We announced the new ChatGPT Search Engine back in May 2024—five months early! Our crystal ball was working overtime. But now, OpenAI’s AI-powered Search Engine is finally here, so say hello to SearchGPT, marking a new era in search engines. I’ll make sure not to jump ahead on future developments again—my bad! Let’s stick to sources that report on things after they happen.
Leaving our level of information aside and without further ado, let’s move to the subject. I consider myself an expert in SEO and quite versed in AI tools, so I was eager to see how SearchGPT would change the way I find information online. After using it for a day and testing it against ‘common’ search engines, I will share my experience with SearchGPT and explain the pros and cons I believe AI-powered search technology brings, along with the challenges it will face.
And more importantly: the challenges we will face!
What Is SearchGPT?
Launched on July 26, 2024, SearchGPT was officially introduced by OpenAI (just two months after our own prediction!). Available to a select test group that includes media publishers and other partners, SearchGPT has redefined search by accessing real-time data without limitations. OpenAI’s design means it delivers current responses on topics like breaking news, live sports, and financial updates—a notable departure from other search engines’ static models.
Like ChatGPT, SearchGPT enables natural language interactions and follow-up questions, making information-seeking conversational and fluid. This design lets users bypass keyword search methods, instead engaging in full queries and receiving comprehensive, context-rich answers.
A caveat
OpenAI says it “can provide up-to-date responses on topics such as news updates, sports events, and stock market information.”
While this is true, I don’t think it makes any difference compared to other known search engines, and certainly not against the almighty Google. But let’s continue dissecting this—I promise it will be quite juicy.
Core Features of SearchGPT
Summarized Information
Benefits: One standout feature of SearchGPT is its ability to deliver concise, structured summaries. Unlike traditional search engines that return lists of links, SearchGPT provides direct summaries of relevant content, saving users from navigating multiple sites. These summaries offer a fast overview, enhancing user efficiency.
Limitations: Despite its convenience, the summarization can lack context and, at times, misrepresent the original source’s intention. I encountered instances where SearchGPT’s summary misinterpreted details, offering answers that felt misleading. While this issue isn’t exclusive to SearchGPT, its approach to summarization sometimes risks presenting partial insights as complete facts, underscoring the need for cautious interpretation.
Source Attribution
Advantages: SearchGPT enhances transparency by providing clear source citations, allowing users to trace information to its original context. This layer of accountability promotes trust and invites users to explore the material further, ensuring better access to depth and nuance.
Drawbacks: Although the citations are helpful, they aren’t immune to flaws. At times, authoritative sources may be prioritized over smaller, perhaps equally relevant sites. This could influence SEO dynamics, particularly for small-scale webmasters and bloggers, a point I’ll explore more in an upcoming article.
Ad-Free Experience
Upsides: In its current prototype stage, SearchGPT offers an ad-free environment, removing distractions and allowing for a smoother, more immersive search experience. The uncluttered interface enables users to focus directly on their queries, contributing to an overall more pleasant experience.
Downsides: Really?
(And yes, I know it won’t last, at least for free users, but I’m reviewing what’s available now. You don’t want me predicting things months in advance AGAIN, right?)
Integration with ChatGPT
Since October 31, 2024, OpenAI has embedded web search directly within ChatGPT, allowing real-time data queries with citation links. This feature blends the conversational flow of ChatGPT with real-time search capabilities, supporting a seamless experience that lets users ask follow-up questions and explore topics further in one place.
Notable Recent Developments
Enhanced Search Capabilities
ChatGPT’s search functionalities have improved significantly, with faster responses and inline citations that make verification easier.
Accessibility Improvements
Initially, SearchGPT access was limited to ChatGPT Plus users and a waitlist, but OpenAI has since extended access to all users.
Publisher Collaborations
OpenAI’s partnerships with reputable news organizations, like Reuters and the Associated Press, ensure reliable content sourcing, fostering collaboration between AI and media.
Planned Updates
OpenAI has plans to refine SearchGPT further, focusing on user feedback and exploring areas like local content and e-commerce.
The “Search Chats” Feature in ChatGPT
Along with the new search capabilities, OpenAI introduced a “search chats” feature within ChatGPT (it was about time!). This feature allows me to search through my chat history to find old conversations. For example, if I had a previous conversation about branding, I can use the search function to locate it quickly. This is especially helpful when I want to revisit a topic without scrolling through all my past chats.
The search chats feature is available on both desktop and mobile apps. A magnifying glass icon in the interface lets me search for keywords in my previous chats. This addition enhances the usability of ChatGPT, making it easier to manage and reference past conversations.
How to Use ChatGPT Search Engine
ChatGPT can now search the web more effectively than before. It can provide fast, timely answers with links to relevant web sources. When I ask a question, ChatGPT decides whether to search the web based on the content of my query.
I can also choose to search manually by clicking the web search icon. When I do, the globe icon transforms into a button with a blue background, as shown in the image below.
A curious UI design detail that caught my attention: once you click the button, it also changes the border-radius of the text area, but it only happened when I zoomed in. In normal mode, it keeps the same radius of 24px.
Another detail, this time on the UX and accessibility side: ChatGPT and SearchGPT use Radix, a component library with amazing accessibility features. Despite this, and as simple as the page is, ChatGPT fails almost every existing accessibility standard with flying colors. I mean, it’s black and white. A form. A list. A few monochrome icons. And you’re using a great accessibility library. And you failed on everything! Even doing it on purpose is difficult!
Does SearchGPT Provide Better Answers Than Google?
In my opinion, this was the feature where SearchGPT needed to shine—and it did not disappoint. Here’s why: Getting useful answers on a traditional search engine often requires multiple searches, digging through links, and skimming paid results. With ChatGPT Search, I can ask a question in a natural, conversational way, and ChatGPT provides an answer that considers the full context of our conversation. Even better, follow-up questions let me explore the topic further, with ChatGPT refining its responses as our conversation continues.
For example, if I ask about the weather forecast for a specific location, ChatGPT provides both the current weather and the upcoming forecast. If I then ask for dinner recommendations in that area, it understands the context and suggests relevant options. This conversational approach saves time and makes the search experience more personalized and less frustrating.
If you ask for my opinion on whether SearchGPT surpasses Google in result quality, I’d say yes—at least for certain types of answers.
AI Search Engine with Sources? Yes.
Now, a word of caution: SearchGPT is not the first AI chatbot to offer sources. Perplexity.ai did it first, and even ChatGPT, though in a very rudimentary (and often misleading) way.
Let me be clear: ChatGPT’s habit of inventing sources and URLs that never existed has been one of my main critiques. It was highly unethical, and I still find it hard to believe it was allowed.
However, it seems they’ve stepped up, and chats now include links to real sources, such as news articles and blog posts. When I receive an answer, I can click the “Sources” button below the response to open a sidebar with references. This feature helps me learn more about the topic and verify information by accessing the original content. I think this also benefits content creators by directing users to their websites—a major complaint content creators have had with Google, and a justified one. As I mentioned, I’ll cover this in detail in an upcoming article, once I have some ongoing research results.
In any case, having direct access to sources improves transparency and allows users to explore different perspectives. Whether they’re reading about current events, researching a topic for school, or satisfying personal curiosity, the source links add valuable information.
How Does SearchGPT Work?
The search model behind ChatGPT Search is a fine-tuned version of GPT-4o . OpenAI used novel synthetic data generation techniques called Model Distillation to train the model further. ChatGPT Search leverages third-party search providers and content provided directly by partners to provide the information users are looking for.
Model Distillation: A Simplified Explanation
I’ll explain this in simple terms, as it will have significant implications for webmasters, programmers, and SEO experts.
In a nutshell, model distillation is a process that trains smaller, cheaper models using the outputs of larger, more advanced models. This lets the smaller models perform like the advanced ones on specific tasks, but at a lower cost.
OpenAI’s new Model Distillation suite makes this process easier by bringing all the necessary steps into one platform. Here’s how it works:
- Stored Completions: Developers can save and reuse input-output pairs from bigger models (like GPT-4o) to create datasets, making it easier to train smaller models with real data.
- Evals (Beta): Evals allows developers to test and measure how well a model is doing on specific tasks, using either saved completions or their own datasets. It’s a simple way to check model accuracy and track progress.
- Fine-tuning: Stored Completions and Evals work together with OpenAI’s fine-tuning tool. Developers can fine-tune smaller models with saved data and evaluate results, all on one platform.
This setup makes it faster and easier to develop cost-efficient models that can handle tasks like their larger counterparts.
SearchGPT Limitations (and Opportunities)
One of the biggest limitations, besides the mentioned “cons” of the model, is that, at least for now, it only works with the ChatGPT 4.0 model. That means you can’t use 01-preview , 01-mini or any other. None other than 4.0. Don’t get me wrong, I think it’s more than enough, but it’s woth mentioning.
Another aspect that disappointed me was image search. I searched for “Rome Colosseum Images.” The images provided were fine, but they came from Google’s image search results for “colosseum wallpaper rome” (text case is exact; I searched in Capital Case, but Google changed it to all lowercase when I looked for wallpapers).
What’s interesting about this? First, despite what most people in SEO and IT think, SearchGPT isn’t pulling results from Bing but from Google. Try the same search I did and see for yourself. Additionally, SearchGPT interpreted my query and looked for wallpapers instead. In my opinion, this isn’t good.
The other interesting aspect is the text. It isn’t related to the images at all; both image and text sources are completely different. So, if I simply wanted Rome Colosseum images, as I requested, the text results should probably be the most relevant (with the exception of the third one). However, if I wanted background text on the Rome Colosseum, these sources would be the last I’d choose. And the text, especially the one for Indigo Sahara, is a lesson in poor marketing copy. In short, it’s a complete disappointment; it didn’t fit any of the potential outcomes.
These issues leave plenty of room for improvement, which I believe will happen soon, given that OpenAI often releases new versions almost monthly.
Additionally, OpenAI announced plans to continue improving search, focusing on areas like shopping and travel. They also aim to leverage the reasoning capabilities of the OpenAI o1 series for deeper research. Future updates may include integrating search with advanced voice features and expanding access to free and logged-out users.
In short, the technology behind ChatGPT Search allows it to understand complex queries and provide detailed answers. As OpenAI refines the model, I expect the search experience to become more intuitive and powerful. And hopefully, they’ll make improvements to the image search!
Summing It Up
Overall, my experience with SearchGPT has been positive. It didn’t surprise me—it’s essentially a better version of perplexity.ai.
However, the combination of real-time data access, a conversational interface, and integrated search within ChatGPT provides a fresh approach to finding and interacting with information online. The ad-free experience and clear source attribution build trust and improve usability.
Considering all these factors, I wonder now, more than ever, if this poses a real threat to Google. Initially, I wasn’t convinced, but after using it for less than a day, I believe SearchGPT is a significant advancement in AI-powered search technology, and I look forward to its continued evolution and improvement. I’d like them to show at least some consideration for accessibility; what they have now is really bad sad.
Key Points from Case Study: SearchGPT: The Powerful New ChatGPT Search Engine
Don’t want to read everything? Here are the takeaways!
Launch & Features
- Launched July 26, 2024; integrated into ChatGPT.
- Real-time answers with sources, ad-free prototype.
- Summarized results, conversational follow-ups, “search chats” for history.
Strengths
- Faster, contextual results vs Google for some queries.
- Reliable media partnerships (AP, Reuters).
- Clear citations boost trust.
Limitations
- Weak image search (pulls from Google, mismatched).
- Only works with GPT-4o.
- Accessibility standards largely failed.
Future
- Plans for shopping, travel, better reasoning (o1 models).
- Voice features and wider free access expected.
Overall
- Strong step beyond Perplexity.ai.
- Serious Google competitor, but needs major fixes in UX and accessibility.