Is an AI-generated website good enough to be a business's only website?

I build these, so take the bias as read. But I think the honest answer is yes for a lot of businesses and clearly no for others, and the line between them isn't where most people put it.

Here's the short version, and then I'll argue with myself for a while.

An AI-generated site is good enough to be the only site when the site's job is to inform and reassure — hours, services, proof, a way to get in touch. It stops being good enough the moment the site has to do something: take money, hold inventory, book a slot, or make a claim somebody could sue over.

That's the whole thesis. The rest is why the usual objections aim at the wrong target.

What are you actually comparing it to?

This is where nearly every version of this argument goes wrong, mine included at first.

The instinct is to compare a generated site against a good custom site. Real designer, real discovery calls, real photography, a few weeks, a real invoice. And against that, sure, the generated one loses on several axes. Fine.

But that's not the choice in front of the businesses I'm building Cadence for. Go actually look at the web presence of the small businesses near you. What you'll find is a Facebook page treated as the homepage. A site built in 2013 that shoves you sideways on a phone. A one-page thing from a template with lorem ipsum still in the footer. A Google listing and literally nothing else.

Against that, the comparison isn't close, and it's not close in the directions that matter — mobile layout, load speed, whether the hours are visible in three seconds, whether there's a tappable phone number above the fold.

So the real question isn't "is generated as good as bespoke." It's "is generated better than the nothing they've had for six years." And I've stopped pretending that's a hard question.

Where does it genuinely fall down?

Now the other side, because I'd rather say this myself than have someone say it for me.

Photos. This is the ceiling and it's the one I can't engineer around. A generated site can write your copy, pick your palette, and build your layout. It cannot show the inside of your restaurant. And the temptation — drop in a beautiful AI-generated interior, or a stock photo of a dining room in a different country — is a lie about a physical place a customer is about to drive to. That's not a quality problem, it's an honesty problem, and it's the fastest way to make a nice-looking site actively worse than no site. Real photos are the one input the pipeline can't fake, and a business that won't send five decent phone pictures of its own premises is a business I can't fully serve.

Anything transactional. Checkout, inventory, real-time booking, memberships, accounts. This is where "generated" stops being a document and starts being software, and software has states, edge cases, refunds, and a support burden. Different job. Don't let a generation pipeline near your payment flow because it produced a nice hero section.

Regulated and high-stakes claims. Medical, legal, financial, anything where a confidently-worded sentence about outcomes is a liability. The model writes marketing copy in the register of marketing copy — assured, fluent, definitive — and in those industries that register is exactly what gets you in trouble. That needs a human who knows the rules reading every line.

Sameness at scale. If I generate sites for thirty businesses in one trade, and they're all shaped by the same pipeline, the differentiation decays. Any advantage from looking sharp erodes the more of them exist. I don't have a clever answer to this one. It's a real limit on the model and I'd rather name it than pretend the twentieth one is as distinctive as the first.

What about the input problem?

There's a failure that looks like quality but isn't, and it's the one I worry about most.

The whole Cadence approach is to build a site from a business's real Google reviews — because reviews are the customer describing the business in their own words, which is better source material than the owner's own about-page instincts. That works when there's enough of it. How much is enough is a question I've written about separately, and the answer isn't a nice round number.

The dangerous part is that the output doesn't degrade visibly. A site built from a hundred detailed reviews and a site built from six thin ones look about equally polished. Same confident headline, same clean layout, same tone. The second one is just generic underneath — it says quality service and friendly staff because that's all it had to work with, and there's no visual signal telling you the tank was empty. I've called this confidence not being sufficiency and it's the thing I keep having to build guardrails against: fluent output is not evidence of good input.

So my honest answer here is conditional. Generated is good enough when there's enough real material to generate from. Below that floor, what you get is a well-dressed template, and the business would be better served by four true sentences someone actually wrote.

Isn't the website kind of secondary now anyway?

Worth saying, because it changes the stakes of the whole question.

For a lot of local businesses, the primary front door isn't the website at all. It's the Google Business Profile — that's where the hours, reviews, photos, phone number and directions get consumed, often without anyone clicking through. The website is what the listing points at, and what someone checks after they've already mostly decided.

Which reframes "only website" considerably. It doesn't have to carry the entire brand. It has to load fast on a phone, confirm the business is real and currently operating, say what they do, and make contact effortless. That's a job a generated site does well.

It also means the ranking of priorities is unintuitive: for many of these businesses, getting more reviews matters more than getting a better site. I'll say that to someone's face even though I'm the one selling the site.

The part everyone forgets: who maintains it?

A website isn't a deliverable. It's an ongoing obligation, and it decays.

Hours change. Services get added. A holiday closure needs posting. Someone new joins. The most common way a small business site becomes embarrassing isn't bad design — it's being stale, advertising a service they dropped or hours they don't keep. Wrong information is worse than plain information, because now the site is actively costing them a customer who showed up to a closed door.

Generation makes the first version nearly free. It does nothing about month fourteen. If anything the risk goes up, because a site that cost almost nothing feels like it needs nothing, and one that cost thousands at least gets looked at.

So when I'm asked whether a generated site is good enough to be the only site, I've started answering with a question: who's changing the hours at Christmas? If nobody has an answer, the design was never the problem.

Where I've actually landed

For a service business — trades, salons, clinics-that-aren't-regulated-medicine, restaurants, local operators whose site exists to inform, prove and connect — yes. It's good enough to be the only site, provided there's real review material behind it, real photos of the real place, and a person who'll keep it current.

For anyone selling online, holding inventory, taking bookings, or making claims that carry legal weight — no. Not as the only site, and not without a human in the loop who knows more than the model does.

And I should be straight about where I'm standing while saying all this. Cadence is built and live, I've written about how I'd price this work, and the pipeline can find businesses and draft a pitch — but nothing sends without me pressing the button myself. What I don't have yet is a wall of client outcomes to point at. When I do, that's its own post, and it'll say what actually happened rather than what I expect to happen.

The thing I keep coming back to: "is it good enough" is the wrong shape of question. Good enough for what is the whole thing. An awful lot of businesses need a site that loads fast, tells the truth, and gives someone a way to call. We spent fifteen years making that expensive. It isn't anymore, and I don't think that's a tragedy for anybody except the people who were charging four figures for exactly that and nothing more.

This one's auto-drafted from my notes on a schedule. If a number isn't in the notes, it doesn't show up here — I'd rather leave a blank than make something up.