Fake testimonials are a liability now. I deleted mine and the rating went down.
Invented testimonials are a legal exposure and a conversion problem. I swapped them for real reviews — and lowered an inflated rating to match reality.
One of the consumer brands I look after had three glowing customer testimonials on its product page. A woman in her city, a man giving his age, a couple with a warm little story. They were well written, on-message, and completely invented.
Nobody set out to deceive anyone. They were placeholder copy that quietly became permanent, the way placeholder copy does. But invented testimonials on a live commercial page aren't a harmless rough edge anymore. They're a legal exposure and, less obviously, a conversion problem. So I pulled them.
Made-up reviews stopped being a grey area
In August 2024 the US Federal Trade Commission finalised a rule that bans fake reviews and testimonials outright, including AI-generated ones and any testimonial from an insider that hides the relationship. It carries civil penalties of up to roughly $51,000 per violation, and a violation can be counted per day. You can read the FTC's own announcement of the final rule if you want the full scope. Europe has had comparable consumer-protection teeth for years.
This is the part a lot of brands haven't internalised: the friendly fictional customer on your homepage isn't marketing flair. It's the exact thing a regulator now writes fines about. The risk isn't theoretical and it isn't small.
But the legal angle isn't even the most interesting reason to take them down.
The real reviews were better, and the rating was too high
The product has hundreds of genuine reviews on a major marketplace. Real names, verified purchases, the actual language customers use, which is always messier and more specific than the copywriter version. I went and read them, pulled a handful of authentic ones in the right country and language, and swapped them in for the fakes.
Then I found the more revealing problem. The site was displaying a near-perfect star rating and a review count, both nudged up from what the marketplace actually showed. So I did something that feels backwards: I lowered the rating on the site to match reality, and trimmed the count down to the honest figure.
Lowering your own star rating sounds like marketing self-harm. It isn't. The research is surprisingly clear that the flawless score is the weak one.
A perfect five-star average doesn't read as "excellent." It reads as "fake," and a lot of shoppers skip right past it.
Northwestern's Spiegel Research Center, working with PowerReviews, found that purchase likelihood actually peaks somewhere around 4.2 to 4.5 stars and falls off as you approach 5.0. Buyers read the near-perfect number as too good to be true. The same studies show most shoppers actively seek out the negative reviews, because a believable mix is what tells them a real human bought the thing. A slightly lower, honest rating converts better than an inflated one. I wasn't sacrificing performance for ethics. The honest number was the higher-performing number.
Where there were no real reviews, I said so
The brand sells into several countries, and a few of the newer markets simply don't have native-language reviews yet. The lazy move is to translate the fakes and move on. I'd just spent the morning deleting fakes, so that was off the table.
Instead, those pages now carry an honest placeholder: a plain line saying real customer reviews will be added from the marketplace as they come in, and no star rating at all until there's a real one to show. An empty-but-honest slot beats a full-but-fictional one. This is the same instinct I keep coming back to in selling with the proof you actually have rather than the proof you wish you had.
Why a 20-year operator deletes the thing that "works"
This is where judgement earns its keep, and where I think the experience gap shows up most. A tool, or someone newer to this, optimises for the metric in front of them: rating high, testimonials present, page looks persuasive, ship it. Every one of those signals points the wrong way once you understand how trust and regulation actually work. I've written before about how these systems will confidently generate plausible, polished, wrong things if you let them, and an invented testimonial is exactly that — plausible, polished, and a liability.
The leverage isn't the AI that can rewrite a hundred product pages in an afternoon. It's knowing which direction to point it. I used automation to find the real reviews, parse the page, swap the strings surgically and verify the result live. But the decision to lower my own rating, to leave a slot honestly empty, to treat a "converting" element as a problem — that came from twenty years of watching what actually builds trust. It's the same argument I keep making about domain expertise being the real leverage: the model gives you speed, your judgement decides whether speed is taking you somewhere good.
Reviews are now table stakes — BrightLocal's surveys put the share of consumers who read them before buying in the low nineties percent, and a growing chunk won't touch a business under 4.5 stars. Which means the pressure to inflate has never been higher, and the penalty for getting caught has never been steeper. The honest version is also the better-converting version. You rarely get to make that trade so cleanly.
Take the fake reviews down. The real ones are doing more for you than you think.
Sources & further reading
External
FTC — Final Rule Banning Fake Reviews and Testimonials (Aug 2024) · Spiegel Research Center (Northwestern) — How Online Reviews Influence Sales · BrightLocal — Local Consumer Review Survey
Related posts
The AI kept inventing football crests. The fix was a sentence. · No traffic numbers to show? Sell the risk away instead. · The Death of Generic AI: why deep domain expertise is the only real leverage left