What a generator can honestly build from a business's real public facts, what it's built to refuse to make up, and where a human still has to step in.
Short answer: a well-built generator is given a small set of a business's real, public facts — its name, trade, town, phone number if listed, star rating, review count, photo count — and instructed to build a page using only those facts, never inventing services, prices, certifications, years in business, staff names, or testimonial quotes. Anything it doesn't know becomes visible placeholder copy, not a guess dressed up as fact. That's a constraint the software has to be built with on purpose — left unconstrained, a model will happily fill gaps with plausible-sounding invented detail.
"AI-generated" gets treated as a red flag by a lot of people, often for a fair reason: a model asked to write a business a website, with no constraints, will readily invent things that sound plausible — a service the business doesn't offer, a certification it doesn't hold, a testimonial nobody actually said. That's not a hypothetical risk, it's the default behaviour of an unconstrained model asked to fill in gaps. What separates a generator worth trusting from one that isn't is whether it's been deliberately built to refuse that.
A generator working from real public data starts with a narrow, factual set of inputs pulled from a business's own listing — typically its name, trade or category, town, phone number if one is listed, star rating, review count, and how many photos it has. That's the entire factual basis for what gets written. Nothing about the business's history, its actual services beyond the category it's listed under, its pricing, or its staff is known, and none of it is guessed.
A well-built prompt explicitly forbids the model from inventing services never mentioned, prices, certifications, years in business, staff names, or testimonial quotes — the exact list of things that would otherwise be the easiest, most tempting gaps for a model to fill in convincingly. This has to be an explicit instruction, checked and enforced, not an assumption that a capable model will simply behave this way on its own; left to its own judgement, it usually won't.
Where a real fact is genuinely missing — a price list, a specific credential, an "About us" story — the honest options are to leave it out entirely or mark it clearly as something to be filled in, visibly, on the page itself. A guess dressed up as a stated fact is the one thing this kind of tool cannot do without becoming actively misleading to whoever receives the finished site.
A generated preview built this way is a strong, honest starting draft — not a finished product. A real person still needs to review it before it reaches a business owner, add the real detail the generator was never given (actual prices, real testimonials once there are some, exact opening hours), and have the actual conversation with the owner. The generation step replaces the slow, mechanical part of building a first draft; it doesn't replace judgement.
The two get conflated because, most of the time, they go together — a lot of AI-generated content genuinely is unreliable about facts, because most of it was never built with a rule against inventing them. That's a property of how the tool was built, not an inherent limit of the technology. A generator that's explicitly constrained to real, supplied facts, with everything else flagged rather than guessed, can be honest by construction rather than by luck.
It shouldn't, if it's built with that in mind — but the risk is real. Left unconstrained, a language model will often fill in gaps with plausible-sounding invented detail: a made-up service, a fabricated certification, a testimonial nobody said. A generator that avoids this has to be explicitly instructed not to invent anything beyond the real facts it was given, with anything unknown flagged as a placeholder rather than guessed.
The basics from a public listing — the business's real name, trade, town, phone number if listed, star rating, review count and photo count — is enough to build a genuinely accurate starting page. Nothing else is invented to fill the gaps; a service, price or credential that wasn't supplied stays out or becomes a clearly marked placeholder.
Treat it as a strong, honest starting draft rather than a finished product — it's built only from what was publicly known at the time, so a real business will usually want to add its own prices, exact hours, and its own words before it goes live properly.