When OpenAI debuted SearchGPT, the demonstrations suggested everything about how people look for things online would immediately change forever. But, “wow” became “wow, that’s “embarrassing” when examples of the AI search engine at work proved somewhat flawed. The challenge to Google’s reign as search engine king is still undergoing revisions.
According to a new piece in The Washington Post, SearchGPT is still wobbly on the facts. Google may not have to worry about losing its digital search throne any time soon, even as it moves at breakneck speed to implement its own AI search tools.
The issues are not hard to understand. SearchGPT is supposed to meld OpenAI’s AI models with real-time web data for faster, more accurate answers. Questions and keywords return a summary of the requested information instead of the standard Google links. It can be fast and informative. Unfortunately for OpenAI, that initial error of fact is starting to look more like the rule than the exception. As the Post pointed out, early testers saw SearchGPT claim that OpenAI CEO Sam Altman was speaking at a tech conference in the near future that he wasn’t actually scheduled to attend. That’s a hallucination just as bad as anything made up by ChatGPT.
And even if SearchGPT were guaranteed to only state the truth, that’s not much of a salve when it has no way to answer your questions. Tests shared with the Post especially disparaged SearchGPT’s abilities to help out with local information. That information has to come from somewhere. Google’s decades of refined data on vast numbers of businesses and the products and services they provide make it a snap to find most of the information people might want about the places around them. And if there is anything it doesn’t already have in its databanks, its partners and subsidiaries can likely fill it in for them. SearchGPT and OpenAI have no such database access and so the responses are either nonsense or nothing.
AI Search Parties
OpenAI’s prominence in the AI scene makes SearchGPT stand out, but the idea of mixing AI and search isn’t exclusive. Google’s aforementioned ambitions include AI Overviews and answers relayed by its Gemini AI assistant. Of course, Google’s AI search tools have faced their own struggles, with occasionally bizarre and outright dangerous advice. Not that it stopped Google from going global with AI Overviews after some revision.
OpenAI is still being cautious and has only released SearchGPT to a very limited number of users. Those eager to experience how generative AI models might transform online search have other options from well-funded startups trying other strategies. Perplexity AI is among the most prominent one. Rather than creating an AI model and then grafting it to a search engine system, Perplexity leverages existing models built by OpenAI, Anthropic, and other developers. Perplexity swerves around hallucinations with stricter guardrails and insisting on including links to back up what the AI writes, an innovation it was very early in adopting. Perplexity built its own web index while still using Google and Bing to fill in the gaps, while SearchGPT is apparently more reliant on external data. The result is a more structured approach that dodges some of the mistakes reportedly plaguing SearchGPT.
Google’s long-time approach to search doesn’t look likely to face any formidable opponents for now, even from itself. No matter who is making an AI search engine, it will need to at least come close to Google’s speed, precision, and accuracy to entice the Google-loving world. Considering the years and billions of dollars it took for Google to reach its current heights, OpenAI may need to be a little more patient or a lot more profligate in spending to do the same. And those resources will probably take more than just ChatGPT’s subscriber base to achieve
The stumbling blocks for SearchGPT highlight the broader challenges that generative AI tools face when trying to compete with established search engines. While AI tools like ChatGPT and SearchGPT can offer impressive conversational abilities, they lack the deep, structured, and real-time data needed to tackle everyday search queries at scale. SearchGPT may be on the right trail, but finding the ideal AI search engine will take some more searching.