I built our new website with AI. Done blindly, it would have been terrible.
Years ago I wrote an article about Wix. The gist: it makes it very easy to build a nice-looking website, and very easy to build one that quietly loses you money. It ran for years and picked up a pile of comments, some of them genuinely entertaining. To be honest Wix was never really our target market, so it brought in all the wrong traffic and never much decent work. I've now pulled it, because there's a newer version of the same story, and it's a far more useful one to tell.
I've just rebuilt our own website, and I built it with AI.
If I had let the AI run blindly, I would have had a site live in a day. It would also have been slow, close to invisible in search, and it would have quietly cost the business money. Same trap as Wix. Same trap as Facebook's old "Boost" button. New tool, same trap.
The stuff that makes a website work is invisible at the front end
I've said this for years: most of what makes a website actually work is invisible at the front end. It sits behind the scenes, and that's exactly where you need a bit of knowledge. That was true with Wix. It's just as true with AI. Times change, the tools and platforms change, the principles don't.
Facebook understood this better than anyone with its old Boost button. You wrote a post, Facebook offered to "boost" it, you tapped a button, money left your account, and it felt like you were advertising. You weren't, really. There was no proper targeting, no structure, no tracking of what it brought back. It felt like progress. It was mostly vanity reach. Easy, satisfying, and quietly wasteful.
Building a website with AI is the 2026 version of that button. Type a prompt, watch a site appear, publish it. It looks fine. And like a Boosted post or a drag-and-drop Wix site, "looks fine" is doing a lot of heavy lifting, because what you see is not what Google sees, and it's not what your customer's slow phone sees either.
What "blindly" would have skipped
Here is the part that decides whether a website earns its keep, and every bit of it is invisible on the surface. If I had just let the AI build and hit publish, it would have skipped or fumbled nearly all of it. I had to push on every single one:
- Proper heading structure, so the page tells search engines what it's actually about.
- Page titles and meta descriptions that earn the click in search results, not just fill a box.
- Load speed. A fast site on a real phone, not a pretty one that takes five seconds to appear.
- Image compression and correct sizing. One oversized image can cost you more load time than the rest of the page combined.
- Clean URLs and internal linking, grouped into proper silos so related pages support each other.
- Articles written with the right content in the right format, so they actually get picked up in search rankings and by AI. (Our new core pages were in Google's index within two hours of going live, and some were cited by Google Gemini within three hours, all because of the technical steps we took.)
- Schema, analytics and conversion tracking, so you actually know what's working.
- Accessibility and contrast, so everyone can use it.
None of that shows up in the "wow, it built a website" moment. All of it shows up in whether you get found and whether you get business.
We didn't let it guess. We built to a standard, and I corrected it constantly.
This is where it went past a normal build. I didn't let the AI wing it. We built to a formal web specification, the standard modern browsers and search engines actually expect, and validated the result using not just one tool but three, plus Google's own. On top of that I corrected the AI constantly. It would suggest a way of doing something, and often there was a better way, so I pushed back. That matters more than people realise, because AI tends to agree with you and avoid being controversial, so if you don't know enough to challenge it, you take the easy answer and never see the better one.
The honest result is it ended up more thorough than a lot of professionally-built sites I've audited over the years. And I've audited plenty that shipped without so much as analytics installed.
The real unlock is context, not "we used AI"
Here's the bit that actually matters, and it isn't "we used AI." Anyone can use AI.
Don't get me wrong, AI is trained on an enormous amount of content and a lot of it is genuinely fantastic. We use it as the first port of call for almost every piece of work we do, precisely because we've built our own well-trained AI brain for the business. But what you have to remember is that most AI large language models are trained on the lowest common denominator, the most common stuff out there. Left to their own devices they give you content that reads like everyone else's.
The difference for us was the context we'd already loaded into ours: our clients, the sectors we work in, our data, our voice, and years of experience. That is what made the content come out fast and genuinely ours, not a template with the names swapped in. And to be clear, not all of it was written by AI. Plenty of it was written or reworked by us, usually starting from dictated notes of what we wanted on the page. The AI was a starting point, not the finished article. That combination, our knowledge steering a well-briefed tool, is the whole game.
The proof
When our new site was first built, before we did the technical work, it scored 52 on Google's PageSpeed test. We got it to 100 on desktop and 96 on mobile by fixing the real issues: deferring scripts until they're needed, self-hosting fonts, loading videos only when someone clicks play, sizing every image to what it's actually displayed at.
And the old site? On paper it scored well. But only because it leaned on a plugin that deferred the slow parts until after the speed test had finished, so the number looked great while the page was still doing all that work a moment later for real people. That's the difference between a good-looking score and a genuinely fast site.
That gap, 52 to 100, is the whole point in one number. "AI built it" versus "built right."
What a website that can't be found actually costs
This isn't abstract, and I ran the numbers on it years ago. In the old Wix article I told the story of a garage owner whose cheap website couldn't be found. When a proper site and the right business listings went live, he was inundated with work.
Put an estimate on what the poor-performing site had been costing him. Say just one extra job a week would have come through a site that could actually be found, at an average of $200. That's $10,400 a year, or $41,600 over the four years the old site sat there invisible. Now take it closer to reality for a busy garage found online, nearer five jobs a week at an average repair of around $305, and the estimated loss climbs to roughly $317,782 over those four years. All from a website that quietly couldn't be found, and he never once got that bill in a lump.
Those are estimates based on real-world data, but that's the point: the cost of a poor-performing site is real, it's large, and it's invisible. That maths hasn't changed. A site that looks great but can't be found, loads slowly, and never turns a visitor into an enquiry costs you far more than you saved by getting it up in a day.
And that cost isn't just lost search rankings. It's spread across everything you do online. A slow, poorly-built page makes your paid ads cost more, because the ad platforms charge you extra for a weak landing experience. It drags down how many of your visitors turn into customers. It holds back your business performance overall, not just where you sit in Google. One weak site quietly taxes the lot.
So, should you use AI to build your website?
Yes. But not blindly.
The tool isn't the problem, and I'm not here to tell you AI is bad any more than I was telling you Wix was bad. It's usually not the platform that lets you down, it's the output. Platforms are designed to make things easy for you, not necessarily to make them the best. The gap between "easy" and "best" is where the knowledge earns its keep.
And I'll be honest, I know how it feels to let this slide. We hadn't properly updated our own website in years, because we were too busy growing our clients' businesses. So this rebuild was overdue, and it was the perfect chance to prove the point: AI plus real expertise, done properly.
Your business doesn't need a website. It needs customers. The website is just one way of getting them, and only if it's built to do the job.
If you want an honest read on your current site, or you're thinking about building one with AI and want it done properly, get in touch.