SearchGPT explained: Details about OpenAI's search engine
SearchGPT is OpenAI's answer to Google search. It's a generative AI search engine that can ingest information in real time.
In the world of generative AI, a major challenge is knowledge cutoff.
An early concern with OpenAI's ChatGPT was the knowledge cutoff issue -- which means the large language model (LLM) only has knowledge of the world until a certain date. For example, the GPT-4o model that powers ChatGPT as of this writing has a knowledge cutoff of October 2023. With the older GPT-4 model, that cutoff was September 2021.
In contrast, a traditional search engine such as Google has no such knowledge cutoff. With Google search, the internet is constantly crawled to add the freshest possible information to the index. The challenge of knowledge cutoff in an LLM is one that multiple vendors are looking to solve with search capabilities powered by generative AI (GenAI). OpenAI is one of those vendors.
What is SearchGPT?
SearchGPT is a GenAI search engine built by OpenAI, first announced July 26, 2024.
GPT stands for generative pretrained transformer. The GPT family of LLMs including GPT-3, GPT-4 and GPT-4o are the foundational LLMs developed by OpenAI that power ChatGPT. With SearchGPT, OpenAI will bring some of the power of a traditional search engine and pair it with the capabilities of its GPT LLMs. There will be no knowledge cutoff with SearchGPT, as it mines real-time data from across the web.
SearchGPT is currently an OpenAI prototype, launched to a limited number of test users. There is also a waitlist for those who want to be part of the test group in the future. Test users include individuals as well as publishers. OpenAI is inviting publishers into the SearchGPT evaluation to ensure they are satisfied that their content is appropriately represented in search results.
The service isn't just a prototype, it's also positioned by OpenAI as being a temporary service. The goal is to test and evaluate how it works over time. When the evaluation is complete and OpenAI is satisfied with the progress, the plan is to integrate SearchGPT functionality directly into the ChatGPT interface.
As of August 2024, OpenAI has not explicitly stated when the SearchGPT service will be generally available or integrated into the primary ChatGPT experience for users.
SearchGPT features
SearchGPT will offer users many features to extend the capabilities of ChatGPT. Key features include the following:
- Conversational AI. As SearchGPT uses the same LLM as ChatGPT, the service provides a natural language interface for asking questions.
- Follow-up questions. The service will let users ask follow-up questions, making the experience more interactive and similar to a human conversation.
- Up-to-date responses. There is no longer an arbitrary knowledge cutoff date. SearchGPT can ingest real-time information from the web to provide users with up-to-date information.
- Summarization. In contrast with a traditional search engine that just provides a series of links to web addresses where more information can be found, SearchGPT provides concise summaries of the information.
- Attribution. OpenAI is including clear attributions and source links enabling users to verify the information and explore further details on the original websites.
- Visual results. A feature of SearchGPT is the delivery of visual results in the form of pictures and video to answer user queries.
OpenAI's answer to Google Search
Google has long dominated the search space. It's a position that OpenAI is looking to challenge with the SearchGPT prototype.
Answers, not links
A long-standing issue with Google search and traditional search engines in general is they are functionally indexes or pointers to other sources of information. A Google search doesn't necessarily provide an answer to a question, but it does point to where the answer can be found.
Google does provide an AI Overviews capability (formerly known as Search Generative Experience or SGE) that provides AI-generated summaries and snippets for quick answers. However, it primarily relies on presenting a list of links to third-party websites.
The goal with the SearchGPT prototype is to provide answers to user questions. Instead of just providing links, it will provide a summarization of the source, giving the user the information they need.
Contextual continuity
A traditional Google Search query is a point-in-time query where there is no carryover from one query to the next. With SearchGPT the goal is to provide the ability to maintain context across multiple queries.
Search accuracy
Traditional Google Search focuses on keyword matching, which can sometimes require users to sift through multiple pages to find relevant information. There are also concerns with traditional search engines over the practice of search engine optimization, which uses various to techniques to optimize search results. SearchGPT will combine real-time information with an LLM so more contextually relevant information will be found.
Ad-free experience
SearchGPT offers an ad-free interface, creating a cleaner and more user-friendly experience compared to Google, which displays advertisements in search results.
AI-powered search engine comparison
Here is a chart comparing the AI-powered search engines available today.
Search engine | Platform integration | Publisher collaboration | Ads | Cost |
SearchGPT (OpenAI) | Standalone prototype | Strong emphasis | Ad free | Free (prototype stage) |
Google SGE | Built on Google's infrastructure | SEO practices, content partnerships | Includes ads | Free |
Microsoft Bing AI/Copilot | Built on Microsoft's infrastructure | SEO practices, content partnerships | Includes ads | Free |
Perplexity AI | Standalone | Basic source attribution | Ad free | Free; $20/month for premium |
You.com | AI assistant with various modes | Basic source attribution | Ad free | Free; premium tiers available |
Brave Search | Independent search index | Basic source attribution | Ad free | Free |
Sean Michael Kerner is an IT consultant, technology enthusiast and tinkerer. He has pulled Token Ring, configured NetWare and been known to compile his own Linux kernel. He consults with industry and media organizations on technology issues.