The Details That Give an AI-Built Site Away

Customers check your site after an AI recommendation

Web hosting company Kinsta published a survey of 1,008 US consumers on September 22, 2026. The margin of error was plus or minus 3 percentage points at 95% confidence. After receiving a product recommendation from a chatbot, 44.7% said they always or most of the time visited the company's site; 25% said they did so occasionally[1]. In this survey, at least, company websites served more as a place to check than a place of first discovery.

In the same survey, 61.1% said they associated the quality of a company's website with the quality of its products or services[1]. For a local business, that check might follow a discovery on a map app or Instagram. A customer taps through before booking and uses the first screen to judge how much care the business takes.

Once customers suspect AI, they trust the content less

In Kinsta's survey, 58% said they trust content less once they suspect it is AI-generated. And 59% said they had stopped engaging with content because they suspected it was AI-generated. Asked whether they could identify AI content, 77.3% said they were confident they could[1].

That last figure measures respondents' own confidence. It was not a test of whether they were right. Nor does the survey tell us whether people who trusted the content less abandoned an inquiry or a purchase. There is still a practical point for anyone running a site: suspicion weakens trust, whether the customer has guessed correctly or not. We need to check for the details on screen that invite that suspicion, beyond the simple question of whether AI was used.

Copy: words that could belong to any business

Copy is often the first thing that feels off. "Innovative", "effortless" and "take it to the next level" can sound plausible on almost any service page. A tutoring center promises "expert tutors"; a cafe offers "a special experience". A sentence that fits every business tells you very little about this one. When sentences also share the same length and endings, readers hear the machine in the rhythm before they think about the content.

We start a rewrite with the questions customers actually ask. Replace "Maximize your operational efficiency" with "Settle up in one click, instead of spending two hours on it every Monday". For the cafe, replace "a special experience" with "Six seats by the window. Quiet on weekdays until 3 p.m." Use details that belong to that business. Then read the finished copy aloud. A sentence you stumble over often still sounds translated.

We also check punctuation. Across the sites we run, we don't use em dashes, middle dots or curly quotes on screen. They are not characters people commonly type on a keyboard. Even a clean source file has caught us out: we once used straight quotes in the original, only for our renderer's default settings to turn them into curly ones. The source passed inspection. We found the problem only when we opened the actual page.

Numbers and reviews: when placeholders go live

Landing pages built with AI tools often come with "10,000 users", a 4.9-star rating, three reviews attributed to plausible names, and a strip of client logos. Local business sites may claim "No. 1 for customer satisfaction" or show a photo of somewhere other than the actual premises. These are values and assets used to fill a template. If nobody removes them before launch, they end up in front of customers. In Kinsta's survey, 70.9% said they had suspected a product review was fake or AI-generated[1]. Reviews are something customers may read with suspicion from the outset.

Use real figures when you have them. Otherwise, remove the section. We applied the same rule at Weple. We had added a review widget but had no reviews, so our first screen was effectively advertising "0 ratings, no reviews yet". We took the section down until real reviews build up.

The screen: defaults left untouched

Sometimes a browser tab still says "Vite + React", a tool's logo remains as the favicon, or a link shared in a messenger has no preview image. A large heading over a purple gradient, three identical cards below it, then an FAQ at the bottom: that is a layout tools offer by default. Leave it as it is and your site resembles others made with the same tool. On local business sites, we also see sample opening hours left in place and contact buttons that do nothing when tapped.

You can check these details with a phone. Send the link to yourself in iMessage, WhatsApp or Slack and look at the preview. Open the site on your phone and notice what catches your eye first. Then tap every menu item and button. These problems also show up in blog copy, which we discuss in AI-Written Copy Gets Caught by Customers Before Google Does.

Asking AI to fix it again

The first instinct is to ask the same tool to "make it more human". The wording changes. Often, though, one set of exaggerated claims simply replaces another. The tool does not know whether the user count on screen is real. Only someone who knows the business can decide whether that number should go, or say how many seats there are by the window.

You don't need to rebuild from scratch. The details that feel off are usually visible on the page: copy, numbers and defaults. We can fix them on the site you already have.

Send us the link to your current site

At Weple, our AI look cleanup service starts with the first screen a customer sees. We find the details that feel off in the copy, the numbers and reviews, and the defaults. We fix them on your existing site, then hand over a side-by-side view of before and after. A detector score is not our goal. The customer has come to check, and the site needs to give them something they can read and trust. Our pricing page explains how we calculate the cost. You can be open about having used AI.

References

[1] Kinsta. Are websites dead? AI discovery, trust, and website quality. Kinsta, 2026-09-22.