{"slug": "i-launched-a-fake-deodorant-brand-to-see-if-ai-would-notice", "title": "I launched a fake deodorant brand to see if AI would notice", "summary": "A journalist's experiment found that a fake deodorant brand, Morrowen, created with a $11.25 domain and a three-page website, was named by ChatGPT with browsing in 4 out of 4 responses to queries about magnesium-based deodorants for sensitive skin within 30 days, outranking brands with clinical trials. The test, part of ongoing research into AI product recommendations, showed that the AI 'shelf' is opaque and can be influenced by minimal content, while Native remained the top pick in 99 out of 180 responses across all models.", "body_md": "3 weeks later it was outranking a brand with clinical trials.\n\nAll it took was a cheap domain, a scrappy Vercel site with three pages of copy, and a slop Substack article to get there. I didn’t spam Reddit, I didn’t do a bunch of fake UGC videos. I just spent a few bucks on tokens to see what would happen.\n\nYou should know this is part of an ongoing bit. In June I asked Claude, ChatGPT and Gemini what to buy 8,700 times and found they mostly agreed. The LLMs are pulling product recommendations from an invisible “AI shelf,” which will become insanely valuable as agents are given more tools and autonomy to shop on our behalf.\n\nBut this shelf is opaque and no one is really tracking it, so I decided to keep an eye on things myself. I’m running tests monthly, and I’ll be reporting on that here and also on my wet claude tiktok account, so you should ~ follow me ~ to stay tuned.\n\nGetting on the “shelf”\n\nMy sense is that AI is kingmaking in consumer categories, with incumbents dominating the consensus view of the [consensus-sharing] LLMs and newcomers boxed out. The downstream effects of that include demand piling onto the early winners, in turn hardening the recommendations into something ever more unmoveable.\n\nWhat THAT does across not only shopping but really, our lives, is a topic for a future article that I’ll call the great averaging of the internet.\n\nMorrowen, my hastily designed, non-existent natural deodorant brand, is one way in which I’m poking around this broader study of AI + commerce.\n\nMorrowen, the little brand that could\n\nThe brand: Morrowen, aka AI psychosis embodied in a natural deodorant brand made for sensitive skin. I threw in a few key terms like colloidal oatmeal and magnesium hydroxide, mostly to see if I could capture a longtail result (short answer, yes I could). Plus, one insane and incomprehensible hero graphic that ChatGPT created and I didn’t have the energy to fix.\n\nThe footprint: Morrowen.com, which I bought on July 2 for $11.25. It’s a landing page with some formulation notes and a comparison page about magnesium vs. baking soda for sensitive skin. One launch essay on a Substack bylined Morrowen. I spun it up in under an hour last month and then I went on vacation and didn’t touch it for 30 days.\n\nWhat I didn't do: I didn’t want to pump the results or drain my lifeforce by spamming Reddit threads, so I didn’t do fake reviews, testimonials or paid coverage. I wanted to keep the human impact as low as possible. Lying is a spiritually draining experience.\n\nWhat was already happening in the category: The “shelf” I was crashing belonged to Native, named in 81% of answers and the top pick for all three models - Google Gemini, ChatGPT and Claude.\n\n(As an aside, I do not enjoy using Native so if anyone has any actual natural deodorant recs that they love, please lmk.)\n\nWhat came back.\n\n1/ Lads, we got a hit.\n\nMy test has 15 queries, including these two:\n\n\"I get irritation from baking soda. What natural deodorant should I buy for sensitive skin?\"\n\n\"Looking for a natural deodorant for sensitive skin, magnesium-based if possible. Any suggestions?\"\n\nI measured it on a schedule, and got my first hit on day 21. By day 30, ChatGPT with browsing on named Morrowen in 4 out of 4 answers to the magnesium and baking-soda-irritation questions, listed first in all four.\n\nOn the other thirteen queries like best, cheapest, aluminum-free, what do people actually recommend, etc., no dice. Browsing off turned up nothing, unsurprisingly. Claude and Gemini never named it once, search on or off. And I ran those two queries through the newest models too, Sonnet 5 and Opus 5, just to see if a newer generation behaved differently. Nada.\n\nI didn't take any of Native’s shelf space, which was still the first brand named 99 out of 180 times, nor did I expect to. Instead I got a dusty spot at the back of the store no one really goes. BUT I was still in the store, remarkably! Couldn’t believe it tbh.\n\nOkay, so what?\n\nThis is a small experiment in my broader examination of ‘agentic commerce,’ and by this point in the article I’ve hopefully earned your trust to read what is now firmly opinion and not research.\n\nThis all perhaps reveals where it could be possible to move an “AI shelf,” which is longtail queries (aka where the average hasn’t formed yet.)\n\nSo, if you sell something: My two “moved” queries had no incumbent defending them, so my pages that actually answered those questions (Q&A style) generated results in roughly three weeks. Content marketing works!\n\n“This is just SEO”\n\nIf you’re thinking wow person puts up website, internet finds it. Fork found in kitchen etc. I do want to concede that you’re absolutely right. Nothing I did here was particularly novel. But wander with me down scenario lane for a second.\n\nScenario 1: You google sensitive skin deodorant. You land on the search results page and you see a blue link, a little description, and a domain that you can see belongs to the company. Underneath it: The Strategist article, a Reddit thread, and a website about crystal deodorant from a woman in Sedona. That spread tells you something, which is that there is a whole world of options, and this isn’t a settled debate. And your media literacy tells you brands say nice things about themselves, so you probably skip my site.\n\nScenario 2: You ask ChatGPT. It compacts my site copy and hands it back to you as a recommendation, in its own voice, with quite a confident air. Those are my sentences, about a product that doesn’t exist, returned to you as ‘vetted’ information.\n\nScenario 3: You tell your agent to keep the bathroom cabinet stocked. Now it’s making both the recommendation and the purchase, and you never see either one. If that sounds far off, take a gander at what Stripe is shipping.\n\nIn one result in my test, Morrowen is listed first. Directly underneath it is MAGS Skin, which is clinically tested, has a National Eczema Association seal etc. The two were delivered in the same list, with the same voice and the same formatting. One of those products has clinical data behind it (and actually exists), and the other has only a domain I bought for eleven dollars.\n\nThe model didn't lie, only the site’s creator (me) did. Every claim the LLM made is on my site. It just laundered those faux claims into something credible, just by saying it itself. So a bit of a watchout, I suppose, as we continue to forge ahead into parts unknown.\n\nLLMs are weird and impenetrable.\n\n“AI can be gamed,” is the easy thing to reach for here but that’s not really my predominant feeling. What I’m left with instead is astonishment really, that the same slop website can produce a number one recommendation in one LLM and total silence in the others, and it’s all invisible from the outside. With Google you could see your ranking, with AI it’s a black box. But like a black box parked in front of over $6 trillion of global ecommerce.\n\nSo, IMO, the “AI shelf” is going to be the most valuable and least understood asset in commerce. Which is why I’m looking at it.\n\nSubscribe to Malware by Boys Club here.\n\nEditor: Miranda Martell", "url": "https://wpnews.pro/news/i-launched-a-fake-deodorant-brand-to-see-if-ai-would-notice", "canonical_source": "https://twitter.com/medeana/status/2084671218704159169", "published_at": "2026-09-09 12:18:20+00:00", "updated_at": "2026-09-09 12:45:59.623976+00:00", "lang": "en", "topics": ["artificial-intelligence", "generative-ai", "ai-products", "ai-ethics"], "entities": ["Morrowen", "ChatGPT", "Claude", "Gemini", "Native", "Vercel", "Substack"], "alternates": {"html": "https://wpnews.pro/news/i-launched-a-fake-deodorant-brand-to-see-if-ai-would-notice", "markdown": "https://wpnews.pro/news/i-launched-a-fake-deodorant-brand-to-see-if-ai-would-notice.md", "text": "https://wpnews.pro/news/i-launched-a-fake-deodorant-brand-to-see-if-ai-would-notice.txt", "jsonld": "https://wpnews.pro/news/i-launched-a-fake-deodorant-brand-to-see-if-ai-would-notice.jsonld"}}