OpenAI Announces SearchGPT, Leaving Perplexity to Die and Google to Cry
Kylie Robison, reporting for The Verge:
OpenAI is announcing its much-anticipated entry into the search market, SearchGPT, an AI-powered search engine with real-time access to information across the internet.
The search engine starts with a large textbox that asks the user “What are you looking for?” But rather than returning a plain list of links, SearchGPT tries to organize and make sense of them. In one example from OpenAI, the search engine summarizes its findings on music festivals and then presents short descriptions of the events followed by an attribution link…
SearchGPT is just a “prototype” for now. The service is powered by the GPT-4 family of models and will only be accessible to 10,000 test users at launch, OpenAI spokesperson Kayla Wood tells The Verge. Wood says that OpenAI is working with third-party partners and using direct content feeds to build its search results. The goal is to eventually integrate the search features directly into ChatGPT.
First, for consumers: This is amazing. Google Search, the most popular search engine by a long shot, has been degrading in quality for years, and it finally has competition in the form of a search product, first and foremost, made for searching, not chatting. SearchGPT’s interface, from the start, looks eerily akin to Google, albeit with more artificial intelligence sprinkled throughout the website. The main page isn’t a chatbot interface, but a giant, inviting text field. Entering a search shows a list of links at the left with an AI summary at the right. Yes, there is a follow-up text field at the bottom, but it can be ignored, and it’s out of the way — the main focus is the list of links.
The AI summaries themselves aren’t prose-heavy, unlike Google’s or Perplexity, where the links are buried behind a Show More button. They’re visual, using “visual responses” like graphs, images from the web, and other widgets, presumably provided by some selected partners, just like Google’s Knowledge Graph-powered info panels. Any search interface should focus on short blurbs and links to more information — search engines should not be text-generation machines. Most Google searches are short, about one to two words long, and they’re mostly for finding quick bits of information, such as a link to a news article or something else on the web. AI companies like Google, Anthropic, and Perplexity like to highlight complicated queries like “What are the best vacation spots in Italy?” but the volume of such specific and well-formatted questions like that is small.
Google has failed at its most basic job: showing 10 blue links related to a search query. What the world needs is not another AI chatbot-powered summarizer, but a search engine fully powered by large language models rather than archaic crawlers and PageRank. For example, I just typed into Google “PageRank” to grab the link to Wikipedia and make sure I got the name and capitalization correct — I would never type “What is PageLink” into Google because I know Google isn’t a chatbot, it’s a search engine; a librarian for the internet. Natural human instinct encourages speaking to a chatbot the way one would to a fellow human, but search engines are different, and their results should be, too. Google mastered this perfectly, but now it’s falling apart and going down the deep end of AI. Users want 10 blue links fetched by a smart AI crawler better equipped to understand language and filter cruft on the web, not AI summaries to upsell products or advertisements.
Perplexity aims to solve the issue of Google being so comically inept that it can’t even find 10 blue links by shoving a chatbot down people’s throats, which is not what anyone wants. There’s a reason both Google search summaries and Perplexity are unpopular by Google Search metrics: they’re too complicated. Neither product prioritizes links to other parts of the web; instead, they aim to steal the internet for themselves. This ridiculous practice has infuriated publishers, who allow Google access to scrape their sites not to book it with their content without attribution, but to help attract new visitors. Links shouldn’t be added to summaries, the summaries should be added to the links. That is what, in essence, separates chatbots from search engines. If OpenAI can nail attribution, it has won the search wars and Google is dead. And regardless, we can already begin planning Perplexity’s funeral now.
But that ties into my second point: What about publishers? They’ve had their content ripped off by every LLM on the face of the planet, and now another one has taken a seat at the dinner table. What is the promise that this one will actually drive, not drive away traffic? Deepa Seetharaman, for The Wall Street Journal:
OpenAI said it partnered with publishers to build the search tool. In recent months, OpenAI representatives have shown mock-ups of the feature to publishers, who have grown increasingly uneasy about the way AI could reshape their newsrooms and newsgathering amid recent declines in online traffic for many publishers.
Publishers are broadly concerned that AI-powered search tools from OpenAI or Alphabet’s Google will serve up complete answers based on news content, eliminating the need to click on an article link and starving publishers of online traffic and advertising revenue.
It isn’t clear how much traffic a product such as SearchGPT could send publishers’ way. “We expect to learn more about user behavior” in the test, an OpenAI spokeswoman said.
Clearly, OpenAI’s main way of showing it is a more moral company (it isn’t) is by making deals with publishers, like The Journal and Vox Media. That isn’t a bad strategy, but OpenAI couldn’t possibly pay every website interested in showing up in SearchGPT results. For instance, would I show up in SearchGPT, even though I’m obviously way down the priority list of must-pay publishers? I think I will since I haven’t blocked ChatGPT’s search crawler — only its training one — but with the immense self-inflicted reputational damage AI companies have done to themselves, why wouldn’t weary publishers block SearchGPT before it even launches? OpenAI has said that it will obey Robots Exclusion Protocol directives, which is good, but OpenAI needs to do a lot of work to prove to the world that it is capable of attribution.
OpenAI has not contributed much to the open web yet, but it has the potential to do so with SearchGPT. Until it does that — until it drives traffic to publishers without blatantly ripping their content off — OpenAI will continue to be known as the company that steals from hardworking people. For users fed up with Google’s erroneous search results, SearchGPT will be great. For publishers fed up with OpenAI stealing their work, SearchGPT is just another product to bemoan, even if it might actually do good to the information superhighway in the long run. Either way, Perplexity, which clearly hasn’t even pondered this dilemma in the slightest, can go to hell, and Google has work cut out for itself.