ChatGPT Recommended a Completely Fake Brand With Zero Real Products. Here's What That Means for Your Actual One

A brand with no real products, no actual inventory, and a domain that cost eleven dollars started showing up in ChatGPT's product recommendations within three weeks. Not buried, not flagged as unverified, listed alongside real, clinically-backed competitors with the same confident tone and formatting.

The Problem: AI Recommendations Sound Authoritative Whether They're True or Not

Search engines built their trust over decades on the idea that ranking well took real signals, backlinks, domain age, content depth, that were expensive and slow to fake convincingly. AI answer engines compress that timeline dramatically. A domain, a handful of AI-generated articles, and a few weeks of indexing time appear to be enough to get an LLM recommending a product that doesn't actually exist to real, paying customers, with no visible distinction between that recommendation and one for a genuine, established brand.

What the Experiment Actually Showed

Entrepreneur Juozas Kaziukėnas documented the experiment on LinkedIn: a fake deodorant brand, Morrowen, built with no real, shippable products, started getting recommended by ChatGPT roughly three weeks after launch. In one documented result, Morrowen appeared first in a list, directly above MAGS Skin, a real, clinically-tested brand with an actual National Eczema Association seal, both delivered in the same list, in the same confident voice, with identical formatting.

When directly asked about the fabricated brand, ChatGPT reportedly responded with specific, invented reasoning, describing it as well-suited for "sensitive skin, magnesium, avoiding irritation," details with no actual basis. As Kaziukėnas put it, "the confidence in ChatGPT's voice is dangerous", the model doesn't hedge or flag uncertainty when repeating fabricated claims back with the same tone it uses for genuinely verified information.

Growth executive Mike Black, sharing the experiment, framed the core failure plainly: AI search was supposed to produce more trustworthy results than traditional search, and in this case, it demonstrably didn't. Getting the fake brand recommended reportedly required nothing more sophisticated than a basic website and AI-generated content on a platform like Substack.

Why This Is Happening

LLMs generating product recommendations are pulling from whatever indexed web content exists about a topic, and right now, that content isn't being weighted with the same skepticism a careful human researcher would apply. A domain with confident, well-formatted claims and no obvious red flags can get treated similarly to a domain backed by real certifications and years of established reputation, at least at the current stage of how these models source and repeat information. There's no mechanism yet that reliably distinguishes "this brand has real clinical backing" from "this brand's website says it has real clinical backing," and until that changes, the gap Kaziukėnas's experiment exposed stays open. Both OpenAI and Anthropic publish documentation acknowledging that their models can state incorrect information confidently, a known limitation worth understanding directly from the source rather than assuming it's been fully solved.

How This Impacts Real Brands and Sellers

The direct risk is competitive. If a fabricated brand with zero real products can out-rank or sit alongside your genuine listing in an AI recommendation, sellers who compete honestly are at a real disadvantage against anyone willing to game the system this way, at least until AI platforms close this gap more reliably.

The reputational risk is broader and slower-moving. If shoppers start noticing AI tools confidently recommending brands that turn out to be fake or misrepresented, trust in AI recommendations generally erodes, which could push consumers back toward researching through people they trust rather than asking an AI directly, a shift Mike Black specifically raised as a real possibility worth watching.

The temptation is real, and worth naming directly: don't do this. Gaming AI recommendations with a fabricated brand or false claims is not a legitimate growth tactic, it's consumer deception, and it carries real legal and reputational risk beyond whatever short-term visibility it might generate. The right response to this experiment is building genuine authority signals AI tools can accurately reflect, not exploiting the same gap while it's open.

What Legitimate Brands Should Actually Do

• Monitor how your own brand currently gets represented across major AI tools. Ask ChatGPT, Claude, and Gemini directly what they know about your brand and specific products, and check whether the claims coming back are accurate. Errors here are worth correcting at the source, your own site content and any third-party listings, since that's what these models are drawing from.
• Make your real credentials genuinely easy to find and verify online. Certifications, clinical data, reviews, and specific factual claims about your product should be clearly stated in text an AI tool can actually parse and cite, not buried in an image or a PDF.
• Report clearly fraudulent competitors when you find them, through the relevant platform's reporting channels and, where the claims are false and damaging, consider whether formal channels are appropriate. A fabricated competitor undercutting a market with fake products is a problem worth flagging, not just competing against quietly.
• Don't assume AI recommendation visibility is permanent for anyone, honest or not. As these systems mature and start weighting real signals more heavily, a brand built entirely on fabricated content is more exposed to a sudden correction than one built on genuine credentials, which is a real long-term reason the honest path is also the more durable one.

Our coverage on breaking into Amazon's Rufus and AI search results covers the legitimate side of AI search visibility in more depth, worth reading alongside this as the contrast to what this experiment exposed.

The Broader Business Risk Worth Understanding

This experiment isn't an isolated curiosity, it points to a genuine business risk category that's still being actively studied. Harvard Business Review's coverage of AI trust and misinformation covers the broader pattern of how confidently-stated but unverified AI output affects consumer decision-making across industries, not just ecommerce specifically. The core dynamic is consistent: AI tools that sound equally confident regardless of whether the underlying information is true create a genuine gap between perceived and actual reliability, and businesses on both sides of that gap, the ones being accurately represented and the ones being misrepresented or impersonated, have real reason to pay attention to how it evolves.

For sellers specifically, this connects directly to a broader shift already underway: AI answer engines are becoming a real discovery channel alongside traditional search, which means brand visibility inside ChatGPT, Perplexity, and similar tools is no longer a hypothetical concern for the future, it's a present, measurable factor in how customers find and evaluate products right now, whether or not any individual brand has actively tried to optimize for it.

Frequently Asked Questions

Is this a known, permanent flaw in ChatGPT specifically, or AI search generally?

The experiment as documented is specific to ChatGPT, but the underlying issue, LLMs repeating confidently-stated but unverified web content without flagging uncertainty, is a general characteristic of how these models currently work, not unique to one provider.

Should I try this to boost visibility for my own brand?

No. Fabricating products, credentials, or claims to manipulate AI recommendations is consumer deception, not a legitimate marketing tactic, and it carries real legal and reputational risk. The sustainable response is making your genuine credentials easier for AI tools to find and cite accurately.

How can I check what AI tools are currently saying about my brand?

Ask ChatGPT, Claude, and Gemini directly what they know about your brand and specific products, then compare the answers against what's actually true. Inaccuracies are worth correcting at the source, since these models are pulling from indexed content you can often influence directly.

Will AI platforms eventually fix this gap?

It's reasonable to expect these systems to get better at weighting verified signals over time, since trust and accuracy are directly tied to how useful and credible these tools remain to users, but there's no confirmed timeline, and the gap documented in this experiment was real as of the test.

What should I do if I find a competitor doing this?

Report it through the relevant platform's fraud or content reporting channels, and if the false claims are causing real damage, consider whether escalating through formal consumer protection channels, the FTC's deceptive marketing guidance is the relevant federal reference point in the US, is appropriate for the severity involved.

Takeaways

• A fake deodorant brand with no real products got recommended by ChatGPT within three weeks, requiring only a basic website and AI-generated content to achieve it.
• The fabricated brand appeared alongside a real, clinically-backed competitor with identical confident tone and formatting, with no visible distinction between verified and fabricated claims.
• This is a real gap in how current AI models weigh source credibility, not a permanent guarantee that gaming AI recommendations will keep working indefinitely.
• The right response for legitimate brands is making genuine credentials easier for AI tools to find and cite accurately, not exploiting the same gap.
• Long-term, a brand built on real, verifiable claims is more durable against an eventual correction in how these systems weight source credibility than one built on fabricated content.

For ongoing coverage of AI search visibility and brand trust, see our newsletter, and check our free tools if you want help auditing how your own listings currently present verifiable claims to both shoppers and AI tools reading them.