Back to posts
AINews

ChatGPT Sites just left limited preview: what it replaces, and what it still cannot do

OpenAI rolled ChatGPT Sites into public beta on July 9 as part of the broader ChatGPT Work launch — describe a web app in plain English, Codex builds and hosts it, you get a shareable URL. Before you drop a Lovable or v0 subscription, here are the specific gaps that matter.

ChatGPT Sites had been sitting in a narrow Business/Enterprise preview since June 2. On July 9, as part of the broader ChatGPT Work launch, OpenAI moved it into public beta and opened it to every paid tier except Free and Go. Pro, Pro Lite, Enterprise, and Edu accounts have it now; Plus and Business follow over the coming days.

The pitch is direct: describe a web app in plain English, or tag @Sites in a prompt, and Codex builds it, deploys it to OpenAI-managed hosting, and hands back a shareable URL. That puts it in the same lane as Lovable, v0, Bolt, and Replit Agent — tools plenty of builders already pay for specifically to turn a prompt into a working internal dashboard or prototype.

What is actually different from a subscription you might already have

If you are paying separately for one of those tools just to spin up small internal things — a launch tracker, a review queue, a one-off reporting dashboard — Sites is worth testing before you renew anything, because it may already be sitting inside a ChatGPT plan you are paying for anyway. That is the concrete question today: not "is Sites good," but "does it already cover the specific small tool I keep going to Lovable or v0 for."

The gaps that matter before you commit anything real to it

  • Three countries are excluded entirely. Sites is not available in the EEA, Switzerland, or the UK at launch. If you are in one of those regions, nothing changes for you today regardless of your plan tier — do not expect it to appear via a routine app update.
  • No payment-card data, no protected health information, on any Site. That rules out the two categories of internal tool most likely to justify the switch in the first place — anything touching billing or patient/health data stays on your existing tooling.
  • Cannot serve users under 13. A hard constraint if the internal tool has any chance of reaching that audience.
  • Custom domains need a DNS change, and are not available for Enterprise workspaces at launch. If your test tool needs to live at a company subdomain rather than an OpenAI-hosted URL, budget time for the DNS step, and check your workspace tier first.
  • It is explicitly beta. Terms, limits, and behavior are the ones OpenAI ships this week, not a stable contract — treat anything you build as disposable until the feature graduates out of beta.

What to actually do today

Pick one internal tool you would build in Lovable, v0, or Bolt anyway this week — something low-stakes, no payment or health data, no under-13 users — and build it in Sites with @Sites instead. Compare two things against your usual tool: how long it took to get from prompt to a working URL, and whether the output is something you would actually hand to a colleague. Decide after that single comparison, not before it. Do not touch anything that is already working in production on another platform; this is a test for the next small thing you build, not a migration plan for what you already shipped.


This shipped in the same July 9–10 window as GPT-5.6's general availability. Current tools for prompt-to-app building are tracked in /tools.

Get the next post when it ships

One email on Sunday with the new post and a short list of what shipped that week — new guides, tool updates, and a couple of links worth reading.