What happened when we tested Claude, Perplexity, and ChatGPT on a real one-page website, and what it taught us about where AI ends and expertise begins.
There are two ways to get a great meal. You can go to a restaurant, or you can cook it yourself. For most of human history, that second option came with a real gap in quality, because the restaurant had the tools, the technique, and the taste that you didn’t.
That gap is closing. Home kitchens now come stocked with the kind of gear that used to live only in professional ones. Artificial intelligence (AI) is doing the same thing for digital work, dropping restaurant-grade capability onto everyone’s countertop. So we decided to turn on the stove and see what actually came out (yes, we are going to ride this cooking metaphor all the way to the table).
A friend of the company, Nick Jones, gave us the perfect excuse. Nick is a former Chief Marketing Officer and Chief Growth Officer (CMO/CGO) who now works full-time as a fractional consultant under his own brand, Omni Marketing Solutions. A hosting mishap had wiped out his website and left him with very little content to rebuild from. Low stakes, a real need, and a nearly blank canvas: exactly the kind of test we were looking for.
So we ran a small cook-off. We handed the same job to three platforms that can each write, design, and code: Claude, Perplexity, and ChatGPT. Same ingredients, same recipe, three kitchens.

First, the Mise en Place (ie, the prep)
Anyone who cooks knows the meal is won or lost before the heat ever comes on. Our prep took several hours and produced the least glamorous part of this entire story: the brief. It was also, as you’ll see, the most important.
We wrote more than two pages of content direction, color references, section outlines, and hyper-specific instruction about the result we wanted. The goal was a self-contained HTML page we could drag and drop onto a server with little-to-no manual adjustment. We partnered closely with Nick to lock down the actual words and assets, because a brief is only as good as the substance behind it.
Then we wrote one core prompt to pair with that brief, so every platform received identical material. We defined the output precisely (a responsive, self-hostable web page) and attached the real assets: two brand colors, Nick’s logo, partner brand logos, and personal photos of Nick himself. We also handed over three reference sites whose look and feel pointed in the direction we liked.
That last step matters more than it sounds. Design direction is hard to put into words, and showing is almost always faster than telling.

Three Cooks, Three Dinners
With the same recipe in hand, each platform plated up its own interpretation. Here’s the honest headline: all three did a good job. The real differences were in personality, not competence.
Perplexity took things the furthest. Without being asked, it added a light and dark mode, several on-scroll animations, and a scrolling logo marquee. It clearly came to play.

Claude added the least to what we gave it, producing the shortest page with the most corporate look of the bunch. Beyond a scrolling logo marquee, it kept the animated flourishes to a minimum. Clean and conservative, if a little buttoned-up.

ChatGPT landed the most modern-feeling design treatment, at least to our eyes. Its code was also the simplest of the three, with no animations at all. Pretty plate, plain recipe.

Picking a Plate and Finishing the Dish
Taste is subjective, and this was Nick’s site, so Nick got the vote. He liked ChatGPT’s design best, so that became our base. From there we ran several follow-up prompts to add rollover effects, scrolling content reveals, and a scrolling logo marquee of his own.

Getting the file onto the web server was genuinely as easy as dragging and dropping it to upload. Getting the contact form to actually work was another matter entirely. That required a mostly free third-party service and a surprising amount of fiddling before it behaved.
Was It Worth the Squeeze?
Start to finish, the site took around seven hours of actual effort. Run through our standard process of content to wireframes to design to development, we’d estimate the same one-page site at roughly two to two-and-a-half days. The time savings are real, and for a project like this one, absolutely worth it.
But notice where those seven hours went. The AI handled the fast part. The hours lived in the brief, the content, the judgment calls, and the finishing work.
What We’d Tell You Before You Try This at Home
A few things we learned, in no particular order:
- The brief is the recipe, and the recipe is the work. Our hours went into content, direction, and assets, not into waiting on the machine. Rigorous prompting with clear intent, relevant examples, and actual files is the whole ballgame. Garbage in, garbage out isn’t a cliché here; it’s the rule.
- Open-ended direction is a feature, not a bug. Our design constraints were light, just a couple of brand colors and a logo. With less to fight against, every platform produced something we were happy to look at. A dense, rule-heavy brand system would have made for a much tougher test.
- Show, don’t just tell. The three reference sites steered the design more than any sentence we could have written. Hand the AI real assets and real examples, not a pile of adjectives.
- Each platform has a personality. Same recipe, three genuinely different dinners. If the output matters, run more than one and compare, because the defaults vary wildly.
- The last 20% still bites. The page came together fast, but the connective tissue (a working form, hosting quirks, third-party services) still took patience and know-how. AI builds the thing; a human still has to make it work out in the wild.
- You have to know what “good” looks like. Someone had to recognize the stronger option and know exactly what to fix and how to ask for it. That’s taste and experience, and no prompt manufactures either one.
So, Who Needs a Chef?
This experiment had a second, quieter purpose. We wanted to understand what our clients can now do with us and without us. When anyone can hand a prompt to a machine and get a working website back, it’s fair to ask where a digital marketing agency fits in the picture.
Here’s our answer. The fancy appliances are in everyone’s kitchen now, but the appliances were never the hard part. Knowing what to cook, how to season it, which dish to send out, and how to rescue the one that isn’t quite right: that’s the job. AI didn’t replace that judgment on this project. It made our judgment faster.
That’s the whole thesis, really. We’ve always said you don’t pay us for our hours, you pay us for our years. AI doesn’t cheapen those years; it multiplies them. A client with the same tools and none of the experience walks away with a mediocre dinner and no idea why.
And a beautiful website is still just a plate of food sitting on a table. Without a strategy to bring people to it, you’re right back in field-of-dreams territory: you built it, but that doesn’t mean they’ll come. The build was the easy seven hours. Making a site actually perform is the part worth hiring for.
So no, the restaurant isn’t closing. It just got a much faster kitchen.
Nick is back online, and we had a good time proving a point. If your own site needs rebuilding, rethinking, or rescuing, we’d love to talk. Bring the appetite. We’ll handle the cooking.
Want to visit Nick’s site in action? You can find him here: https://thenickjones.com/