenAI, declined to provide a demo or an interview about SearchGPT. However, she confirmed that the company has given access to some unnamed partners and publishers and has made improvements to the search engine based on their feedback.
Microsoft, which invests in OpenAI, was one of the first companies to launch a generative AI search engine when it released an AI version of Bing in 2023. This version used OpenAI’s language models and has since been renamed Copilot.
Since then, other companies like Google and Perplexity have also introduced their own AI search tools. Google’s AI Overviews provide AI-generated summaries of articles, often appearing at the top of search results. OpenAI’s SearchGPT seems more like Perplexity’s approach, where a chatbot offers a list of relevant links and allows users to ask follow-up questions.
After OpenAI launched ChatGPT in November 2022, early users saw its ability to find and summarize information from the web as a possible replacement for traditional search engines. However, large language models have limitations that make chatbots less reliable as search tools. They often use training data that is outdated, and if they don’t know an answer, they might make up facts.
Microsoft’s initial efforts with Bing weren’t very successful. The AI-powered search engine often provided strange, inappropriate, or incorrect answers, leading to only a slight increase in its market share.
When Google introduced AI Overviews in May, it also faced reliability issues, like suggesting people add glue to pizza. OpenAI’s SearchGPT might use a method called retrieval-augmented generation (RAG), which is a standard approach in AI search. This method aims to reduce errors in chatbot answers by referencing trusted information, like reliable news websites, while generating its responses and providing links to the original sources.
There are also concerns about copyright violations. Perplexity, in particular, has faced criticism from publications like Wired for copying elements of original journalism and ignoring requests from some websites not to use their content. In a blog post, OpenAI emphasized its commitment to publishers, stating that “SearchGPT is designed to help users connect with publishers by prominently citing and linking to them in searches.” Several companies, including Vox Media, The Atlantic, News Corp, and the Financial Times, have signed licensing agreements with OpenAI this year.
Despite the success of ChatGPT, OpenAI is looking for new ways to make money as it invests heavily in training larger AI models. A report from The Information estimates that the company could lose $5 billion this year due to costs related to hiring more staff, launching new products, and training more advanced AI.
Getting into web search could also help OpenAI compete with Google. After initially falling behind, Google quickly created its own competitor to ChatGPT called Gemini. The company is also working to integrate this technology into its products, including office software and the Android mobile operating system.
Richard Socher, CEO of You.com, a tool that uses generative AI to find information online, says it’s very difficult to create a search tool that is clearly better than what people are used to. For this reason, You.com has shifted its focus to being a more business-oriented research tool. Socher adds that OpenAI’s move from focusing on developing general artificial intelligence to commercial projects will bring new challenges. “It’s a very different mindset,” he says.
