OpenAI takes on Google with new SearchGPT prototype
The AI engine will provide users with information from the web. The AI vendor is working with publishers such as Newsmax. It will incorporate some features in ChatGPT.
The generative AI vendor and ChatGPT creator revealed that it is testing a temporary prototype of new AI search features that provide users with fast and relevant sources.
SearchGPT will provide users with up-to-date information from the web and links to relevant sources, OpenAI said in a July 25 blog post introducing the prototype.
Users will be able to ask SearchGPT follow-up questions.
The best features of the temporary prototype will later be incorporated directly into ChatGPT, according to the vendor.
OpenAI is partnering with publishers including Newsmax to build the system.
The vendor said it has also launched a way for publishers to manage how they appear in search. With the prototype, OpenAI is challenging a Google-dominated search market.
Despite alternatives like Microsoft Bing or Yahoo, and although You.com and Microsoft have incorporated chat AI features into their browsers, and despite startup Perplexity providing a popular AI-powered answer engine, no vendor has been able to reach or come close to matching Google's lead in search.
Strengthening ChatGPT
For OpenAI, SearchGPT appears to be about both taking on the search market and strengthening ChatGPT.
"What they really want to do is to change their experience for the users," said Arun Chandrasekaran, an analyst with Gartner.
OpenAI wants to change the experience with summarization and more conclusive responses, Chandrasekaran said.
Although OpenAI's gambit could potentially challenge disrupt Google's dominance in search, it's also an extension of how OpenAI wants to make ChatGPT a go-to product for consumers, he added.
With OpenAI's introduction of the SearchGPT prototype, the vendor used citations from publishers such as Thomson Reuters.
It also showed graphics of users searching for tomatoes and the AI search engine citing back to sources.
The approach shows that OpenAI is working to get its models to the point where they can accurately and repeatedly source information, said William McKeon-White, an analyst with Forrester Research.
What they really want to do is to change their experience for the users.
Arun ChandrasekaranAnalyst, Gartner
OpenAI and other generative AI vendors have been criticized for failing to provide reliable sourcing information about their models' outputs.
"That's something the entire market is still working through," McKeon-White said.
Even though there's been progress, generative search from all vendors sometimes cites the wrong source, causing some complications, he added.
OpenAI is "going after most relevant, potential correlated information with someone's search," McKeon-White continued.
For example, in a graphic the vendor used in the blog post, the AI engine searches for the best time to visit Half Moon Bay, a picturesque small city south of San Francisco. This is a complicated "ask" for the engine to process, he said.
This is because a user specifies something they want to do when they want to do it and where. The system has multiple tasks to do in searching for information, retrieving and then synthesizing it to the users' objective, which is more complicated than how current systems work.
Addressing the market
Meanwhile, it's unclear how much of the search market OpenAI wants to address, McKeon-White said.
"I do believe that they want to be an overlay or an intermediary, but as for providing and curating and managing that underlying search index, I don't know how much appetite they have for that," he said.
While OpenAI plans to work with other organizations such as news publishers, it's not clear if they plan to develop the search product itself; work with its partner, Microsoft, which has been putting a lot of money into its search data index; or create a synthetic data index of the web like Google.
Creating a web data index will be difficult for many reasons, including that the information is contextual as well as prioritizing relevant sources and retrieving the right information for a complex query.
Despite Google's problem in the past few years of sometimes favoring unhelpful sources, it has maintained its leadership because of its distribution on so many devices worldwide and because consumers associate search with Google, Chandrasekaran said.
"It's just baked into people's brains," he said. "Creating that habit change is not going to be very easy."
Moreover, OpenAI will also face problems in "ensuring accuracy and reliability of AI-generated information, creating a seamless user experience and addressing data privacy concerns," Futurum Group analyst Paul Nashawaty said.
SearchGPT's success will depend on overcoming hallucinations, maintaining data quality and building a robust infrastructure, Nashawaty added.
If successful, OpenAI could gain some of Google's search users, he said.
"SearchGPT's focus on conversational search and its potential to deliver more comprehensive responses could attract users away from Google's traditional search model," he continued.
OpenAI will also have to think about whether this is a task it can handle without stretching itself too thin as a startup, Chandrasekaran said.
"You don't want to expand into too many areas where you're not able to allocate resources and monetize opportunities," he said. "It's one thing to build a product, but it's another thing to monetize that very well as a market opportunity."
Esther Ajao is a TechTarget Editorial news writer and podcast host covering artificial intelligence software and systems.
