Anyone can build a good website now. That was never the hard part.

Something genuinely shifted in the last couple of years, and it's worth saying plainly rather than pretending it didn't happen. It used to take a developer — a real person, real hours, real money — to build a professional website. A custom small-business site still runs somewhere between three and fifteen thousand dollars when a person builds it, and a clean landing page on its own is commonly two to five. Those are the going rates. For an independent veterinary practice watching every dollar, that has always been a meaningful number.

And now there's a version of the work that costs almost nothing. AI tools can take a plain-English description and generate a working website — write the code, style it, put it online — in the time it takes to drink a coffee. Not a toy. A real, fast, decent-looking site. The people who build websites for a living are not in denial about this. In a 2026 survey of web development firms, one of them put it about as plainly as you can: the build is now cheaper, but the think is not.

That distinction is the whole subject of this piece. Because here is the trap a practice owner can fall into when they hear "AI can build a website now": they conclude that the website was the thing, and the thing is now nearly free. It wasn't, and it isn't. Producing the site was never where the difficulty lived. The difficulty lives in everything the production step doesn't touch — and that part did not get cheaper. If anything it got more valuable, because now it's the only part left that's hard.

What actually got cheap

Be precise about what the tools did and didn't do, because the precision is the point.

What got cheap is production — turning a description of a website into a working website. Layout, styling, the code, getting it online with a domain pointed at it. That used to be skilled manual labor billed by the hour. Now it's close to instant. This is genuinely good news, and I'm not going to pretend otherwise to protect anyone's pricing. A practice that needs a basic, fast, presentable site can get one for a tiny fraction of what it cost three years ago, and the change that used to mean emailing a developer and waiting a week can happen while you watch.

What did not get cheap is knowing what the website should be. That's a different skill entirely — not typing, judgment. The AI will build, beautifully and instantly, exactly what you describe. It has no opinion about whether what you described is the right thing. It won't tell you the layout you asked for buries the one action you need a visitor to take. It won't notice the page is gorgeous and also invisible to Google. It won't mention that the booking button you put front and center quietly hands every conversion to a website you don't control, so you'll never know which of your visitors actually booked. It builds what you ask. The asking is the hard part, and the asking is what almost nobody can do well.

Building it was always the smallest piece

Here's the part that gets lost. Imagine the website is finished — fast, clean, professionally built, costing almost nothing. Sitting there at your domain. Now answer the question that actually matters: how does anyone find it?

A website is not a destination people stumble into. It's the bottom of a funnel. Someone in your city has a sick cat at 9pm and types something into Google. Something has to put your practice in front of that search, at that moment, ahead of the chain hospital three miles away with a bigger budget. Something has to make sure that when they click, they land on a page that loads before they give up, that says the thing they need to hear, and that makes calling you the obvious next move. That is the work. The website is just the thing waiting at the end of it.

This is the distinction worth holding onto: anyone can build a good website now. Not everyone can drive the right people to it, in the right way, and prove it worked. The first is production, and production is solved. The second is judgment under uncertainty — which searches are worth paying for and which quietly drain the budget, what a worried owner actually types versus what a marketer assumes they type, how to read whether the spend produced a booked patient or just produced clicks. None of that is in the website. None of it got automated. All of it is the part a practice is actually paying for — and now, the only part with any scarcity left in it.

I've written before about the architecture most agencies get wrong and about why most practices can't prove their Google Ads work. Both are this same idea from a different angle: the value was never in the asset, it was in knowing whether the asset is doing its job. A site you can't measure is a site you're funding on faith, no matter how cheaply it was built.

What a web developer will tell you about an AI-built site

If you ask a web developer what they think of a website built by AI, you'll often get a quick, firm answer: that it's cheap, that it's not real work, that it'll break, that you get what you pay for. Some of that is true. Some of it is fear. The useful skill — for you, trying to decide — is telling which is which. And the cleanest way I know to explain the difference comes from your world, not mine.

Think back to when affordable in-house lab analyzers arrived in veterinary practice. A machine on the counter that could run bloodwork in a few minutes, instead of drawing the sample, shipping it out, and waiting for the reference lab to call back the next day. Plenty of experienced veterinarians were skeptical, and they had a real point. The reference lab is staffed by board-certified clinical pathologists and trained technologists — genuine expertise a box on your counter doesn't have. A cheap analyzer, run by someone who skips the maintenance and doesn't know how to read a flagged result, will mislead you. The skeptics weren't being precious. They were describing a real failure mode.

But notice what actually happened. The analyzer didn't make the reference lab worthless, and it didn't make the careless clinic good. What it did was move the value. Producing the numbers got fast and cheap. Reading them — knowing that this potassium level, in this patient, with these symptoms, means act now — stayed exactly as hard as it ever was. And because the producing got cheap, the reading became the whole job. Today most practices run an in-house analyzer, and the veterinarians who learned to use one well didn't get replaced by it. They got faster, and the part of their work that was actually scarce — the judgment — became the part that mattered most.

The same thing is happening with websites, and the same two kinds of criticism are getting mixed together.

The criticism that's true: an AI can build a site that looks finished and is quietly broken — slow on a real phone, invisible to Google, with no way to tell whether anyone who lands on it ever calls. That's the careless clinic running labs on a machine nobody maintains. A developer who says "a site built fast and cheap, with no one checking it, will let you down" is right, the same way the analyzer skeptics were right. Listen to that part. It's the reason this isn't a do-it-yourself afternoon you can forget about.

The criticism that's fear: "AI sites aren't real," "that's not real work," "anything that cheap is worthless." Listen for the tell — that's not a critique of a specific flaw you could fix. It's a verdict on the tool existing at all. And it usually comes from the same honest place the analyzer skeptics came from: a person watching a tool collapse the cost of the thing they were paid for. That reaction is human. It isn't a conspiracy and it doesn't make anyone a villain. But it conflates two things — it treats "the building got cheap" as "the thing has no value," when the value didn't vanish. It moved, off the building and onto the knowing.

I'll be straight about the part that makes the skeptics partly right, because pretending otherwise would be exactly the kind of unbacked claim this whole practice exists to avoid: the AI is not always right. It will confidently produce a layout that buries your phone number, or describe a service you don't offer, or build something fast that quietly drops the tracking. An analyzer throws a bad value sometimes too. Neither fact is a reason to throw the tool out. Both are the reason you keep someone in the room who knows what right looks like — who can glance at the output and say that's wrong, run it again. The imperfection of the tool isn't the argument against it. It's the entire argument for judgment.

So when a developer talks down an AI-built site, take the half that's about a real, fixable flaw, and set aside the half that's really about the floor falling out of an old price. The good developers — like the good veterinarians who learned the analyzer — are about to be more valuable, not less, because the scarce thing now is knowing whether what got built is any good.

Why I make your site excellent, even now that building is cheap

There's a fair question hiding in all of this. If building a website got this cheap, and I build the page your ads point at — what stops me from doing exactly what this piece warns about? Cutting the corner, shipping something that looks done, and moving on?

The answer isn't that I'm more honest than the next person. You shouldn't have to take that on faith, and I wouldn't ask you to. The answer is that I can't afford to — and the why is the whole reason this works.

A traditional web developer is paid to build the site. Their job, and their income, ends at delivery. Whether the thing they handed over brings you a single patient is, from that point, your problem. That's not a knock on them — it's just where the money is in that model. The incentive stops at handoff.

Mine doesn't. I don't make my money building your site. I make it when the ads bring you patients — and ads cannot perform on a bad page. This isn't a sentiment, it's mechanics. Google scores the page your ads send people to, and that score sets what you pay per click and whether your ad shows at all. A slow page, a page that buries the one action that matters, a page Google can't make sense of — every one of those raises my costs and sinks the campaign I'm running for you. Google welds the quality of your page to the performance of my ads. I don't get to care about one and not the other. The platform won't let me.

So turn the usual worry around. The fear with any vendor is will he cut corners once he's been paid. In this model, the site isn't the thing I've been paid for — it's the foundation everything I'm paid for stands on. A corner cut on your page is a corner cut on my own campaigns. The fast load, the single clear action, the structure Google understands — those aren't favors I'm doing you. They're the conditions under which my work succeeds at all. I am, in the most literal sense, unable to profit from giving you a mediocre site.

That's the part that took the whole piece to earn. Building the page got cheap — genuinely, and for everyone. What stayed expensive is making it perform, and proving it did. So I don't compete on who can build a page; anyone can build a page now. I compete on being the person whose own livelihood depends on yours working. The website was never the product. The patients are. The site is just what the ads stand on — which is exactly why I can't let it be anything less than the best-performing page in your market.