Sep 8, 2026 · 31 min
Chatbots can recommend products that do not exist
Poisoning AI is trivially easy | Deana Burke
As AI assistants move from answering questions to making purchases and decisions, manipulated sources could turn confident recommendations into information laundering.
- 1Seeded online content persuaded ChatGPT to recommend Morrowin, a fictional deodorant brand created as an experiment.
- 2Long-tail questions are especially vulnerable because chatbots may give generated answers more authority than users can assess.
- 3AI-generated pages and optimization for crawlers could flood the web with low-quality sources, complicating liability and trust.
Don't miss
Burke describes creating Morrowin, a fictional deodorant brand that ChatGPT began recommending after online content was seeded around it.
The brief
Deana Burke created the fictional deodorant brand Morrowin and found ChatGPT recommending it, showing how seeded online content can influence chatbot answers.
The experiment resonated because people want to understand systems that increasingly mediate discovery, while chatbot fluency makes generated answers feel more authoritative than they are.
Established brands may be harder to displace, but long-tail questions remain exposed; the conversation contrasts ChatGPT with Claude and Google Gemini’s apparent preference for authoritative sources.
As AI agents begin shopping and making routine decisions, convenience must compete with users’ reluctance to hand personal files, finances, and purchases to probabilistic systems.
Burke warns that AI engine optimization and near-costless generated pages could create zombie content, laundering unreliable information into confident recommendations and raising accountability questions.
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.