Open AI’s New GPTs - How to Create & Share!

OpenAI’s new GPTs feature is rolling out a way for non-developers to build customized versions of ChatGPT—then share them via links—using a guided “GPT Builder” workflow. The core value is practical: creators can define a GPT’s purpose in plain language, set its behavior and guardrails, and optionally attach tools like web browsing, image generation, file-based knowledge, and custom API actions. That combination turns a chat model into a more specialized assistant, from creative image generators to real-world decision helpers.

The walkthrough starts with the “Create a GPT” beta on the OpenAI site, where a left-side builder and right-side live preview update together. A simple example—“comic book image Creator”—shows how natural-language prompts can generate a name, starter suggestions, and even a profile picture using DALL·E 3. The builder then tailors behavior through an instruction set: the GPT is designed to produce comic-style visuals, guide users on character design, settings, and plot visualization, and keep outputs strictly within the comic/illustrative format. The creator also chooses how the assistant should handle ambiguity (ask for clarification vs. use judgment) and sets the communication style (casual and friendly).

A key part of the configuration is capabilities. The creator can upload files as “knowledge” so the GPT can reference specific documents or datasets for that custom use case. Capabilities can include web browsing, DALL·E image generation, and code interpreter (left enabled in the example). Most notably, the “actions” section allows a GPT to call external APIs—effectively enabling agent-like workflows such as retrieving data or triggering services—provided the creator sets up authentication (e.g., API keys). This is positioned as the path to “superpower” GPTs that do more than chat.

After testing in the playground, the GPT can be saved and published with different visibility settings: only the creator, anyone with a link, or public (with a note that public listing in the GPT store is expected in the coming weeks). Sharing is immediate via a copyable URL, and the GPT appears in the creator’s Explore sidebar so it can be edited, deleted, or opened for new chats.

The transcript also demonstrates a more functional GPT, “Auto advisor,” aimed at car buying. By uploading a screenshot of a car listing, the GPT breaks down the listing and uses web search to check pricing context and reliability—though it struggles when key variables like location aren’t provided. The creator highlights that entertainment and utility both benefit from the same customization pipeline.

Finally, community-made GPTs are tested from a “share gpts” area on the creator’s Discord: “grug’s wisdom” (caveman-style replies with web search), “deal Scout” (finding a Best Buy deal via web search), and “GPT Genesis” (a GPT idea generator). The overall takeaway is that the barrier to building and distributing specialized assistants is low enough that many users will experiment immediately, even before a formal GPT store launches.