Scrolling Hacker News yesterday, the top of the feed had a post titled "ChatGPT can now build you a website" — 112 points, 124 comments. Clicking through, OpenAI has actually pushed site-building into the ChatGPT conversation: no Wix signup, no Framer tab, no v0-then-deploy dance. Describe what you want in the chat, and a site goes live.
My first reaction was: here we go again, another demo toy. But reading the wording on the official page more carefully, this one feels different.
What Sites actually is: one sentence, one website
The entry point is simple: open chatgpt.com/features/sites/, or just tell it "build me a website for X" in a ChatGPT conversation. The official pitch emphasizes that what comes out isn't a preview — it's an actual publishable, hosted site with a shareable link.
The flow runs roughly like this:
- Describe what you want in ChatGPT (e.g., "a landing page for my pottery class this Saturday")
- ChatGPT generates HTML/CSS/JS and renders it inline
- Hit publish and get a public link on a chatgpt.com subdomain
How it differs from the HTML code chunks ChatGPT has occasionally spit out: that code, you'd have to copy and host yourself. Sites wraps the whole thing in a button.
I tried it: what a chat-built site actually looks like
I asked ChatGPT to build me a single-page "AI tools directory." The prompt was:
Build a single-page website listing 10 AI coding tools. Each tool should have
name, one-line description, pricing, and a link. Use a clean dark theme,
add a search box that filters tools by name. Make it look like a real product,
not a template.
About 20 seconds later, the site was there. Clean structure, dark theme, working search box, clickable links.
A side-by-side comparison:
| Tool | Speed | Customization depth | Deploy friction | Price |
|---|---|---|---|---|
| ChatGPT Sites | 20–40 sec | Medium (code editable) | None (auto-hosted) | Included with ChatGPT sub |
| v0 | 30–60 sec | High (prompt-iterable) | Medium (Vercel needed) | Free / $20/mo |
| Framer | 1–2 hours manual | Very high | Low (built-in) | Free / from $15/mo |
| Wix | Several hours manual | Very high | None (built-in) | From $17/mo |
For someone who doesn't write code, the barrier is almost zero. For designers, v0 still gives more control. But for the 80% case of "I just need a landing page," Sites is already enough.
How low did the bar drop, and how much control did you give up
This is where I'd pump the brakes.
Building a site through conversation sounds great, but what you hand over isn't just the "code now" step. It's a whole control chain:
- Custom domains: you get a
xxx.chatgpt.comsubdomain. Want your own domain? No sign of support yet. - SEO control: title, meta, sitemap, robots.txt, Open Graph tags — Sites' handling isn't spelled out on the official page. HN commenters noted the generated pages barely get indexed by Google.
- Data migration: say the site goes live and you want to move it to WordPress or your own server. Where's the code? Can you export it? No docs on this.
- Complex interactions: search boxes and forms are fine. Stripe payments, databases, third-party APIs? No.
- Analytics: GA, Plausible, anything? No integration yet.
In other words, Sites solves "having a site" — not "having a proper site."
This is the core difference from Wix/Framer: Wix hands you a tool to build with; Sites hands you an artifact. You can use a tool long-term. An artifact, you have to regenerate.
What this means for Chinese AI vendors
Kimi, Tongyi, Wenxin, Zhipu — which one copies first?
Using the Apps SDK rollout as a reference: OpenAI shipped Apps SDK late last year, letting third-party apps run inside ChatGPT. Chinese vendors followed about 2–3 months behind. Kimi moved first (plugging in a batch of office apps), Zhipu went down its own GLM tool-calling path, Tongyi and Wenxin lagged.
For this Sites wave, my read:
- Kimi fastest: they already generate PPTs, docs, and spreadsheets. Adding "webpage generation" is a one-meeting decision for a product manager.
- Zhipu next: GLM's tool-calling is strong by domestic standards, but they'll likely ship an enterprise version before a consumer one.
- Tongyi, Wenxin: half a step behind, most probably copying once a competitor validates the feature.
The real battle isn't Sites itself — it's whether the generated sites can carry an API ecosystem. If ChatGPT Sites eventually plugs into Stripe, Calendly, and Notion, it stops being a site builder and becomes ChatGPT morphing into a new web platform. Whether Chinese vendors can keep up depends on whether they have comparable ecosystem ambitions.
Editorial take
What makes Sites genuinely interesting isn't "you can build a site in chat." It's the next step on OpenAI's path of turning ChatGPT into an OS.
Stepping back: Apps SDK lets ChatGPT run apps, Sora generates video, the Atlas browser takes over web browsing, Memory remembers you — add these up and ChatGPT isn't "a chatbot" anymore. It's becoming an app entry point plus execution environment. Sites is the latest piece of that story: not just "call apps inside ChatGPT," but "produce apps inside ChatGPT."
How is this different from what came before? When we used to say "AI lowers the creative barrier," we meant the coding barrier and the drawing barrier. Sites lowers the deployment and hosting barrier — and those are different. The former got handled years ago (think GitHub). The latter has been owned by Wix, WordPress, and Cloudflare Pages infrastructure. OpenAI just swallowed the deployment step too.
My read: for individual users and small projects, Sites will eat into part of Wix's long tail within six months. For real commercial sites, the threat is limited in the short term — until SEO and data-migration get solved, nobody's betting their primary site on it.
What I'm not sure about: the official page doesn't clarify whether Sites is bundled with all ChatGPT subscriptions or limited to Plus/Pro tiers. I also haven't confirmed whether you can edit and re-publish a site on the same subdomain after publishing — if every change generates a new link, then yes, it's just a toy.