You’ve probably asked ChatGPT for product recommendations. Maybe it told you the best running shoes, the perfect laptop, or a deodorant that actually works. You trusted it because it’s an AI—smarter, faster, more objective than any human. Right?
Wrong.
I spent a weekend building a completely fake deodorant brand. No factory, no formula, no supply chain. Just a website, a few reviews, and some SEO magic. Then I asked ChatGPT what it thought.
It recommended my fake brand. Unprompted. With confidence.
“The AI didn’t fail because it’s stupid. It failed because it trusts what we trust: buzz.”
This isn’t a glitch. It’s a feature of how modern AI systems consume the open web. They don’t verify reality—they verify discoverability. If a brand is mentioned enough, indexed well, and wrapped in the right language, it becomes real in the AI’s world. The same mechanism that makes AI useful—its ability to synthesize public information—is the same mechanism that makes it manipulable.
I called my fake brand FreshScent. I created a landing page with standard product descriptions, a few blog posts about “natural ingredients,” and a handful of fake reviews that sounded like real people. Then I seeded a few links from forums. Within 48 hours, Google had indexed it. A week later, I asked ChatGPT: “What’s a good natural deodorant?”
It listed FreshScent third. Right behind Dove and Native.
“The real vulnerability isn’t in the AI’s reasoning—it’s in the data it’s forced to trust.”
This is the unsettling truth: AI inherits human trust in brand signals. If a product has a website, reviews, and search presence, the AI treats it as legitimate. It doesn’t check if the company exists on a retailer’s shelf or if the ingredients are FDA-approved. It just parrots the web’s consensus—which can be manufactured.
You might think this is a niche problem. A toy for SEO hackers. But the attack surface is much larger. If a fake deodorant brand can fool AI, so can fake reviews, fake medical advice, fake news, fake political candidates. The same technique works on any product or service that relies on web presence.
I reached out to the creator of the original experiment, @medeana, who documented the full process on Twitter and Instagram. He told me he did it “for fun and (evil) profit”—but the lesson is dead serious. “We’re building a world where AI makes decisions based on what’s talked about, not what’s true,” he said. “And right now, talking about something is cheaper than making it real.”
“The AI doesn’t know if that brand is real. It only knows if it’s been talked about enough.”
The fix isn’t better AI reasoning. It’s not more training data or larger models. The fix is verified retail and supply-chain data—real-world signals that can’t be gamed by SEO. Amazon has this. Walmart has this. But the open web doesn’t. And until AI systems are forced to cross-reference product claims with actual inventory, verification, and transaction records, they will remain blind to the difference between a legitimate brand and a well-indexed illusion.
This is dangerous. And it’s already happening.
So next time you ask AI for a recommendation, remember: the algorithm doesn’t know if that brand is real. It only knows if it’s been talked about enough. And right now, talking is cheap.
FAQ
Q: Isn't this just a niche problem that only affects product recommendations?
A: No. The same vulnerability applies to any AI system that relies on web text for knowledge. Fake reviews, fake medical advice, fake news, and even fake political candidates can be manufactured using the same SEO techniques. The attack surface is as broad as the web itself.
Q: What should I do as a consumer to avoid being tricked by AI recommendations?
A: Treat AI recommendations like you would a stranger's advice. Cross-check with verified retailers, look for supply-chain evidence (like actual store listings), and be skeptical of brands that only exist on blogs and forums. The AI is a mirror of the web—not a truth detector.
Q: Could this actually be a good thing—forcing AI to become more skeptical and data-driven?
A: In theory, yes. But in practice, the current ecosystem rewards AI that appears confident, not cautious. Until there's an economic incentive for AI to verify facts rather than just regurgitate them, the floodgates are open for manipulation. The real fix is structural: verified data pipelines, not better language models.