OpenAI has updated ChatGPT Sites, the feature that lets anyone describe a website or a simple app in a chat window and have it built and hosted automatically. No coding required. OpenAI says people have created more than 5 million of these sites since the feature launched three months ago.
The update adds five things. Teammates can now edit, save, and publish to a shared site together, and sites can be shared privately with specific named people instead of being made public. Deployment is roughly twice as fast as before, you can ask ChatGPT to inspect the database behind your site directly in chat, and you can now connect your own web address instead of using a generic one.
For a business, the private sharing option is the standout change. It means someone in operations or finance can build a small internal tracker, a client portal, or an event sign-up page and hand it only to the people who need it, without it ending up searchable by anyone with the link. Combined with a custom domain, a small team can now put together something that looks and feels like professional software in an afternoon, not a multi-week project with an outside developer.
The more interesting story sits one layer up. ChatGPT already has close to a billion monthly users, and a wave of separate companies built their entire business on the same basic idea: type a description, get a working app back.
Lovable has raised money at a valuation of 13.3 billion dollars while pulling in 500 million dollars a year in revenue. Replit's valuation moved from 3 billion dollars to 9 billion dollars in about six months, and the company is now chasing 1 billion dollars in yearly revenue. Wix bought a smaller player in the same space, Base44, for 80 million dollars just to get into the market.
Every one of these companies builds its product on top of AI models supplied by companies like OpenAI. That is now a real problem for them. When the company that makes the underlying model builds the same app-building experience directly into a product it already gives to hundreds of millions of people, the businesses paying to rent that model and wrap it in a nicer interface suddenly look a lot less necessary.
This has happened before in software: whenever a big platform absorbs a feature that used to be sold separately, the smaller company selling just that feature struggles to justify its price. There is a real limit worth knowing before anyone rushes to build company tools this way. Independent security research on AI-generated apps has found a large share ship with genuine weaknesses, including exposed passwords, API keys, and personal data left open to anyone who finds the site.
A tool built in an afternoon should get the same scrutiny as any new piece of software before it touches customer data or payment information. Building small internal software just got faster and cheaper for almost any business, and that is a genuine gain. But any company whose whole product is a friendly wrapper around someone else's AI model is standing on ground that can shift under it without warning.