I recently got access to SearchGPT, which is either OpenAI’s big play to be the next Google or, you know, just something they’re tinkering with. As I’ve been using the ChatGPT-powered search engine the past week and a half, I was struck not just by the quirks of the user interface, but also what those quirks say about the difference between searching with AI and just using AI. Also: How does the experience compare to Perplexity, or Google, or even ChatGPT itself?
Before that, a quick update: Today I’m attending the Ai4 conference in Las Vegas. As you read this, I’m probably in a session to learn more about applied AI in media, taking a meeting with an innovator who’s building something interesting, or eating yet another brownie from a random snacks tray. Come say hi if you’re here! Just try not to grab me when I’ve got a mouth full of sweets.
Also, a quick reminder that The Media Copilot’s next AI Fundamentals class is happening at the beginning of September. This is a 100% live class that I recommend for any journalist, PR pro, marketer, or content creator that wants to learn how to use AI in practical creative work. You’ll also learn how to build healthy AI habits, one or two advanced prompting techniques, tons of great AI tools, and lots more. Still on the fence? Hit me up.
A MESSAGE FROM INCOGNI
Keep Your SSN Off The Dark Web
Every day, data brokers profit from your sensitive info — phone number, DOB, SSN — selling it to the highest bidder. And who’s buying it? Best case: companies target you with ads. Worst case: scammers and identity thieves.
It’s time you check out Incogni. It scrubs your personal data from the web, confronting the world’s data brokers on your behalf. And unlike other services, Incogni helps remove your sensitive information from all broker types, including those tricky People Search Sites.
Help protect yourself from identity theft, spam calls, and health insurers raising your rates. Plus, just for The Media Copilot readers: Get 55% off Incogni using code COPILOT.
How SearchGPT Compares to Perplexity, Google, and ChatGPT
When OpenAI finally unveiled SearchGPT in July, it was underwhelming. The product — its most direct shot across the bow of Google — was just a simple web page with a search box. It also wasn’t widely available: A select few users had access to it, but everybody else could only get on the waitlist. At first it looked like we might have another Sora on our hands, but many on the waitlist quickly got access, including me.
SearchGPT is a different website and service than ChatGPT. Instead of tapping into a knowledge base to architect an answer, SearchGPT applies natural language processing to a web search to summarize the results. The idea is to provide an answer that, for many searches, will be the final stop.
If this sounds a lot like how ChatGPT works already, you’re right. The white space between the two is subtle, but it’s important: ChatGPT primarily relies on its training data to create answers, sometimes supplementing that with data it finds on the web. It typically alerts the user when it’s browsing with a message, “Searched [X] sites.” SearchGPT, on the other hand, is all about the search, applying the interpretive powers of large language models to information it finds elsewhere.
In practice, a user is typically interested in good and accurate information, and doesn’t care much how it’s generated. Generally, in the case of ChatGPT, there’s the danger of hallucinations — an error of interpretation with the LLM when it flat-out makes something up. But SearchGPT’s answers are dependent on the sources it’s looking at being accurate. I’m simplifying of course, but the point is your mileage may vary depending on the vehicle used.
Similar to Google, SearchGPT serves up a search box that’s the main attraction of an otherwise spartan page. In the top right corner is your account icon, along with an option to set SearchGPT as your browser’s default search engine. It does this through a Chrome extension, but I wasn’t able to access it.
The size of the search box for SearchGPT is considerably larger than a normal search engine’s. The idea here seems to be that the large window will encourage longer, more specific queries — not just “Kelly Ripa” but “why is Kelly Ripa not on Live today” It makes sense: an LLM tends to work better when you give it more to work with. And the whole point is to answer specific questions, not throw links at you.
In a virgin SearchGPT window, the left side of the page has only the OpenAI icon. Once you’ve done a search, 3 simple icons appear beneath it:
Search: calls up a new search by overlaying the search box on your results page
Links: shows the list of sources for your search in the left rail
Photos: A grid of images from those sources in the left rail (I guess they needed a third thing?)
The Competition
For this review, I compared SearchGPT answers with results from the same query on Google, Perplexity, and ChatGPT itself.
For the simple query, “What were the main highlights of the Paris Olympics,” SearchGPT surprised me: it produced a list, not an answer. It selected 16 highlights — mostly events, but a few of the list entries were about the medal count and the opening ceremonies. Every item had a thumbnail image, but not all of them listed a source (although the complete list of sources was in the left rail).
With the same search, Google surprisingly did not return an AI Overview of highlights, instead giving me a page with widgets with videos, news stories, and Reddit posts before the actual search results, which started with the official Olympics page. Ever since the infamous glue-on-pizza incident, Google is being selective about when to serve up Overviews, but this one strikes me as a strangely missed opportunity
.
Similar to SearchGPT, Perplexity’s Pro answer also picked various events to highlight, though fewer (9 to SearchGPT’s 16, despite it being a Pro search), and with a different set of athletes. Perplexity chose Cole Hocker’s 1500m record and Ryan Crouser’s gold in the shot put, while SearchGPT was a fan of Summer McIntosh’s medal haul for swimming and Letsile Tebogo’s win in the 200m. both agreed Simone Biles and the U.S. women’s soccer team were tops, though. ChatGPT gave a much smaller list of 5 highlights, with no images and only 3 links at the bottom of the page.
Aspects of Attribution
From a UI perspective, one of the key things SearchGPT brings to the table is attribution. The links that SearchGPT uses for the answer, appear in the left sidebar, but the sidebar is easy to dismiss. This is a change from Perplexity, which puts prominent link cards up top, and you can’t click them away. That said, links also appear in the SearchGPT answer text, so content creators aren’t cheated out of attribution if the user ditches the sidebar.
When a link is from a news website, the brand’s icon appears next to the link. I initially thought this was a feature reserved for publisher partners (that is, those who have signed deals with OpenAI to license their content), but I saw that NBC and the BBC had icons too, and those aren’t official partners. SearchGPT seems to not be playing favorites with its results, which sounds like a good thing from a user perspective, but maybe not a great thing from a publisher perspective, who will certainly want more visibility in the tool.
For the search, “Should I but a robot lawnmower?” SearchGPT and Perplexity (non Pro) served up similar answers: a list of five pros and five cons of having AI cut your grass. There was a lot of overlap, but also some surprising variety (Perplexity got very specific about boundary wire). Google’s AI Overview was a very different list of factors to consider, with less detail. Most interestingly, ChatGPT produced a nearly identical list to SearchGPT of pros and cons about robot lawnmowers, but with no reference links at all.
Finally, the query, “What was Tom Cruise doing at the Olympics?” provided an opportunity for SearchGPT to perform against a query about a very specific news story. It handled it well, producing a three-paragraph answer of Cruise’s stunt, with links to media sources in the text and the sidebar. Google’s Overview was mercifully one paragraph. Perplexity’s answer was also three paragraphs, but it put prominent links to videos front and center — handy for anyone who hadn’t seen the stunt. ChatGPT’s answer was a little more basic than SearchGPT’s, but it got the job done. It also provided links, but only three.
OpenAI calls SearchGPT a prototype, but that has more to do with the decision to launch it as a standalone site instead of folding it into ChatGPT. However, OpenAI has said the best aspects of SearchGPT will eventually be merged into the main ChatGPT interface. The sticking point won’t be whether it works or not, but whether it’s a useful and seamless experience switching between the “search” and “chat” modes.
Is it better than Perplexity? Not even close. But SearchGPT wasn’t made to compete with upstarts in the AI industry — OpenAI is placing its mousetrap next to Google’s exceedingly busy search results page. On that score, SearchGPT’s clean take on internet search feels like a beautiful walled courtyard inside a sea of digital chaos.
Leave a Reply