GPT Search
Last reviewed
May 11, 2026
Sources
10 citations
Review status
Source-backed
Revision
v4 · 2,197 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 11, 2026
Sources
10 citations
Review status
Source-backed
Revision
v4 · 2,197 words
Add missing citations, update stale details, or suggest a clearer explanation.
| GPT Search | |
|---|---|
 | |
| Information | |
| Name | GPT Search |
| Platform | ChatGPT |
| Store | GPT Store |
| Model | GPT-4 |
| Category | Research and discovery |
| Description | A finder GPT that searches the public web for other Custom GPTs matching a topic or use case. |
| Developer | starterbuild.com |
| OpenAI URL | https://chat.openai.com/g/g-RCXRXgLOX-gpt-search |
| Internal name | CustomGPT Directory |
| Chats | 1,000 (approximate, as of early 2024) |
| Web Browsing | Yes |
| DALL-E Image Generation | Yes |
| Free | Yes |
| Available | Yes |
| Updated | 2024-01-24 |
GPT Search is a Custom GPT for ChatGPT, listed in the GPT Store under the OpenAI URL slug g-RCXRXgLOX-gpt-search. It is published by the indie maker behind starterbuild.com and was created to help users find other Custom GPTs by topic, since the official GPT Store does not expose a full-text search across every public GPT. The GPT works by issuing constrained web queries of the form site:chat.openai.com/g GPT [topic], reading the results, and returning a short list of matching GPTs with their source URLs. Internally the assistant is named CustomGPT Directory, the same name used in the original Product Hunt launch for the project, although it is listed in the store under the public title GPT Search.
The GPT was published shortly before the official GPT Store opened in January 2024, when ChatGPT users had no built-in way to browse community GPTs and instead relied on Google searches and third-party catalogs. It is part of a small wave of finder-style Custom GPTs that wrapped Bing or Google site searches in a conversational interface, and it remains a useful entry point for anyone looking for a niche GPT that is not surfaced by the official store's category leaderboards.
When OpenAI announced Custom GPTs at DevDay on 6 November 2023, builders could create and share GPTs immediately, but the marketplace itself was not ready. For roughly two months, the only way to share a GPT was to copy and paste a long chat.openai.com/g/g-XXXXXXX-name URL. This produced a wave of community-built directories trying to index every public GPT, and a smaller wave of GPTs whose entire purpose was to find other GPTs.
GPT Search belongs to the second group. According to its own Product Hunt listing, the developer (handle @hellowilly) had previously built a website directory of Custom GPTs that ranked at the top of Google for related queries before being acquired. The Custom GPT version of that directory was published as a companion product, with the developer noting that the search tool would probably become less necessary once OpenAI rolled out an official store. The GPT remained live in the GPT Store after that launch and is still listed by third-party catalogs such as the AIPRM directory.
GPT Search is a thin conversational wrapper over the standard ChatGPT web browsing tool. When a user describes the kind of Custom GPT they are looking for, the assistant turns that description into one or more Google or Bing search queries restricted to the chat.openai.com/g domain, reads the result pages, and returns three to five matching GPTs together with their direct URLs.
The system prompt makes the methodology explicit. It tells the assistant to use queries like site:chat.openai.com/g GPT [user's topic], to identify three to five suitable GPTs, and to return their names and source URLs. The prompt also instructs the assistant to ask for clarification rather than guess, to keep the tone neutral and professional, and to match user criteria as closely as possible. A line in the prompt explicitly attributes the search results to Bing, which is the engine ChatGPT's browsing tool used at launch.
In practice the assistant behaves like a niche search engine. A user can describe a domain, a workflow, or a problem, and the GPT replies with a short ranked list. Because the underlying search is just a Google or Bing site search, results are limited to public GPTs that have been indexed by those engines. Private GPTs, very recently published GPTs, and any GPTs hidden from indexing will not appear.
The GPT ships with four conversation starter chips that users can tap to start a session:
| Starter | Likely intent |
|---|---|
| Search Math GPTs | Find Custom GPTs focused on mathematics, problem solving, or tutoring. |
| Search Travel GPTs | Find Custom GPTs for trip planning, itineraries, or local guides. |
| List Coding GPTs | Find Custom GPTs for programming, debugging, code review, or specific languages. |
| Help me with Data Science | Find Custom GPTs for statistics, machine learning, notebooks, or analytics workflows. |
The starters are intentionally broad. Each one maps to a high-level Google site search, and the assistant then asks follow-up questions to narrow the topic before returning a final list.
GPT Search uses three of the standard Custom GPT capabilities: web browsing, DALL-E image generation, and the default chat interface. It does not load any external knowledge files and does not call any custom Actions. The list below summarizes what is and is not enabled.
| Capability | Status | Notes |
|---|---|---|
| Web Browsing | Enabled | Core capability; powers the site search queries on chat.openai.com/g. |
| DALL-E | Enabled | Available but not central to the GPT's purpose. |
| Code Interpreter | Not enabled | The GPT does not run code or analyze files. |
| Knowledge files | None | No uploaded documents in the knowledge base. |
| Actions | None | No third-party API calls. |
| Memory | Not used | Each session is independent. |
Because the only real tool is browsing, the quality of any answer depends almost entirely on what Bing or Google has indexed at query time and on how well the assistant interprets the user's request.
GPT Search is most useful for short discovery sessions when a user knows what kind of GPT they want but does not know its exact name. Some examples that come up repeatedly in community discussion:
The assistant is less useful when the user wants the single best GPT for a task, since it returns a small ranked list rather than a curated recommendation. It is also less useful when the user wants reviews or ratings, since those are not exposed through the underlying site search.
A short exchange that follows the system prompt's pattern looks like this:
User: I need a Custom GPT that can help me prepare for the GRE quantitative section.
GPT Search: I can run a search for GRE-focused math GPTs. Do you want a tutor that explains concepts, a practice question generator, or both?
User: Both, if possible.
GPT Search: Here are five GPTs I found on the GPT Store that match. [List of names with
chat.openai.com/g/g-...URLs and a one-line summary for each.]
The assistant rarely volunteers an opinion on which entry is best; it presents the list and lets the user click through.
GPT Search is one of several ways to discover Custom GPTs. The table below sketches the main alternatives that were available around the same period and that are still maintained.
| Tool | Type | How it works | Notes |
|---|---|---|---|
| GPT Search | Custom GPT | Conversational, runs Bing or Google site searches inside ChatGPT. | Returns three to five GPTs per query with direct links. |
| Official GPT Store explore page | Native ChatGPT | Browse curated categories and a trending list. | Limited free text search; surfaces popular GPTs first. |
| GPTStore.ai | Third-party website | Crawls public GPTs and lets users search and rank by category. | Includes ratings, review counts, and example prompts. |
| Featured GPTs, AllGPTs, AIxploria | Third-party websites | Curated lists, updated daily, with categories. | Heavier on editorial picks than full-text search. |
| Custom-GPTs-Directory on GitHub | Open source list | A public repository where builders submit GPTs via pull request. | Useful for developers, less polished as a search experience. |
Google site:chat.openai.com/g/ searches | Manual web search | Direct Google site query with no wrapper. | The raw form of what GPT Search automates. |
The main advantage of GPT Search is convenience. Users already inside ChatGPT do not have to switch tabs or learn a new website; they describe what they want in natural language and click into a candidate GPT. The main disadvantage is that the underlying search is constrained to whatever Bing has indexed, with no scoring layer, no review data, and no filtering by capability such as DALL-E, code interpreter, or browsing.
GPT Search has a small but visible footprint. The Product Hunt page for the underlying CustomGPT Directory project gathered upvotes and comments from the indie hacker community at launch, and the GPT itself accumulated around one thousand chats during its first weeks before the official GPT Store opened. The third-party AIPRM directory listing records a rating count of one, which is consistent with a tool used mostly inside ChatGPT rather than through external GPT directories that surface ratings.
Community comments tend to focus on two themes. On the positive side, the assistant is praised for its simplicity and the fact that it stays inside ChatGPT. On the negative side, users note that the search is only as good as Bing's index of the chat.openai.com/g domain and that the assistant sometimes returns dead links to GPTs that have been unpublished.
Three limitations are worth highlighting for anyone considering GPT Search for regular use.
First, it does not see private or unindexed GPTs. A GPT that has not been crawled by Bing or Google, or whose author has opted out of being shared, will not appear in any result list. This means very fresh GPTs and many enterprise GPTs are invisible to the tool.
Second, it does not surface usage data. The assistant cannot tell the user how many chats a GPT has had, whether it has reviews, or whether it appears on any leaderboard. For those signals, users still need to visit a third-party catalog or the official store.
Third, its answers depend on browsing being available in ChatGPT. If the browsing tool is degraded or rate limited, the GPT can fall back to its own training data, which is not a reliable source of current GPT Store listings.
The broader context of GPT Search starts with the OpenAI DevDay announcement on 6 November 2023. The GPT Store itself opened on 10 January 2024 for ChatGPT Plus, Team, and Enterprise subscribers, with a leaderboard organized into categories such as DALL-E, writing, research, programming, education, and lifestyle, and with a revenue-share program announced for US-based builders. GPT Search was updated shortly after the store launch (the article infobox lists 2024-01-24 as the last update date) and continued to operate as a complement to the store's official discovery surface rather than as a replacement.
The maker of GPT Search has been candid that the project was an experiment timed to a market gap. Once the GPT Store rolled out its own browsing and ranking pages, the underlying website directory was sold and the Custom GPT was left running as a public utility. That history is part of why the GPT's footprint is modest: it was never marketed aggressively after the initial Product Hunt launch.