Eden Stelmach typed a note to a contractor. His supplement company, TruHeight, was running bots on social media, and users were flagging them. They did not sound like real people, he wrote. He sent the note. The bots kept posting. Millions of dollars kept coming in at $45 a bottle.

Stelmach co-founded TruHeight with Justin Rapoport. Since 2020, the two have built their sales on a wall of five-star reviews and a crowd of social media profiles praising the product. In April 2026, the FTC filed a complaint. The agency finalized its order in July 2026. It all traced back to that email.

"Let the reviews speak for themselves." You have heard that line. You have probably used it to close a trust call on a vendor, a partner, or a product. Stelmach built his entire operation on the other side of that instinct. He bet that if the volume looked right, no one would look further. He was right for years.

I have made this mistake. Not running bots. Trusting the count. Most people in this position make this mistake. You see a thousand five-star ratings, and your brain reads them as a crowd of satisfied buyers. A thousand ratings do not prove a product works. They prove someone put in the hours.

Here is what the FTC complaint laid out.

TruHeight employees wrote thousands of five-star reviews and posted them as if they came from real buyers. That was layer one. The company offered free and cut-rate products to outside buyers. The deal was simple: leave a five-star rating on Amazon or on the company's own site. Layer two. Then it bought more than 150 fake social media profiles rigged with AI chatbots to post praise on Facebook and Instagram. Layer three.

Not one rogue employee. A built system with a budget and years of run time.

The company's own "clinical evidence" rested on a single study of 32 people. It showed no real gap between the product and the control group. The study lacked proper controls and tested only one product. Even the science was a prop.

The FTC's Reviews and Testimonials Rule, which took effect in late 2024, now carries a fine of $53,088 per violation. The order hit Stelmach and Rapoport with a $4 million judgment, later cut to $750,000 based on what they could pay. The supply side of fake reviews now has a price tag. But there is a gap that the rule does not close.

Researchers at Copenhagen Business School showed 375 people a mix of real social media profiles and fakes built by AI. The task was plain: sort the real ones from the false ones.

The average score came in at roughly 50%. A coin flip. One well-built fake was flagged as a bot by only 10% of the viewers who saw it. Nine out of ten people thought it was a real person. Viewers tended to rate the AI accounts as more likely to be real than the true ones. The machine was more convincing than the human.

You look at a full comment section, a profile with the right photo and the right tone, and your brain reads it as a crowd of real people. Your brain is wrong about half the time. That is not a guess. That is what the experiment found.

The FTC can fine the seller. It can shut down the bots. It can name the principals in a federal complaint. What it cannot do is fix the way you and I read the signal in front of us.

"Let the reviews speak for themselves" puts the burden on the wrong side. It assumes the signal is clean. It treats volume as a stand-in for truth. What you think you want and what will actually work are two different things.

The better rule is short. Stop reading reviews for sentiment. Start reading them for structure.

Once that is clear, three moves follow from it.

Pull the full spread of ratings. A real product with real buyers almost always has a messy curve. Ones, twos, and threes mixed in with the fives. A wall showing 95% five-star ratings and almost nothing else is not beloved. It is managed. This move asks you to slow down at the exact moment the reviews are designed to speed you up.

Skip the praise. Go straight to the one-star and two-star reviews. Look for named details: a use case that failed, a concrete flaw, a timeline. Fake review systems rarely build detailed low-star content on purpose. If the negatives are vague or nearly absent, the top of the stack is worth less than it looks.

Pick three of the highest-rated reviews. Click the name. Look at the profile. A real buyer has a pattern: other purchases, other reviews, a trail that predates the product you are checking. A fake profile, even a good one, tends to have a thin trail. The 150-plus chatbot accounts TruHeight ran looked fine on the surface. The history behind them was hollow.

Running this for a few weeks does something the old advice never did.

You start to see which review walls were built and which ones grew on their own. You spot the tells: clusters of five-star posts inside the same short window, generic praise with no named detail, reviewer profiles that begin and end with one product. You catch the gap between a vendor's public face and the structure holding it up.

You also find the suppliers and partners whose reviews hold up under weight. Those are worth more to you than a thousand unread stars.

At the end of a month, ask three things.

→ Which trust calls held up after you ran the check?

→ Which ones looked solid on the surface but fell apart once you read the spread?

→ Where did the same pattern, clean top and thin trail, show up more than once?

That is the gap between advice that sounds right and a system that proves itself.

Stelmach knew the bots were getting flagged, and he put it in writing. The system kept working because the people on the other end had no audit. The FTC now covers the seller. The reading side is yours.