An AI website builder that only uses facts from your clinic

Most clinics that come to us already have a website. It is usually not that the information on it is wrong — it is that the site is slow, hard to read on a phone, and expensive or awkward to change. Throwing away the accurate information along with the design is a bad trade, so the builder starts by reading what you already have.

You paste your web address at /start. We fetch up to three pages of that site, with a 1.5 MB cap and an eight second timeout, through a fetcher that is guarded against server-side request forgery so a pasted address cannot be used to reach anything private. From those pages we parse the title and meta description, any JSON-LD structured data, your logo image, your theme colour, your phone number, your email address, your postal address, your opening hours, your social links and any external booking link. That parsed set of facts is what the model is given.

If you do not have a site yet, the same builder runs from scratch. You describe the practice and the builder works from your description. There is less confirmed material to build from, so you should expect fewer sections and more editing afterwards. That is the honest trade, and it is the same constraint at work: the builder writes from what it has been told, not from what it can imagine about a clinic like yours.

This is the part that separates a clinic website builder from a general one. Ask a generic AI site builder for a medical practice page and it will happily produce a polished result containing a Dr. Sarah Chen who does not exist, a list of accepted insurers nobody checked, a starting price for a consultation nobody set, and five-star reviews from patients who were never seen. It reads well. Every sentence of it is a liability.

On a clinic website that is not a cosmetic problem. A fabricated clinician name is a claim about who is licensed to treat people at your address. A fabricated insurance list tells a patient their visit is covered when it is not, and they find out at the desk. A fabricated price is a quote you did not give. A fabricated testimonial is an invented patient — which is both a false advertising problem and, if it had been real, a disclosure problem. Whoever publishes it owns it, and that is you, not the model.

So the constraint is written into the system prompt the generator runs on, in the imperative: use only real facts from the provided content, never invent providers, prices, insurers or contact details, and omit a section rather than fabricate it. That last clause is the load-bearing one. The failure mode of a language model asked for a complete website is that it fills every slot in the template, because an empty slot looks like a mistake. Telling it that a missing section is the correct output when the facts are missing is what makes an absent team section possible.

The practical result: if your current site names three physicians, the generated site names those three. If it names none, the generated site has no team section, and you add one yourself with names you can stand behind. We would rather hand you a shorter page than a longer one you have to audit line by line.

Generation ends with a draft, and drafts need editing. Every generated site opens in a block-based visual editor. You click a block to change its text, drag blocks to reorder sections, add blocks that the generator left out and delete anything you do not want. There is no code to write and no template you are locked inside.

This is the second half of the honest-facts argument. A builder that refuses to invent your provider list is only useful if adding your real provider list afterwards is easy. The same goes for the insurers you accept, the services you have added since your last site update, and the booking link you want people to reach. The generator gets you to a complete, accurate starting point fast; the editor is where you make it yours. Walk through the whole sequence on how it works.

Some of these are deliberate design decisions and some are simply not built yet. We would rather you knew which before you started than found out after you published.