The Fake Deodorant Brand That Won AI Recommendations
A fake deodorant brand shows why AI visibility needs corroborated source architecture, not prompt tricks.
Deana Burke's fake deodorant experiment is a useful AI visibility case study because it separates retrieval from trust. A brand can become visible to a chatbot before it becomes credible to a buyer. The lesson for CMOs is not to game AI recommendations. It is to make source quality harder to fake.
The fake deodorant case shows AI recommendation systems can confuse presence with proof
Burke wrote on LinkedIn that she launched a fake natural deodorant brand to test whether AI systems would notice it. Inc. reported that the nonexistent brand used a cheap domain, a three-page website, and a Substack post, then appeared in ChatGPT recommendations within weeks.
That does not prove a durable ranking formula. It proves a failure mode. When an AI answer surface lacks enough corroborated source signals, it may treat a newly crawled entity as more real than it deserves to be.
The sharper finding is the model split. Inc. reported that ChatGPT recommended the fake product while Claude and Gemini did not. That matters because AI visibility is not one channel. Each answer system weighs sources, retrieval, freshness, and confidence differently.
For brand teams, the risk cuts both ways. A weak or synthetic brand may appear faster than expected. A legitimate brand with messy sources may fail to appear at all.
Visibility is no longer the same thing as credibility
The fake deodorant case is tempting because it sounds like a hack. It is weaker than that and more important. It shows that AI visibility has at least three separate layers:
| Layer | What it proves | What it does not prove |
|---|---|---|
| Entity presence | A model can find a brand name and basic page | The brand is real, trusted, or preferred |
| Recommendation inclusion | A model may name the brand in an answer | The brand has durable authority across engines |
| Corroborated citation | Multiple trusted sources support the same claim | The brand will convert, rank, or stay visible forever |
OpenAI's shopping documentation describes shopping answers as product-oriented responses that can include links and product information when a query suggests shopping intent. That kind of interface raises the stakes: product visibility now happens inside synthesized recommendations, not only on search result pages.
The Burke experiment sits at the bottom of the table. It appears to show entity presence and limited recommendation inclusion. It does not show durable trust. That distinction is the whole operating lesson.
CMOs should audit source architecture, not chase chatbot tricks
AI visibility teams should treat this case as a source-architecture warning. A brand's machine-readable footprint now includes the home page, product pages, FAQ content, third-party coverage, social posts, marketplace data, reviews, schema, and any sources that answer engines can retrieve.
The easiest fake is a clean owned page with an assertive claim. The harder signal is corroboration: independent sources, consistent product facts, stable entity descriptions, and citations from domains answer engines already trust.
AuthorityTech's public publication intelligence tracks which publications AI engines cite across categories. The implication is simple: not every source has the same weight. A brand page, a LinkedIn post, a trade article, and a high-authority publication do different jobs in the source graph.
This is where Machine Relations is a useful lens. The discipline, coined by Jaxon Parrott, frames AI visibility as a system of earned authority, entity clarity, citation architecture, distribution, and measurement. The fake deodorant story exposes what happens when entity presence outruns earned authority.
The measurement standard is moving toward quality, not just mentions
The broader market is already trying to clean up this gap. IAB released "Measuring Visibility in the AI Era" to help brands, publishers, and agencies measure AI-powered discovery. The important shift is that AI visibility measurement cannot stop at "was the brand mentioned?"
A mention is a weak unit. A recommendation is stronger. A cited recommendation across multiple engines is stronger still. A cited recommendation supported by trusted sources, current facts, and consistent entity data is the signal CMOs should care about.
That makes the Burke case less embarrassing for AI and more embarrassing for brand measurement dashboards. If a dashboard counts a synthetic brand mention the same way it counts a corroborated recommendation, the dashboard is measuring noise.
The Machine Relations Stack gives teams a cleaner diagnostic frame:
| Stack layer | Fake-brand failure mode | Legitimate-brand response |
|---|---|---|
| Earned authority | No trusted third-party proof | Build coverage from sources AI systems cite |
| Entity clarity | Thin identity can still be parsed | Keep the same brand facts across every surface |
| Citation architecture | Pages can be easy to extract but weak | Make claims specific, sourced, and structured |
| Distribution | One engine may surface the brand | Test across ChatGPT, Gemini, Claude, Perplexity, and AI search |
| Measurement | A raw mention looks like a win | Track cited, trusted, repeated recommendations |
The real lesson is source discipline
The fake deodorant brand did not expose a magic playbook. It exposed a missing quality gate.
For a real brand, the right response is not to spin up thin microsites and hope models repeat them. The right response is to build a source system that machines can verify. That means product facts that match across owned pages, marketplace listings, reviews, social profiles, and third-party coverage. It means clear entity descriptions. It means pages that answer buyer questions directly. It means enough outside proof that an answer engine does not have to trust the brand's own claim.
This is also why brand teams should be careful with "AI visibility" vendors that report mentions without source context. A brand needs to know whether it was named, why it was named, what source supported the answer, and whether the recommendation survives across engines.
The Burke case is useful because it is small, strange, and easy to understand. It shows the new shelf space. But it also shows the trap: being visible to a machine is not the same as being trusted by one.
FAQ
What did the fake deodorant brand experiment show?
It showed that a newly created brand with minimal source material could appear in at least one AI recommendation context. According to Inc.'s reporting, ChatGPT recommended the fake deodorant within weeks, while Claude and Gemini did not. The result is a warning about uneven source validation, not a reliable growth tactic.
Should brands try to copy the fake deodorant tactic?
No. The practical lesson is to build stronger source architecture, not synthetic brands or thin pages. Legitimate brands need corroborated facts, trusted third-party sources, and consistent entity descriptions so AI systems can distinguish real authority from easy-to-crawl claims.
How should CMOs measure AI visibility after this case?
CMOs should separate raw mentions from trusted recommendations. A useful audit should track whether the brand appears, whether the answer cites sources, which sources support the answer, whether the result repeats across engines, and whether the cited facts match the brand's real offer. A public AI visibility audit can expose those gaps.
Where does this fit inside Machine Relations?
The case sits inside citation architecture and entity optimization. It shows why AI discovery depends on more than content volume: brands need machine-readable identity, trusted corroboration, and measurable answer-surface presence.