How to build a website from a business's Google reviews

So here's the trick, and I'll give it away in the first line because it's not really a secret, it's just something nobody does: if you want a website that sounds like a business, don't ask the business. Read its Google reviews. The positioning is already sitting in there, written by the people who actually pay them.

I found this the annoying way. On the web side of Cadence Studios, I kept trying to get a business to describe itself so the generated site could sound like them. Send me a paragraph about what makes you different. And what comes back is always the same paragraph. Quality service. Family owned. Customer satisfaction. Twenty years of experience. Every single one of them, the same words, and none of those words are why anyone actually goes there. Businesses are genuinely terrible at describing themselves, and it's not their fault — you can't read the label from inside the jar.

Then it clicked. The reviews are the brand voice. What a customer reaches for when they're telling a stranger about the place is better positioning copy than anything the owner would ever write, because the customer isn't trying to position anything. They're just saying the true thing.

Let me walk through what that actually looks like as a process, because "use the reviews" on its own is the kind of advice that sounds smart and helps nobody.

First, you pull the reviews properly. Google's Places API gives you review text, ratings, and dates through the front door — that matters, because scraping someone's listing to build a business on top of it is both fragile and rude. Use the API. And pull enough of them that you're looking at a pattern instead of a mood. Under twenty reviews and you're basically reading three people's opinions with extra steps, which is one of the real limits of this whole approach and I'll come back to it.

Then the part that does the work: you're not looking for nice quotes. You're looking for repetition. The same noun showing up in forty reviews is a positioning statement. Somebody's name appearing over and over — a specific person people ask for — is a headline. A detail that keeps getting mentioned unprompted (the parking, the wait, the one dish, the fact that they called back) is the actual differentiator, and it's almost never the thing on the current website. I've seen a business whose whole homepage was about their equipment while every reviewer talked about one employee by name. The reviews knew. The homepage didn't.

You also get the customer out of it, which is the half people forget. Reviews leak who's showing up and why. Occasions, timing, whether they came in a hurry or made a plan, whether they're locals or one-time. That's your audience definition, and you didn't have to guess or run a survey — it's just written down.

And you get the objections. Read the three-star reviews, they're worth more than the fives. Fives say "great, thanks." Threes say "loved it, but the booking was a mess," and now you know exactly what the site has to answer above the fold. Every recurring complaint is a section of the page. Address it before they have to wonder.

So the site gets assembled out of that. The headline comes from what customers lead with, not what the owner leads with. The section order follows what gets mentioned most, so the page argues in the order people actually care. The tone matches the register of the reviews — if every reviewer is casual and warm, a stiff corporate site is factually wrong about the business. And the recurring complaints get answered on the page instead of ignored.

Now the part I want to be careful about, because this technique has a sharp edge and it's the reason I don't think you can fully automate it and walk away.

Reviews are skewed by nature. People write when they're delighted or furious, and the enormous middle never writes anything. So you're reading the tails and inferring the body, and you have to hold that loosely. Some reviews are fake, in both directions. Old reviews describe a business that may not exist anymore — a place under new management three years ago is being described by ghosts, so weight recent ones harder. And the thin-review problem is real: below a couple dozen reviews there's no pattern to find, and the honest answer is that this method doesn't work there, so don't pretend it does.

The one that matters most, though, is the line between voice and claim. A reviewer saying "best in the city" is not permission for the website to say "best in the city." That's a customer's opinion, and the moment your generated page states it as fact you've written a marketing claim the business can't back and didn't approve. So the reviews shape how the site talks and what it talks about. They don't get promoted into assertions. Direct quotes stay quotes, attributed, doing what quotes do. Everything else is inference — and inference is exactly the kind of thing a human should be glancing at before it goes live under someone else's name. Same instinct as the hard approval gate I put on the outreach side: the machine can do the reading and the drafting, but somebody has to actually look.

The honest status: this runs, it's live, and I've got no client outcomes to show you. No conversion lift, no A/B test, no "sites built this way perform X% better." I'm not going to invent one — that rule is the whole point of writing these. What I can tell you is that the pages it produces are recognizably about that specific business in a way that the template-and-stock-photo version never is, and that's a low bar the entire small-business web somehow fails to clear.

The thing that still gets me is how much of this predates AI. You could have done review-mining by hand at any point in the last fifteen years. Read a hundred reviews, notice the repeated words, write the page around them. It just wasn't worth an afternoon per client. The model doesn't make the idea good — the idea was always good. It makes it cheap enough to do for every business instead of the one you're being paid enough to care about.

Which means the interesting question was never "can it write the site." It's whether reading a business's customers back to them beats letting the business describe itself. I'm fairly sure it does. Ask me again when someone's paid for one.