• 3 min read
ChatGPT Sites turns one prompt into a live website
ChatGPT Sites lets users build and deploy small websites through Codex with a single prompt, though live data connections remain unavailable.

Image: Tom's Guide
A single sentence was enough for Tom’s Guide’s Amanda Caswell to start building a live website in ChatGPT: “Make me a website about the books I’ve read this month.” She uploaded four Word documents containing book reviews, and let OpenAI’s Codex turn them into a browsable reading journal without opening a conventional website builder.
Caswell said the first version was live in under five minutes. The appeal is not simply that ChatGPT can generate HTML and CSS; other AI tools already do that. ChatGPT Sites keeps the process inside the same workspace, from the initial prompt through revisions and deployment.
How ChatGPT Sites builds a website
OpenAI describes ChatGPT Sites as a way to create and share a live website or lightweight app from Codex. Users describe what they want to build, provide source material where relevant, review the generated result, and request changes in plain English before publishing.
For Caswell’s project, the four uploaded reviews gave Codex the content it needed. Rather than specifying page dimensions, templates, or individual sections, she supplied the material and let ChatGPT propose a visual presentation.

Recommended reading
Google Assistant starts disappearing on September 4
“Make me a website about the books I’ve read this month.”
The resulting workflow resembles a conversation with a designer. Users can ask ChatGPT to change colors, rewrite a heading, rearrange the layout, add a feature, enlarge book covers, or reduce the space used by review previews. They do not need to know the technical term for the change they want.
That iterative process also changes when decisions have to be made. Instead of planning every detail before construction begins, a user can react to the generated page and refine it based on what is visible. Caswell steered her site away from a corporate dashboard and toward a reading journal through those follow-up requests.
Hosting is included, but not everything is automatic
ChatGPT Sites includes hosting and access controls, and Codex can deploy the finished site and provide a link. Users still have to pay for the domain and hosting costs, according to the article, but the technical work of building and publishing the site is handled in the ChatGPT workflow.
The feature is designed for focused websites and lightweight apps, not large or complex software projects. It also cannot currently connect directly to live data sources. Caswell’s reading site therefore will not automatically detect a newly finished book or pull updated activity from a publisher’s or author’s website.
Updating the site requires returning to Codex, adding the new review, and deploying a revised version. OpenAI suggests using a separate automation to collect frequently changing information, but the updates still need to be reviewed before the site is refreshed.
Sharing can also depend on workspace configuration. Sites may be shared with specific people or groups, while Enterprise administrators control whether the feature is available and who can use it.
A useful shortcut for small personal sites
OpenAI recommends opening the deployed link in Chrome because its in-app browser is not currently supported. That manual check matters: the site may be generated and hosted automatically, but users are still responsible for reviewing the final result before sharing it.
The feature is a substantial simplification for small, content-driven projects such as family recipe collections, children’s artwork portfolios, or vacation journals. It is not a replacement for a full publishing platform, live-data integration, or complex application stack. Its strongest proposition is narrower and more practical: turn material someone already has into a shareable website without forcing them to learn templates, code, or deployment tools first.
AI Editor
Ava covers the rapidly evolving world of artificial intelligence, from foundational models and research labs to the real-world economics of intelligence. With a background in computational linguistics, she cuts through the hype to find out what actually works. She firmly believes that benchmarks are just marketing until reproduced in the wild.
via Tom's Guide


