How do you tell a real crowd from a bought one?

Picture yourself on a product page at midnight, thumb on the scroll wheel, past the photos and straight down to the reviews. Five stars. Five stars. Five stars. Card after card says some version of the same thing: the shipping arrived fast, the results showed up within a week, this changed everything. You scroll ten more and the pattern holds. Somewhere in your chest a small, useful animal wakes up and says, hold on.

That animal is worth trusting. A real crowd of buyers is loud and uneven: some loved it, some are annoyed about the box it shipped in, one person wrote four hundred words about a totally different product by mistake. A crowd that agrees this cleanly, this fast, on a page nobody but the seller controls, is a crowd worth a second look. That second look is what follows.

Here is the mechanism, and it explains the tell. A real review gets written by one person, in their own time, in their own words, because something actually happened to them. A manufactured one gets written to a brief: post it now, keep it short, hit five stars, work in the phrase the client wants worked in. Whoever runs the operation is paying by the review or by the hour, so speed matters more than variety.

That pressure is what leaves the fingerprint. Real opinions arrive on their own schedule, scattered across weeks. Paid ones arrive on the operator’s schedule, bunched into an afternoon. Real opinions reach for a hundred different ways to say the same thing. Paid ones reach for the same three, because the brief only supplied three, whether the platform is Amazon, Yelp, or Trustpilot. And every message, however many names it wears, still gets typed somewhere, from some account, on some connection. Pull on any one thread long enough and the whole bundle gathers at a single point.

Here is the wall, twelve cards deep, every one reading the same at first glance. Flip the control and follow the threads: watch which cards trace back to a single point, and which ones stand alone with nowhere to trace.

Eight of the twelve trace to the same point. Four stay exactly where they started, unconnected to anything, which is what an honest review looks like on a diagram: alone. The eight that converge carry no proof, by themselves, that any single review is fake. A real product really can go viral in an afternoon. But a whole wall converging on one address is the shape coordination makes, every time someone goes looking for it.

In 2019 the Federal Trade Commission charged Sunday Riley Modern Skincare and its CEO, Sunday Riley, with directing employees to post five-star reviews of the company’s own products on Sephora.com between November 2015 and August 2017, from fake accounts built to hide who was really typing. A staff email the FTC quoted, written by Riley herself in July 2016, laid out the instructions plainly: create three accounts on Sephora.com, registered as . . . different identities, Always leave 5 stars, and DISLIKE any negative review because, in her own words, it is removed. This directly translates into sales!! When Sephora caught on and pulled the fake reviews, the operation adapted rather than stopped, moving staff onto an Express VPN account [to] . . . allow us to hide our IP address and location when we write reviews. Every one of those five-star posts was built to read as an independent shopper. Every one of them, underneath, traced back to the same handful of desks.

Here is the detail that makes the case sharper than an ordinary fine: there was no dollar penalty at all for the conduct itself. Two of the FTC’s own commissioners, Chopra and Slaughter, said so in writing at the time, arguing the agency should have sought real money even where the exact ill-gotten gains are hard to measure. A well-known, Sephora-carried skincare brand ran a fake-review operation for close to two years, got caught, and paid nothing for that specific conduct. Remember that the next time someone assumes this only happens on the cheap end of the internet. It reaches well into the brands sitting on your own bathroom shelf.

Just over two years later, a much bigger fast-fashion retailer showed the mirror version of the same trick: hiding the votes instead of inventing them. In January 2022 the FTC announced Fashion Nova would pay $4.2 million to settle allegations that from late 2015 until November 2019, its review system auto-published only four- and five-star ratings and simply held every lower-starred review for an approval it never gave. In the agency’s own words: Fashion Nova used a third-party online product review management interface to automatically post four- and five-star reviews to its website and hold lower-starred reviews for the company’s approval. Hundreds of thousands of one-, two-, and three-star reviews sat in that queue while the storefront glowed. It was the agency’s first case built entirely around a company hiding bad reviews rather than inventing good ones, and unlike Sunday Riley, this one cost real money.

By 2024 both moves, inventing praise and burying complaints, had their own federal rule. The FTC’s Rule on the Use of Consumer Reviews and Testimonials took effect October 21, 2024, after an August 2024 publication in the Federal Register. It bans fake and AI-generated reviews, reviews from undisclosed insiders, review buying tied to a required sentiment, passing a company’s own site off as independent, threatening or suppressing negative reviewers, and buying or selling fake social-media influence markers, the exact menu Sunday Riley and Fashion Nova ran through years earlier. A knowing violation now carries a civil penalty up to $51,744, per violation, a number that moves a page of reviews from a marketing choice into a legal one.

None of this is confined to a handful of prosecuted cases either. Michael Luca and Georgios Zervas spent years studying the filter Yelp itself runs to catch exactly this kind of review, and published the results in Management Science in 2016: roughly sixteen percent of Yelp restaurant reviews get filtered out as suspected fraud. The filtered ones run to extremes, glowing or scathing, rarely anything in between, and the businesses most likely to fake a review are the ones you would expect: a weak or brand-new reputation, or a competitor crowding the same block. Chain restaurants, which gain little from a Yelp bump, fake the least. Sixteen percent, at scale, across one whole platform, is no longer an edge case. It is closer to a tax.

That is the scale of it. Here is one thread, pulled all the way to the end. A Vine reviewer on Amazon, someone the platform itself vets before it ever sends a free product, ordered a supplement called PeakVigor anyway because, in their words, all the live reviews looked fake AF and they wanted to check for themselves. Their own critical review got rejected three times before a softened version finally cleared. Digging further turned up the sharper detail: the listing’s return address matched other seller accounts Amazon had already closed for the exact same reason. A Nolemy record is one run of exactly this kind of dig, chasing a wall of five-star cards back to the address they share.