ChatGPT Turns Plain-Language Briefs Into Published Web Apps
Corey Ching presents Sites as a ChatGPT plugin that turns a conversational brief into a functional web app, then lets users review, revise and publish it without leaving the interface. In the conference-site example, users give feedback in plain language or annotate the rendered page, while Sites handles implementation; Ching says built-in hosting and storage allow the finished app to be deployed directly. A companion Build Web Apps tool extends the same brief-driven approach to interactive software, illustrated with a locally running pixel-art game.
The interface turns ordinary-language feedback into a web-app workflow
Corey Ching describes Sites as a ChatGPT plugin for creating functional web apps from a conversational brief, then revising and deploying them in the same environment. The division of labor is direct: the user describes the product and responds to what appears on screen; Sites handles project setup, code generation, and—according to Ching—hosting and storage.
That combines specification, implementation, visual review, revision, and release in one surface. Ching says Sites sets up the project, writes the code, and keeps working until an initial version is ready. Once the result is ready, a user can select publish or ask ChatGPT to deploy it, without configuring a separate hosting or storage service.
From an idea to a working web app, all inside ChatGPT.
The conference-site prompt establishes the intended level of input. Ching asks for a design-conference launch site with speaker cards, a filterable agenda, and a clean, minimalist editorial style—without supplying code or a technical requirements document. The preview renders a conference-site direction with bold typography; the filterable agenda remains part of the requested brief rather than a capability visibly exercised in the source.
The important constraint is that the first output need not be final. It is an artifact the user can open, inspect, and use as the basis for a more specific review.
Revision happens against the rendered page
The revision loop is built around concrete criticism rather than a new specification. Ching wants the conference schedule to be easier to scan and the keynote to register as the main event. He can make that request in chat while Sites handles the implementation.
I can keep giving feedback in plain language while it handles the implementation behind the scenes.
Sites also includes an in-app browser with an annotation tool. Ching places a targeted comment on the schedule area itself, using the same request about scanability and keynote prominence. The user can therefore provide both the desired outcome and the relevant location without translating the request into a design-system change, component specification, or code instruction.
Ching says these comments tell the system exactly what to update. The practical model is iterative review of a functioning page: inspect the result, identify a visual or structural issue where it appears, and ask for a revision in ordinary language.
Publishing is a product claim; the conference site supplies the example
Ching says Sites includes built-in hosting and storage, allowing an app to be published immediately without a separate service. In the conference example, the source shows a “Deployed successfully” message for “FORM / FIELD 2026,” followed by the live site loaded in a browser.
That visual confirms a successful deployment in the example. The broader claim—that hosting and storage are built into Sites—comes from Ching’s description of the plugin. Together, they define the intended endpoint of the workflow: a user does not merely generate a prototype inside ChatGPT, but can move from review to a deployed web presence there as well.
Build Web Apps extends the brief-driven approach to interactive software
Ching positions Sites alongside Build Web Apps, another plugin that he says improves site quality through skills for design, implementation, and testing. The second prompt is deliberately unlike the conference launch page: it asks for a colorful early-2000s-style 2D side-scrolling game with a cartoon hero, animated environments, vibrant pixel art, responsive controls, and playful effects.
The request is dictated into ChatGPT and transcribed as a build prompt, preserving the same non-code interaction model. The resulting output is a pixel-art platformer titled “MOXIE & THE MOONBYTE,” shown running at localhost:5173.
The game broadens the range of software Ching is claiming the tools can create from a creative brief. It does not repeat the hosted-publication evidence from the conference project: the game is shown in a local browser environment, while the successful deployment and live site are shown for FORM / FIELD 2026.