By Josh Kilen, Founder & CEO, Cascade Digital Marketing
You have probably seen the pitch by now. Type a few sentences into a tool, wait a minute, and a real website appears. Pages, photos, a contact form, the works. Forty dollars a month, no agency required. And the uncomfortable part, if you run a firm and pay someone to handle your marketing, is that the pitch mostly works. The site that comes out the other end looks fine.
So can AI build your website? Yes. I want to be straight about that, because I run a marketing company and we use AI to build sites constantly. We built our own system to do it. That is the honest starting point, and it is why I can tell you exactly where the magic stops. Including the parts that should make you skeptical of anyone, me included, who says AI simply handles your website now.
Building a website and building your website, the one that works and that you would put in front of a client without flinching, are two very different jobs. The first one got easy this year. The second one got harder to see, which is a worse problem.
The part AI is genuinely good at now
A web page is, underneath the design, a layout problem. Blocks of text, images, and buttons arranged on a screen. That happens to be the exact thing the language behind every website was invented to describe, and the current models write that language fluently, the way they write English. Point one at a blank page and it will hand you a clean, modern, presentable site faster than you can book the kickoff call.
The proof is not just mine. A consulting firm at the publication Every spent months trying to fully automate PowerPoint and burned through real money doing it, at one point 62 dollars in computing cost per slide deck. Their hard-won conclusion was to give up on PowerPoint and move to web code, because the AI was dramatically better at it. Websites already are web code. So when it comes to building sites, we are starting in the best possible case for the machine.
We are. And it still breaks in ways that are worth showing you, precisely because this is the easy case.
”It looked finished” is the most expensive phrase in web design
This spring we migrated a client off an old platform onto a faster, modern build. The AI did most of the construction, and the result looked done. It tested clean on my own machine. All the scores read where I wanted them.
Then we put it on the live internet and the page started shifting under the visitor’s cursor. Images dropping in late, shoving the text down half a second after it appeared, the kind of cheap, jumpy feeling you notice without being able to name. Our quality score for that exact problem read a perfect zero on my laptop and an ugly 0.24 to 0.26 once it was actually online.
That gap is the whole lesson. The bug was invisible in the one place most convenient to check, because on my own machine the images loaded instantly and never had a chance to misbehave. “It passed on my laptop” was a confident, cheerful lie. Now we measure that score on the live site, on a phone, the way a real person on a real network would meet it, not where it is comfortable to test.
The same week, a global styling rule we had not written carefully enough turned a whole section of footer text white, on a white background. No error message. The site built perfectly. The text was simply, invisibly gone, and the only way anyone was ever going to catch it was to look at the rendered page with human eyes. So that is what we do now, on every build: screenshot every section on mobile and desktop, then click every button and feed garbage into every form field to see what cracks, before a client ever sees it.
A confident wrong answer is the dangerous kind
The dangerous failures do not announce themselves. The tooling hit a snag with some redirects and proposed a fix. The fix was clean, well-explained, and textbook correct. For a different hosting company. It does not work on the one we use, and it would have broken links if I had trusted it.
That is the trap with a fluent machine. It is just as confident when it is wrong as when it is right, and a polished wrong answer is far more dangerous than an obvious one, because a polished answer makes you stop checking. The Every team hit the same wall in a more dramatic way: their AI declared it had reviewed and improved a client’s slide deck across eight rounds, and it was lying. It had never actually looked at the slides. It had graded its own homework against a checklist and waved it through.
An 80 percent finished website delivers a lot less than 80 percent of the value. Reviewing a glossy, near-perfect thing for the hidden 20 percent that is wrong is harder and slower than building the right thing in the first place. That is the part the 40-dollar tool cannot sell you, because the tool is the thing doing the overconfident waving-through.
The test that actually matters: does the form reach you?
For a law firm or an estate planning practice, your website has exactly one job that pays the bills. A person in a hard moment fills out the contact form, and that message lands in front of a human who can call them back. Everything else is decoration around that one event.
So we do not assume the form works because we wired it up. We send real test submissions all the way through the live system and watch them arrive on the other end, then delete the test records. It sounds almost too basic to mention. It is also the single most common thing I find broken on sites other people built, including expensive ones. A contact form that silently fails is the worst kind of marketing problem, because nothing looks wrong. The site is beautiful. The phone just patiently stops ringing, and you have no idea why.
What was hard was never the building
Walk back through everything I just described. None of the genuinely hard parts were the construction. The construction took an afternoon. What was hard was knowing what the site needed to be.
Knowing that a handful of this client’s existing page addresses had been compounding in Google for years and could not change, or the traffic would evaporate. Knowing which specific words convince someone who has been burned by a bad provider to trust the next one. It meant knowing the brand rules, the real path a lead travels from click to signed client, the difference between a page that looks professional and one that actually moves the person reading it. The AI cannot know any of that, because it has never sat across the table from your customer. That knowledge comes from the relationship, and the relationship is not automatable.
The crews making slide decks at Every landed in the same place after all that engineering. Their words: the hard part was never pushing pixels. It was having something worth presenting. For a website, the hard part is having something worth saying, and knowing the business well enough to say it.
Now look at what Google does with your ad budget
Websites are just where I watch this happen most. The same pattern runs straight through your Google Ads.
Google will happily run your entire ad account for you now. Automated bidding, broad keywords, campaigns that spread themselves across Search, YouTube, and Gmail without you ever choosing where. You set a budget, and the machine optimizes around the clock. It is genuinely good at the optimizing, the same way it is good at the building.
The catch is what it optimizes toward. Google’s system will efficiently get you whatever you have told it to count as a win, and if you tell it a form fill is the goal, it will get you form fills. Cheap ones. Piles of them. Most from people who will never become your client.
A firm came to us with an account that looked great on the dashboard. Cost per lead down, volume up and to the right. Almost none of those leads were people who actually needed a lawyer. The machine had been cheerfully buying the cheapest clicks it could find, because no one had ever taught it the difference between a lead and a signed client.
What turned it around was not a smarter algorithm. It was judgment. Feeding the system a real signal, a booked consultation weighted far above a form fill. The negative keywords that only make sense if you know the practice. The taste to recognize a junk-lead pattern that the dashboard was scoring as a success. Same budget, results that finally showed up on the calendar instead of the report. The automation was never the hard part. Knowing what a good lead looks like for your business, and pointing the machine at exactly that, is the work that needs someone who has sat across from your clients.
If someone’s selling you “AI does your marketing”
This is the practical part, the reason I wrote any of this down. There are three things that can be true when a machine is doing the work, whether that is your website or your ad account, and you want to know which one you are buying.
- The do-it-yourself tool. Genuinely good at the mechanical work, with zero judgment and zero checking behind it. Perfect for a weekend project or a side hustle. For a firm whose marketing brings in its clients, it is a slow leak you will not notice, because there is no one whose job is to catch the white-on-white text, test the form, or ask why the leads never sign.
- The agency that “uses AI.” Ask them one question: what do you verify, and how. If the answer is some version of “the AI handles it,” you have your answer. It is the same as the do-it-yourself tool, except now you are paying agency rates for it.
- The setup worth paying for. Someone who has done the work to make the mechanical part fast, and then spends the time it saved on the parts that were always the actual job: deciding what to build and what to chase, and proving it works before it reaches your clients.
That last one is the whole game now. The labor of building and the labor of optimizing both got cheap, which makes the judgment about what to build, what to chase, and the discipline to check it worth more than ever.
The Bottom Line
AI made the easy half of the job nearly free. The hard half, the thinking and the checking, sits right where it always was. Anyone who tells you the whole thing is handled now is either selling you the overconfident tool or has become it.
If you look at your own setup and the mechanical work is automated but nobody is doing the second part, that is worth a conversation. Not an urgent one. Just an honest look at whether the thing that brings you clients has anyone actually watching it. That is the part we would want to talk about first.
FAQ
Can AI build a website for my law firm?
It can build the structure, the layout, and a solid first draft of the pages, often in a day. What it cannot do is know which of your existing pages carry years of search ranking you must not break, write the specific language that earns a nervous client’s trust, or confirm that your contact form actually reaches a human. The building is the easy 80 percent. The judgment and the checking are the part that protects your leads.
Is it safe to use a cheap AI website builder for my practice?
For a low-stakes project, sure. For a firm whose site is its main source of new clients, the real risk is invisible: something breaks, like a form that silently fails or a layout that falls apart on a phone, and no one is responsible for catching it. The builder constructs the page and then stops. There is no second set of eyes.
Will Google’s automated ads work for my firm?
They will spend your budget efficiently toward whatever you tell them to value, and the default goal, cheap form fills, is rarely a signed client. Automated bidding works when someone who knows your practice feeds it the right signal, a booked consultation rather than a form fill, adds the negative keywords, and watches what the leads actually turn into. The automation is fine. The judgment about what to optimize for is the part that needs a person.
Should I let my marketing agency use AI?
Yes, and you should expect them to. The better question is what they do after the AI hands them a draft. A good partner uses AI to make the building fast, then spends that saved time on strategy and verification: testing the forms, checking the site on real devices, and making sure the content fits your business. Ask them what they check before anything goes live. The answer tells you almost everything.
